So the question for anyone who owns the index is where that growth is coming from.

Where the growth comes from

FactSet estimates the index’s third-quarter earnings at $800.8 billion. An increase of 29.5 percent means about $182 billion more than a year earlier. To find out which companies supply it, we went through FactSet’s sector data and the companies’ own reports.

Micron alone accounts for about 19 percent. Its non-GAAP net income for the quarter it reported on 30 September was $38.4 billion, up from $3.5 billion a year earlier. Nvidia, which reports in November, is expected to add about 16 percent, based on FactSet’s estimate of $2.47 a share against $1.30 a year earlier. Together, two companies supply about a third of the growth of the whole S&P 500.

Look more broadly and growth is still concentrated. FactSet expects semiconductor earnings to rise 130 percent, which on our arithmetic is about 43 percent of the index’s growth. The energy sector, boosted by oil averaging $85.68 a barrel in the quarter against $64.97 a year earlier, is expected to grow 114 percent, another 16 percent of the total. Chips and oil together account for about 60 percent.

Take chips and oil out, and the rest of the index is still expected to grow about 14 percent. That is respectable, but FactSet’s own breakdown shows how much of the headline depends on a small number of companies. Technology earnings are expected to rise 65.0 percent, but only 24.4 percent without semiconductors. Communication services are expected to grow 51.5 percent, but 12.0 percent without Meta and EchoStar. Financials are expected to grow 3.0 percent, health care 5.5 percent and consumer staples 2.9 percent.

These figures are estimates. Micron’s latest quarter ran 14 weeks rather than 13, which flatters the comparison, and Nvidia’s number stays an estimate until it reports.

The most cyclical product in technology

Micron’s figures are remarkable. Revenue in the quarter was $54.2 billion, compared with $11.3 billion a year earlier. Its non-GAAP gross margin reached 87.0 percent, against 45.7 percent a year ago. For the current quarter, Micron expects revenue of $61.5 billion, plus or minus $1.5 billion, and a margin of about 86 percent.

Memory chips have never been this profitable. Before this cycle, Micron’s highest margin was 61.4 percent, in the quarter to August 2018.

Memory is also the most cyclical corner of technology, because prices follow supply, and supply arrives in waves. After the 2018 peak, Micron’s margin halved to 30.6 percent within a year. After its 2022 peak of 47.8 percent, it fell to negative 31.4 percent a year later.

Micron’s management says this cycle is different in one important way. Chief executive Sanjay Mehrotra said the company does “not have line of sight to when supply and demand will return to balance,” and that he expects conditions to be “much tighter in calendar 2027 and 2028 than they were in 2026.” On the other hand, the finance chief expects “a more moderate rate of price increases” after this quarter, and TrendForce also expects DRAM contract price increases to slow.

Why it matters for the index

The S&P 500 trades at 19.0 times expected earnings for the next 12 months, close to its 10-year average of 19.1, on FactSet’s figures. That looks reasonable. But those earnings include memory margins never seen before, and energy profits that depend on an oil price about a third higher than a year ago.

A simple example: at its 2018 record margin of 61.4 percent, the revenue Micron is guiding to this quarter would produce roughly $15 billion less gross profit, in this quarter alone.

FactSet’s own forward estimates already show the boom fading. Analysts expect earnings growth of 27.6 percent in the fourth quarter, 19.1 percent in the first quarter of 2027, and just 2.3 percent in the second.

The case against us

There are good grounds for optimism. Micron says it has already signed agreements for most of its 2027 supply of high-bandwidth memory, the kind used in AI chips, at significantly higher prices. TrendForce expects that market to stay tight through 2027. Nvidia guides to revenue of about $108 billion for its current quarter. And growth of about 14 percent outside chips and energy is strong by most standards.

If supply stays as tight as Micron expects through 2027, the peak is still ahead, and today’s earnings estimates may turn out to be too low.

What it means for your money

• Know what your index fund is betting on. A third of this quarter’s expected profit growth comes from two chip companies, and about 60 percent from chips and oil.

• Peak margins are not a run rate. A company earning record margins on a cyclical product deserves a lower valuation than one earning the same amount from a steady business.

• Watch prices, not volumes. For memory chips, the signal to look for is slowing contract prices, not falling demand.

What we are watching

• From mid-October: third-quarter results from the rest of the index, especially outside technology.

• 14 October: the September consumer price index.

• TrendForce contract price updates: whether DRAM price increases keep slowing.

• November: Nvidia’s results and outlook.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

Over the past three weeks, new limits have shown up: a regulator’s docket in Wisconsin, a forest in northern Finland, and the rules of the largest power grid in the United States. None of these has anything to do with chips, but all of them decide when a data center can actually start operating.

A power line back at the start

Project Lighthouse in Port Washington, Wisconsin, is one of the largest campuses in the build-out. It needs 1.3 gigawatts of power, according to utility records, and Vantage Data Centers is building it with Oracle as the tenant, as part of OpenAI’s Stargate program. Oracle’s website says customer delivery is expected in the second half of 2027.

Work on the buildings is going ahead. The hard part is the power. The campus needs new high-voltage lines from American Transmission Co., and those lines still need approval from the Wisconsin Public Service Commission.

The commission found ATC’s application complete in December 2025. Over the next eight months, ATC filed or refiled 564 documents, according to Aterio, a data-center research firm. On 6 August the commission voted unanimously to withdraw its finding that the application was complete. “The magnitude and volume of changes have created a barrier to parties and the public’s participation in this proceeding,” said Chair Summer Strand. On 10 September the commission closed the case without ruling on its merits, and on 18 September ATC filed again.

That reset the legal clock. Under Wisconsin law, the commission has 30 days to decide whether an application is complete, so its next decision is due around 19 October. After that, it has 180 days to rule, and the chair can extend that by up to 180 more.

According to Aterio’s report from late September, work on the new lines cannot begin before about March 2027. In its base case, the first part of the campus gets power in December 2027 and the full 1.3 GW in October 2028. In its extended case, full power slips to April 2029. Aterio notes that the commission used the extra 180 days in every comparable large transmission case, and that four 345-kilovolt lines received decisions 355 to 358 days after their applications were complete. These are Aterio’s estimates from public filings, not commitments from Oracle, ATC or the regulator.

The line is also getting more expensive. The current estimate is $2.48 billion to $2.72 billion, up from about $1.4 billion to $1.7 billion in the first application. About $1.1 billion of that is grid-stabilizing equipment that Oracle has agreed to pay for. “We Energies’ largest customer today is less than 100 megawatts, and, at the max, this could be 3,500,” Tom Content of the Citizens Utility Board told Wisconsin Public Radio.

A forest in Finland

On 9 September, Google announced that it would invest at least 13 billion euros in Finland over two years, its largest single investment in Europe. The plan included new data centers in Kajaani, Vaala and Muhos, and a 22-year agreement to buy up to half the output of Fortum’s Loviisa nuclear plant. In other words, Google had brought its own power.

On 6 October, Finland’s Permit and Supervision Agency told Google’s project company, Tuike Finland, to stop preparatory work at Muhos and Kajaani immediately, and by 23 October at the latest. Google has confirmed that about 330 hectares were cleared at Muhos and just under 200 at Kajaani without the environmental impact assessment that the agency says should have come first. Tuike has until 14 October to respond.

In September, Google told the Finnish news agency STT that the trees were felled “in full compliance with the Forest Act.” On 6 October, the company said it had “fallen short of our own high standards” and would plant trees on 130 hectares at Muhos. A citizens’ initiative calling for stricter rules on data centers gathered 50,000 signatures in three days.

The Finnish dispute could be sorted out. The point is what it shows: a company can solve its power problem and still be held up by a different permit.

Paying for power that already exists

On 6 October, Google and Constellation Energy announced agreements covering 3,590 megawatts in PJM, the largest power grid in the United States. Of that, 890 megawatts will be new nuclear capacity from upgrades at 11 existing reactors in Illinois, Pennsylvania and New Jersey, under a 20-year contract. Constellation will invest more than $4.3 billion, with the first upgrade due by 2028. The other 2,700 megawatts will come from Constellation’s existing PJM fleet under a 15-year agreement. Constellation’s shares rose 12.25 percent that day.

For perspective, Constellation’s nuclear deals with Microsoft, Meta and Amazon add up to 2,646 megawatts.

The timing is no accident. PJM’s board has put forward a proposal that Constellation’s release calls “Bring Your Own Power”: new loads of 50 megawatts or more that do not bring their own supply by 1 June 2027 would be among the first to be cut in a power emergency. PJM filed the plan with federal regulators on 13 August and asked them to decide within 60 days. Meanwhile, PJM’s auction price for capacity has risen from $28.92 per megawatt-day for 2024/25 to its cap in each of the last three auctions, and on 29 September federal regulators put PJM’s emergency procurement plan on hold for five months.

Upgrading an existing reactor avoids most of what is slowing down everything else: finding new land, building new lines, getting permits and waiting in the queue to connect to the grid.

Why a delay costs more than time

A campus that is built but not yet powered still has to pay for its buildings, its equipment and its debt. Revenue starts only when the power does.

That matters most for companies that borrowed against a promised delivery date. Oracle has guided to capital spending of $90 billion to $95 billion in fiscal 2027, up from $55.7 billion in fiscal 2026, and S&P rates it BBB-, the lowest investment grade rating. In late September, Oracle sent a force majeure notice on its Project Jupiter campus in New Mexico, Bloomberg reported, and the cost of insuring its debt against default hit a record. Lighthouse is a different site with a different problem, but the result is the same kind of delay.

Delays also create winners. Every month a new campus waits, computing capacity that is already running gets scarcer, and so does power generation that already exists. That is what investors paid for in Constellation on Tuesday.

The case against us

Most of these problems can be solved. Google could file its explanation by 14 October, carry out the assessment and still keep its 2027 and 2028 schedule. The Wisconsin commission may decide faster than Aterio expects, and Oracle has other campuses that do not depend on this line. Demand for computing power is still running ahead of supply, so a late campus could still fill up quickly at good prices.

If Lighthouse gets its approval soon and Finland settles within weeks, this was a difficult month, not the start of a new trend.

What it means for your money

• Treat delivery dates as estimates. When a company says 2027, the permit calendar now matters as much as the chip supply.

• Power that already exists has scarcity value. Owners of running plants, especially nuclear, are being paid for the capacity they bring.

• Know who carries the time risk. Borrowers that promised capacity by a set date carry more of it than owners of computing power that is already running.

What we are watching

• Around 12 October: the end of the 60 days PJM asked federal regulators to take on its plan.

• 14 October: Tuike Finland’s explanation is due.

• About 19 October: the Wisconsin commission’s decision on whether ATC’s new application is complete.

• 23 October: Finland’s deadline for the work to stop.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

Many people think mortgage rates follow the Federal Reserve. This year shows why that is mostly wrong, and why it matters if you are buying a home, thinking about refinancing or holding housing stocks.

What really sets your mortgage rate

The Fed controls short-term interest rates. Since February, it has raised its target range once, by a quarter point, to 3.75 to 4 percent in September.

A 30-year mortgage is a long loan, so lenders price it off a long-term rate: the yield on the 10-year Treasury note. The mortgage rate usually sits about 2 points above that yield. The gap covers the lender’s costs and risks, and it has barely changed this year.

What has changed is the 10-year yield. Since late February, its weekly average has risen about 1.2 points, from 4.02 percent to 5.26 percent. The mortgage rate rose about 1.4 points. The Fed moved a quarter point.

So the thing to watch is not the Fed’s next move. It is where the 10-year yield goes.

What the rise costs

Take a $400,000 loan over 30 years. At February’s rate of 5.98 percent, the monthly payment for principal and interest was $2,393. At 7.40 percent, it is $2,770. That is $376 more every month, or about $4,500 a year, for the same loan.

Over the full 30 years, the total interest rises from about $462,000 to about $597,000.

Put another way, the monthly payment that carried a $400,000 loan in February now carries a loan of about $346,000. A buyer with a fixed monthly budget can borrow about 14 percent less. Compared with the record low of 2.65 percent in January 2021, the payment on the same loan is 72 percent higher.

These figures leave out taxes, insurance and fees.

Who is feeling it

Borrowers are stepping back. The Mortgage Bankers Association said refinance applications fell 8 percent in the week to 2 October and were 56 percent lower than a year earlier. Purchase applications were 15 percent below their level a year ago. The association’s own measure of the 30-year rate rose to 7.49 percent, the highest since November 2023.

“Very few homeowners have an incentive to refinance at these rates, and the jump in borrowing costs has caused many potential borrowers to step back from the purchase market,” said Joel Kan, the association’s deputy chief economist.

What would bring rates down

Mortgage rates will fall when the 10-year yield falls. Right now, three forces are pushing that yield up.

• Inflation worries. In the New York Fed’s September survey, households said they expect prices to rise 3.9 percent over the next year, the highest since May 2023.

• More Fed hikes. Minutes of the Fed’s September meeting show that most officials thought another increase would likely be appropriate by the end of the year. The next meeting is on 27 and 28 October.

• Heavy borrowing. Reuters points to more Treasury issuance and heavy borrowing by tech giants to pay for the AI build-out. On Thursday, the Treasury sold 30-year bonds at 5.618 percent, the highest yield at a 30-year auction since 2000.

Forecasters keep expecting relief. Strategists in a Reuters poll this week put the 10-year yield at 5.00 percent by the end of the year. But the same poll shows that they have underestimated its rise in nine straight monthly surveys this year.

The case against us

Rates can fall quickly. If the war ends, oil prices drop or the economy slows sharply, the 10-year yield could fall and take mortgage rates with it. It has happened before: in late October 2023, the 30-year rate peaked at 7.79 percent. By mid-December, it was 6.95 percent.

Our point is not that rates must keep rising. It is that the Fed is the wrong thing to watch.

What it means for your money

• Watch the 10-year yield, not the Fed. This year, mortgage rates have moved with the 10-year, about 2 points above it.

• Think in payments, not rates. On a $400,000 loan at today’s level, each 1-point move in the rate changes the monthly payment by about $270.

• Refinancing later is not guaranteed. Buying now and refinancing later only works if long-term yields fall, and refinancing costs money.

• If you own housing stocks, track the same numbers. The 10-year yield and the weekly mortgage data say more about housing demand than the Fed’s next move.

What we are watching

• Every Wednesday: mortgage applications from the Mortgage Bankers Association.

• Every Thursday: Freddie Mac’s weekly mortgage rate.

• 14 October: US consumer prices for September.

• 27 and 28 October: the Fed’s next meeting.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

A similar operation on 10 September drew $10.5 billion of offers, and the one in August drew $7.4 billion. On Treasury records going back to 2000, no buyback of bonds this long has ever drawn more.

Buybacks are a regular part of market housekeeping. The Treasury buys older, less-traded bonds so that holders have a way to turn them into cash. When the offers are big, many holders would rather have the cash. We read 1 October as a measure of how much long-dated paper is looking for the exit.

Two caveats are worth remembering. The Treasury recently raised the most it buys in each operation, from $2 billion to as much as $6 billion, and a bigger buyer can draw bigger offers. Some holders may also be swapping old bonds for newer ones. Yet the higher limit was already in place on 10 September, and that operation drew only $10.5 billion.

Oil fell. Long rates rose anyway.

On Monday 5 October, WTI crude fell 1.8 percent to $89.43 a barrel after Saudi Aramco cut its prices for buyers in Asia. The 10-year Treasury yield still closed at 5.31 percent, its highest close since 2002.

It was not a one-day oddity. Between 18 September and 5 October, WTI fell from $100.30 to $89.43. Over the same days, the 10-year yield rose from 5.01 to 5.31 percent and the 30-year from 5.34 to 5.66 percent. The 2-year barely moved, from 4.76 to 4.84 percent. Almost all of the movement was at the long end.

The economy did the work instead. S&P Global’s survey of US service companies reached 58.8 in September, its highest since July 2021, with input costs rising at the fastest pace in almost four years. The prices index in ISM’s services survey rose to 74.0, its highest since July 2022. Strong business activity with rising prices is exactly what long-bond holders worry about most.

The foreign buyer has a better offer at home

For many years, Japanese savers had to go abroad for a decent return. That is no longer the case. On 5 October, Japan’s 10-year government bond yielded 3.085 percent and its 30-year 4.166 percent, on Ministry of Finance figures.

The change is already showing up in Europe. Japanese investors held about 23 trillion yen of French bonds in July, roughly $145 billion, on a Bloomberg estimate. One Tokyo firm, Sumitomo Mitsui DS Asset Management, sold all the French government bonds in its global bond funds and moved into German Bunds and short-dated Japanese bonds, Bloomberg reported. Banque de France Governor Emmanuel Moulin has warned that France risks being “gradually strangled by rising interest rates,” and the euro fell to $1.1161 on Monday, its lowest since May 2025.

We have no figure showing Japanese money leaving US Treasuries. But the arithmetic pulling it out of French bonds applies to every foreign long bond, and Treasuries are the biggest of them.

The borrowers want long money too

Fed Chair Kevin Warsh offered another reason at his 16 September press conference, pointing to the biggest cloud computing companies. “The so-called hyperscalers are out in the market raising funding,” he said. “And so the competition for capital is real, and I think it partly explains the increase in yields.”

So the long end has fewer natural buyers and more borrowers competing for money. Every long bond the Treasury sells has to find a buyer, at a price.

This week’s test

The Treasury sold $58 billion of 3-year notes on Tuesday at 4.932 percent, the highest yield at a 3-year auction since May 2006, on TFTC’s tally. Investors bid $2.62 for every dollar on offer, and indirect bidders, a group that includes foreign buyers and funds, took 57.6 percent.

Next come the long bonds: $39 billion of 10-year notes on Wednesday 7 October and $22 billion of 30-year bonds on Thursday 8 October, both at 1 pm New York time. We will watch three things: whether the yield comes in above where the bond traded just before the sale, which traders call a tail; how much indirect bidders take; and how much is left to the dealers, who buy what nobody else wants.

The case against us

A lot of money is going into bonds. Bond ETFs took in $23 billion over five days, 62 percent of all ETF inflows, Bloomberg’s Eric Balchunas counted on 1 October. At around 5.3 percent, the 10-year pays roughly what many investors expect from stocks, a point we take up separately.

The buyback has a technical side, as we said. If Wednesday’s Fed minutes sound softer, or the consumer price report on 14 October is weak, long yields could fall quickly. And if both auctions this week go smoothly, the buyer problem is smaller than we think.

What it means for your money

• Long bond funds carry the risk. When long yields rise, long bond funds fall the most. The iShares 20+ Year Treasury fund (TLT) lost about 10 percent on price in the third quarter.

• Mortgages follow the long end, not the Fed. A 30-year mortgage is priced off long yields, so it can get dearer even while the Fed holds still.

• Short Treasuries still pay. On Monday, Treasury bills and notes out to two years paid between about 4.0 and 4.8 percent, without the big swings of long bonds.

What we are watching

• 8 October: the 10-year and 30-year auctions.

• The next long-end buybacks: whether offers stay this high.

• 14 October: the September consumer price index.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

For most of the last 30 years, it was the other way around. Stocks earned more than safe bonds paid, and that extra was the reward for taking on more risk. Economists call it the equity risk premium. On these numbers, it is gone.

How unusual this is

The bond has paid more than the S&P 500’s earnings yield in all but one month since August 2023, so this is not new. What is new is the size. At 1.54 points, the gap is the widest since July 2002, leaving aside 2009, when profits collapsed and briefly pushed the earnings yield below 1 percent.

Dividends show an even bigger gap. The S&P 500’s dividend yield is only 1.04 percent, so the 10-year pays 4.27 percentage points more. That is the widest gap since November 2000. The widest on our chart, which starts in 1995, was 5.49 points in January 2000.

Those dates matter. The last time bonds paid this much more than stocks, investors were betting on growth that took years to arrive. The S&P 500 fell in 2000, 2001 and 2002.

The case for stocks anyway

There is a good case on the other side, and it rests on profits.

The 3.77 percent is based on profits already reported. Analysts expect more. On FactSet’s figures, the S&P 500 traded at 19.0 times expected earnings for the next 12 months at the end of September, a little below its five-year average of 19.8 and close to its ten-year average of 19.1. Turned around, that is an earnings yield of about 5.3 percent, roughly what the 10-year pays.

So on expected profits, stocks and bonds are about level. On reported profits, bonds come out ahead. The gap between those two numbers is the earnings growth investors are counting on.

Earnings can also rise with prices, something a bond coupon cannot do. Against the 10-year inflation-protected Treasury, which yielded 2.93 percent on 30 September, its highest since November 2008, stocks still earn a little more. And demand in the economy is strong: S&P Global’s survey of US service companies came in at 58.8 in September, its highest since July 2021.

Why we are cautious

When the safest asset in the world pays more than 5 percent for ten years, every dollar of future profit is worth less today. That hits hardest at companies whose profits are furthest away.

Forward figures are only a promise. If profits fall short, the forward earnings yield moves back toward the reported one, and the gap with bonds widens. The economy under the index is also uneven: the Conference Board’s consumer confidence index fell to 81.9 in September, its lowest since 2014, even as service companies boomed.

The index itself is narrow. From 5 August to Monday, the Dow fell 5.7 percent while the Nasdaq rose 4.2 percent, and Nvidia closed at a record $238.90 on Monday. A few large companies carry the index’s valuation, which means they also carry much of its risk.

What it means for your money

• Know what you are being paid for. On reported profits, owning the S&P 500 now earns you less than lending to the US government for ten years. You come out ahead only if earnings growth arrives.

• Safe yields are real competition. On Monday, Treasury bills and notes out to two years paid between about 4.0 and 4.8 percent.

• Check your concentration. If you own a broad index fund, a large share of it sits in a handful of the most expensive companies.

What would prove us wrong

Fast profit growth would narrow the gap from the stock side. Third-quarter results begin in mid-October, and a strong season would lift the earnings yield. Falling bond yields would close it from the other side. A weak consumer price report on 14 October, or signs that the Fed is less likely to raise rates, could do that quickly.

What we are watching

• 7 October: the Fed minutes and the 10-year auction.

• 14 October: the September consumer price index.

• From mid-October: third-quarter earnings, and whether forward estimates hold.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

At first sight, fiscal 2026, the year to 30 June, was extraordinary. Revenue rose 132 percent to $25.5bn. Income before tax rose from $21.3m to $109.5m. Operating cash flow was $1.22bn, against $152m the year before.

We read the filings behind each of those numbers, and each one needs a second line.

Revenue isn’t the best measure here

A bullion dealer’s revenue is mostly the price of the metal passing through its hands. A better measure is how much metal it sold. Gold ounces sold rose 23.8 percent, from 1.64 million to 2.03 million. Silver ounces were flat.

The rest of the revenue jump comes from higher prices, acquisitions, and one accounting feature worth knowing. Gold.com hedges its inventory by selling metal forward, and those forward sales are booked as revenue even though they earn almost no gross profit. The 10-K says that, excluding an $8.3bn increase in forward sales, revenue rose $6.2bn. On our arithmetic from the same filing, about half of fiscal 2026 revenue was forward sales.

One quarter made the year

Look at income before tax quarter by quarter and you can see where the year came from. July to September 2025: a loss of $0.3m. October to December: $15.8m. January to March 2026: $81.8m. April to June: $12.3m, a little below the $13.0m of the same quarter a year earlier, after a $6.3m charge for contingent acquisition payments. On the company’s adjusted measure, that quarter was up 29 percent.

January to March was the quarter in which silver hit a record $118.45 on the LBMA benchmark, on 29 January. Frightened and excited buyers are this business’s best customers, and in that quarter they arrived all at once. It was also the first quarter to include Monex, which Gold.com bought in January. The other three quarters earned $27.8m between them.

The cash flow that came mostly from Tether

Operating cash flow of $1.22bn gives the impression of a business throwing off cash. The cash flow statement itself shows where it came from: an increase of $1.58bn in deferred revenue and other advances, of which $1.45bn came from related parties. Take the related-party advances out and the year’s operating cash flow is below zero.

Nothing about this breaks an accounting rule. It does mean the headline number is mostly advances from a related party, not cash the business earned.

One counterparty, four roles

That related party is Tether, the firm behind the largest dollar stablecoin. Its relationship with Gold.com has four parts, and all four are set out in the filings.

It is a shareholder. Tether holds a 12.6 percent stake. Its Schedule 13D says it acquired the shares “with the intent to exercise control” of the company. The same filing says it has no current plans or proposals for a takeover or similar steps beyond what it describes, though it may form one. Tether paid $44.50 a share for $150m of new stock in February and May, a Tether nominee joined the board on 16 March, and on 3 September Tether bought another 100,000 shares at about $39.

It is the main funder. At 30 June Gold.com owed Tether $1.454bn in payables and advances, against nothing a year earlier. That is about 1.7 times the equity that belongs to Gold.com’s shareholders. Over the same year, bank borrowings went from $345m to zero, though the bank facility of up to $427.5m is still in place, undrawn.

It is a customer. Tether bought $1.19bn of metal from Gold.com during the year, all of it after March.

It is a vault client. The 10-K says Tether uses Gold.com’s secure storage and logistics services. As part of the share deal, Gold.com used $20m of the proceeds to buy Tether’s XAU₮ gold token, in April. At 30 June it was carried at $17m.

The one set of terms on file

The only Tether lease terms Gold.com has filed sit in an appendix to its February share purchase agreement with Tether. They describe a gold lease facility of $100m at 1.75 percent a year, with draws of up to 18 months. One line is headed “Lessee Redemption Right”. It reads: “Upon 60 days prior notice at the option of Tether.”

That was a term sheet, conditional on a definitive agreement. Tether’s May 13D says the two sides have since entered into metals leasing arrangements, but no definitive lease agreement has been filed. The 10-K describes the company’s metal leases in general as short-term arrangements, with terms set when each lease starts and settled in metal or cash. It doesn’t say how quickly the $1.45bn owed to Tether at 30 June could be recalled. For a holder, that’s one of the most important terms to know, and it isn’t in the filings.

The week in September

On 9 September Grant Thornton signed a clean audit opinion on the fiscal 2026 accounts, as it had every year since 2015. On 14 September the audit committee dismissed it and appointed KPMG for the new year. The filing reports no disagreements.

The same day, the chief financial officer handed in his resignation, effective 18 September. The company calls it a planned retirement, and he will stay on as a consultant for a year. The controller took over. She spent twenty years at Grant Thornton, including as an audit partner, and has since been an audit partner at RSM and a chief financial officer elsewhere.

None of this is evidence of a problem, but all of it is a reason to read the next 10-Q closely.

Why the stock can fall when gold rises

This is the part that’s easy to miss. A dealer is not a gold fund. In our reading, when prices climb steadily, customers who bought coins and bars start selling them back. That is volume the dealer has to take at thin margins. Meanwhile the cost of carrying and financing inventory rises with the price of the metal. The best quarters come from fear and spikes. A gold price that just keeps rising can be bad news for the stock.

The customer numbers are worth watching. Direct-to-consumer active customers were 160,700 in the April to June quarter, against 170,600 a year earlier. The 2026 figure includes Monex, which Gold.com bought in January.

The case for the stock, from the same filings

The same filings also make the case for the stock.

Rising prices didn’t hurt every quarter. Between October and December 2025, when the LBMA gold price rose about 14 percent, income before tax nearly doubled to $15.8m from $8.0m a year earlier.

The spike quarter came before most of the Tether money. At 31 March the company owed Tether $362.6m; the big step up came between April and June.

Leasing is cheap on paper. Borrowed metal and leases, with most of the leases coming from Tether, cost a weighted 2.2 percent last year, against an effective 8.4 percent on the bank facility.

There’s a backstop. Gold.com ended June with $578m of cash and an undrawn bank facility of up to $427.5m running to September 2027. It also held $2.36bn of inventory, and leased metal can be returned as metal.

Tether keeps buying shares. A bid at a premium would reward holders, though its 13D says no such plan exists today. The board also declared a $1.00 special dividend alongside the regular 20 cents, payable on 28 September to holders of record on 16 September.

What the full report adds

The report rebuilds owner cash flow line by line. It values the company on four scenarios, including one where gold wins and the stock loses and one where Tether’s funding pulls back. It also names the levels of quarterly profit and Tether balance we are watching. Those stay in the report.

What to keep an eye on next

The next 10-Q. Income before tax in a quarter without a price spike, and the first numbers under a new CFO and a new auditor.

The Tether balance. Whether it grows or shrinks, and whether a definitive lease agreement, with its recall terms, is ever filed.

Retail customers. Active customer counts, now that a record silver quarter is in the past.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

To find out who pays if a data center runs late, you have to read the contracts. We took what has been reported about this one and put it next to three other AI infrastructure deals whose terms are public, in company filings, a company announcement or a federal order.

On our reading, all four deals point the same way: when an AI data center is late, the cost lands on whoever can’t leave.

What the notice does

A force majeure clause excuses a party when something beyond its control stops it doing what it promised. Think of a war, a flood, or a government refusing a permit.

On 24 September Bloomberg reported that Oracle had sent one to the developer of Project Jupiter, its planned campus in Doña Ana County, New Mexico. The developer is Stack Infrastructure, which Blue Owl owns. Reuters then reported the terms that matter, from a person familiar with the matter.

According to that person, securing power for the site is Oracle’s responsibility, and Oracle cannot terminate the lease under any circumstances. Blue Owl has about $3 billion of equity in the project, and Oracle pays the debt costs. During construction, Blue Owl earns a 9 percent yield on that equity. After completion, the levered yield is expected to be about 11 percent.

By invoking force majeure, the same person said, Oracle extends the period in which it pays the lower rent. Blue Owl still gets the higher rent for the full planned term. It just starts later.

So Oracle isn’t walking away. Under this lease, as that person describes it, it can’t. It is arguing that the clock on the full rent shouldn’t start if the power is late.

Why the power is the risk

In April Oracle switched Jupiter’s power plan to up to 2.45 gigawatts of Bloom fuel cells, replacing the gas turbines it had planned. Fuel cells still need gas.

The pipeline meant to deliver it, proposed by a subsidiary of Energy Transfer, had a target in-service date of 15 August. After repeated refusals from New Mexico’s State Land Office, the operator’s filing now says 1 February 2027. An analyst at Energy Aspects told Bloomberg that date is probably not realistic, expects the second half of 2027 at the earliest, and sees a risk of cancellation if progress stalls. In May, Oracle itself told federal regulators that time was of the essence for that pipeline.

So the part of the project most at risk of running late is the power, and power is Oracle’s job, according to Reuters’ source. Oracle says the campus remains on its planned schedule. The pipeline is already behind its own target.

That’s why the notice matters. A delay that is the tenant’s own doing would normally be the tenant’s cost. A delay caused by a state agency refusing a pipeline route is the kind of event force majeure clauses are written for. Whether that argument holds depends on a contract nobody outside the parties has seen.

Who carries the months

Put the reported terms together and the three companies are in very different positions.

Oracle carries the largest share of any delay. It pays the debt costs on a campus that earns it nothing yet, and if the notice fails, it starts paying the full rent sooner. The brokerage William Blair noted that Jupiter contributes no revenue to Oracle this fiscal year, so the near-term accounts barely change. The cost shows up as cash going out before any cash comes in.

Blue Owl’s cost is smaller. On the reported figures, a year of delay postpones a step up of about two percentage points on about $3 billion of equity, roughly $60 million a year on our arithmetic. Bloomberg reported that funds Blue Owl manages committed the equity, so on our reading most of this falls on the investors in those funds, and Blue Owl’s shareholders feel it through the firm’s fees. The higher rent still runs for the full term, so the money is postponed, and the $60 million a year is the price of waiting.

The lenders, about 20 banks behind an $18 billion loan, are owed debt costs that Oracle pays, according to Reuters’ source. Bloomberg reported, citing a person familiar, that the notice is not an event of default (the kind of breach that lets lenders demand their money back), and that the debt was already trading below 90 cents on the dollar. A gauge of Oracle’s credit risk hit a record on 24 September. On our reading, with Oracle paying the debt costs, the loan is largely a bet on Oracle’s credit. We haven’t seen a loan price since the notice.

Bloom Energy is the one we can say least about. Its fuel cells need the gas line, and its terms with Oracle aren’t public. Bloom said on 24 September that Oracle remains committed to its contract.

One day of share prices can’t sort any of this out, least of all a day when long Treasuries also fell about 1.3 percent. Our concern is wider than this notice. Oracle itself calls notices like this commonplace, so we doubt it will be the last.

How big buyers write their contracts

Now look at deals where the terms are public.

On 5 June SpaceX signed a contract to supply Google with computing capacity built on about 110,000 Nvidia GPUs. SpaceX’s filing with the SEC sets the fee at $920 million a month from October through June 2029. Add that up and the headline is about $30 billion.

Two clauses change what it’s worth. If SpaceX hasn’t delivered the committed GPUs by 30 September, then after a one-month grace period Google can walk away, or keep what has arrived and pay less. That exit is Google’s alone. The second clause runs both ways: after 31 December either side can end the contract on 90 days’ notice. SpaceX’s own quarterly report says its cloud contracts generally work like this.

On our reading, if SpaceX delivers on time, the part neither side can escape is about six months of fees. That’s roughly $5.5 billion, against a headline near $30 billion.

If Google uses that exit, it stops paying, and the capacity goes back to SpaceX to use itself or rent to someone else. Until the GPUs are delivered, though, only Google can walk.

Meta took a similar approach at Hyperion, its campus in Louisiana. Funds managed by Blue Owl own 80 percent of the joint venture that owns the campus, Meta owns 20 percent, and part of the money came from bond investors. Meta’s lease has a four-year initial term with options to extend, which Meta’s announcement said gives it “long-term strategic flexibility”. To balance that flexibility, Meta guaranteed part of the campus value for the first 16 years. The guarantee is capped and only pays out if certain conditions are met.

On our reading, both buyers kept a way out: Google if the capacity is late, and Meta if it no longer needs the campus after four years, at the price of a capped guarantee. Google’s other exit, 90 days’ notice after 31 December, is one SpaceX holds too. Now notice where Blue Owl sits. At Hyperion its tenant can leave after four years. At Jupiter, as reported, its tenant can’t end the lease early. Same manager, opposite ends of the risk.

What Oracle signed

Oracle is the other kind of company in this story. It sells computing capacity to customers such as OpenAI, and it leases buildings to do it.

Its quarterly report for the three months to 31 August shows $288 billion of data center leases that are signed and haven’t started. They run for 15 to 19 years and are generally expected to start between now and Oracle’s fiscal 2029, which ends in May 2029. In May 2024, when the figure covered operating leases only, it was $22.9 billion.

For scale, every remaining payment on the leases Oracle has already started adds up to $63 billion. The pile waiting to start is more than four and a half times that.

Big signed pipelines aren’t unique to Oracle. Microsoft reported $329 billion of leases not yet started at the end of June, some of them subject to conditions, with terms of up to 20 years. The differences are proportion and timing. Microsoft’s pile is about three times what it still owes on its running leases, and it starts over fiscal 2027 to 2033. Oracle’s is more than four and a half times, and it is generally expected to start by the end of May 2029, less than three years away.

Set that against what Oracle has sold. Its contracted revenue still to be delivered is $664 billion, and it expects to recognize about 84 percent of it within five years. About 16 percent, roughly $106 billion, falls after that. The leases waiting to start run 15 to 19 years from their start dates. New contracts may well fill those later years. The rent for them is already signed.

There’s one more detail worth watching. Between February and May this year the total barely moved, from $261 billion to $260 billion. Over the same three months, the far end of the window in which those leases are generally expected to start moved from fiscal 2028 to fiscal 2029. The filing doesn’t say whether new leases were added or older ones slid. It’s the right question to ask in December.

And one line has gone. In February, Oracle’s quarterly report said one of these leases came with an Oracle guarantee of up to $2.2 billion of the landlord’s borrowing, maturing in September 2026. By May the annual report put it at up to $3.3 billion. The August report doesn’t mention it. We don’t know which lease it is, and we’ll look for what happened to that loan.

The grid has the same problem, with one dollar in it

The same fight is happening one layer down.

PowerHouse Hillwood is developing a 1.8-gigawatt, $20 billion data center campus in Joliet, Illinois. Its agreement with ComEd, the Exelon utility, covering the transmission work requires credit support. A letter of credit is a bank’s promise to pay if its customer doesn’t. The developer says a letter of credit for $1 met its first credit requirement.

ComEd tried to cancel the agreement. On 22 September FERC, the federal power regulator, rejected the cancellation and declined to decide the credit dispute, which is before a federal court. Four commissioners added concurring statements, two of them jointly. One called the idea that $1 is appropriate security “an embarrassing legal fiction”, and said ratepayers, among others, bear the real risk of the project. Another said agreements like this exist so that project risk stays with the developer, not the public.

So far, the utility has tried to leave and been refused. For now, the security behind a gigawatt-scale grid agreement is one dollar, and whether that’s enough is for the court to decide. Grid operators have until mid-November to answer FERC’s June orders on how very large users connect to the grid, and those answers will shape how future agreements are secured.

The rule we take from this

Four deals, one pattern on our reading. The buyers kept a way out: Google can walk if its GPUs are late, and Meta signed a four-year first lease. The other side carries the wait: Oracle with a lease it reportedly can’t end, a developer waiting on its step up, SpaceX until its GPUs are delivered, and a utility that tried to leave and was told no by FERC, with ratepayers behind it.

When a data center is late, someone still pays: whoever can’t leave.

Three practical things follow.

Read the exit before the headline. On our reading, if the GPUs arrive on time, this $30 billion contract is about six months of committed fees plus an option on the rest, and both sides hold that option.

Read both sides of a backlog. Oracle’s $664 billion of contracted revenue arrives as capacity is delivered, mostly within five years. Its $288 billion of leases are generally expected to start by fiscal 2029 and then run 15 to 19 years. At Jupiter, on the reported terms, Oracle pays the debt costs and a lower rent while it waits, so a late building costs it money before any revenue arrives.

Watch the dates. The next ones are public.

• 30 September: the delivery date in the SpaceX and Google contract, with a month’s grace to the end of October.

• Mid-November: grid operators answer FERC on connecting very large users.

• 23 November: New Mexico’s deadline to decide Jupiter’s air permit.

• Mid-December: Oracle’s next quarterly report.

• 31 December: after this date either side of the SpaceX and Google contract can give 90 days’ notice.

• 1 February 2027: the pipeline’s target in-service date.

What would make us wrong

Oracle calls notices like this commonplace in developments of this scale, says Jupiter remains on its planned schedule, and says such notices do not, by themselves, establish a project delay. Blue Owl says the notice doesn’t change the financial commitments to the project. Bloom says Oracle remains committed to its contract. On 17 September New Mexico’s Supreme Court turned away two challenges and let permitting resume.

Oracle’s $664 billion is contracted revenue, which is a much stronger thing than a headline deal value, and it grew from $455 billion a year earlier.

Microsoft’s signed pile is larger than Oracle’s. Big lease commitments are how this industry builds. On their own they aren’t a warning.

The $1 letter of credit may be exactly what the agreement allowed at this stage. The developer’s case is that $1 met the initial requirement. FERC didn’t rule on it.

And SpaceX can leave the Google contract on 90 days’ notice too.

The test we will score

In Oracle’s next quarterly report, due around mid-December, we will print one paragraph: the total of leases signed but not started, the window in which they are generally expected to start, and whether Oracle still guarantees any landlord’s borrowing.

We’ll read it four ways, set out now.

• More than 5 percent below $288 billion, with payments on running leases up by at least half the fall: buildings are being handed over.

• More than 5 percent lower without that rise: leases may have been cut or renegotiated, and we’ll say so.

• More than 5 percent higher: Oracle is still signing, and we’ll report the new window.

• Within 5 percent either way: if the far end of the window has moved to fiscal 2030 or later, that’s the February to May pattern again, and we’ll read it as starts moving later while saying the filing can’t prove it. If it hasn’t moved, we’ll call it no change.

Whichever it is, we’ll also report whether any landlord guarantee is disclosed.

We’ll print the paragraph whichever way it reads.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

On 10 September OpenAI launched ChatGPT for Financial Services, a product built for bankers and research analysts. Look past the headline, though, and there’s a detail that doesn’t fit the story. OpenAI says it is building shared sign-in with S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva and Moody’s, so its users can reach data they already pay those firms for.

In other words, the AI will log in through the data owners. The product also hosts licensed data from Daloopa, PitchBook, LSEG News and Crunchbase, which makes it a bundler of data as well as a door to it.

That doesn’t make every data vendor safe. It changes the question. AI will be used in finance; that part is settled. What isn’t settled is which layer of the data business AI can go around, and which layer it has to pay for.

The draft is getting cheap. The proof is not.

Stanford’s AI Index tracks how fast the price of AI is falling. For a model performing at the level of GPT-3.5, the price per million tokens dropped from $20 in November 2022 to 7 cents in October 2024. That’s more than 280 times cheaper in under two years.

That is the drafting layer: summarizing filings, pulling out numbers, writing the first version of a note. It is turning into a commodity.

Checking is a different job. Before a fund can act on an answer, it needs data somebody is allowed to use, cleaned and verified. The owners of that data are spelling out that AI use falls under their licenses. This year Cboe told its data customers that putting its data into AI products may need additional licenses, and that any AI output which can substitute for its data needs a license of its own. Nasdaq’s AI policy says much the same in its own words: AI use sits inside the license, and the customer has to buy it.

There’s also an early warning on quality. A working paper posted last December, not yet peer-reviewed, used the 2023 launch of FactSet’s AI platform as a natural experiment. Analysts who adopted it wrote richer reports, drew on 40 percent more sources and worked faster. Yet their forecast errors rose by 59 percent. The authors put the extra errors down to the cost of digesting more information, not to speed. It’s one unrefereed study, but it’s a warning that more data and a faster draft don’t by themselves make a better answer.

Four layers, and the question that matters

We split financial data into four layers and asked one question of each: can AI go around it?

Public facts. Filings, statistics, anything already in the public domain. Yes, and cheaply. A business that mainly repackages public filings is now competing with a model that costs pennies.

Exchange data. Prices and trades, owned by the exchanges. No. The license is the product, and the exchanges have said in writing that the license covers AI.

Benchmarks. The indexes written into fund mandates and prospectuses. This is the hardest layer to go around. A fund can’t swap its benchmark just because a cheaper AI tool exists. Changing one takes lawyers, trustees and a letter to every client.

Licensed content and identifiers. News, and the codes that identify securities. Only with permission, and permission is exactly what OpenAI is negotiating.

Once you have the layers, the rest follows. Value moves toward whoever controls the license. The revenue most exposed is the kind priced per person, once one person with an agent can do the work of several.

Three companies in three different positions

FactSet sells the screen and the data behind it. So far its numbers are better than the story about it. Organic subscription value grew 7.1 percent in the year to May, driven mainly by data solutions and workstations. Its user count rose about 12 percent, mostly wealth managers. Client retention slipped to 90 percent from 91, while retention measured by subscription value stayed above 95 percent. FactSet also runs CUSIP Global Services, the securities identifier business, which sits in the layer our map puts behind a license.

More people are using FactSet, not fewer. What matters next is the price of each seat when one analyst with an agent does more, and how much of FactSet’s revenue still moves with the number of seats. The company says that share is smaller than it used to be. Its results on 30 September and its investor day on 10 November are where it has to show it.

MSCI owns the layer AI finds hardest to touch, and its numbers show it. Index retention was 97.5 percent in the second quarter. BlackRock extended its ETF license to March 2035.

Look one line lower and the picture changes. The fee MSCI earns on each dollar of ETF assets tied to its indexes fell from 2.43 basis points (hundredths of a percentage point) to 2.28 in the year to June, a drop of 6.2 percent. MSCI says the BlackRock amendment revises fees for some funds from January 2026, with more changes in January 2027, trading price for volume; the fee schedule itself is redacted in the filing. Asset-based fees still rose 26.6 percent in the second quarter. And of the $793bn those ETF assets grew by over the twelve months, $538bn came from markets rising and $255bn from new money, on our sum of MSCI’s own quarterly table.

What a moat like this is worth is the valuation question the report answers. The inputs are public: the fee trend, and how much of the growth came from markets.

LSEG is the cleanest test of the layer argument. In the first half, organic revenue in Data & Feeds grew 7.5 percent. Workflows, the business built around its Workspace desktop, grew 2.8 percent. The feed outgrew the screen.

The margin needs more care. Adjusted EBITDA margin, the share of revenue left as operating profit before depreciation and one-off items, rose 260 basis points at constant currency, which strips out exchange rate moves. LSEG itself says 140 of those came from a change to its SwapClear revenue share. In October 2025 it agreed to pay £1.15bn to cut the banks’ share of SwapClear’s revenue surplus to 10 percent. That payment is amortized outside adjusted earnings, so the benefit shows up in the adjusted margin and the cost doesn’t. LSEG’s interim report says the year-over-year uplift will largely reverse in the fourth quarter, because the benefit for all of 2025 was booked in the second half of 2025. The lower revenue share itself runs to 2045.

What the full report does

The report values each of the three companies on three scenarios, including one in which AI wins outright. It sets out what we think each company is worth, and the single result in 2027 that would tell us the framework is wrong. Those stay in the report.

Three things to check by 22 October

FactSet, 30 September. Does user growth hold near 12 percent and organic subscription growth near 7 percent now that OpenAI has a banking product? Does the company say anything about AI revenue, and how it is priced?

MSCI, 20 October. Watch the fee on each dollar of ETF assets. If assets keep rising while the fee keeps falling, the index business is growing on volume and markets rather than on price.

LSEG, 22 October. The trading statement covers revenue only. When margins come back with the full-year results, take the SwapClear change out before you compare.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

A more detailed perspective can be taken of Oracle’s force majeure notice concerning Project Jupiter. While it might be claimed that the primary tenant intends to vacate, Reuters, quoting a source, gives an alternative account. It is reported that Oracle cannot terminate the lease, is still obliged to pay the project’s debt costs, and pays reduced rent while the project is under development. Higher rent commences when the campus is completed and, as the source states, will continue for the original term.

The terms mentioned are from a private agreement, not one we have examined ourselves; they do serve to make the notice more informative. Nevertheless, an AI contract can remain in effect even if investor returns vary over time.

What the notice actually changes

Reuters reports that approximately $3 billion of the project’s equity is held by Blue Owl. Our source says the equity yield at the development stage is about 9%, and it is expected to increase to around 11% once the project is completed and leveraged. The two yields might be based on different calculations, so it would not be correct to subtract them and refer to the difference as ‘lost rent’. The newspaper later added, based on a source, that the project could be postponed by a year.

Oracle states that Jupiter remains on schedule; Blue Owl maintains that the notice has no effect on its financial obligations. According to Bloomberg, Oracle would only be able to secure a possible three-year delay in rent if both parties agreed to a power-related force majeure event. A one-year delay in the project and a possible three-year contract remedy are not identical. As the reports do not refer to the same phase or milestone, we should not regard ‘on schedule’ and ‘one year late’ as applying to the same date without further information.

The notice doesn’t show that any relief has been granted or that there was a default on the project financing, but it does raise a significant question concerning large AI projects—namely, whether income increases when construction is completed, when power becomes available, or only when the customer accepts usable capacity? It can happen that one requirement is fulfilled while waiting for another.

The return from waiting

The method we use to determine whether a delayed project is a problem can be explained as follows. Let V represent the present value of the expected net cash flows to an investor under the original timetable. If the entire stream is shifted by one year, keeping the same term and level of risk, the investor will receive an additional amount of net development-stage cash equal to C at the end of that year after having paid for financing and holding costs. When an annual rate of return of r is assumed, the new value becomes (V + C) divided by (1 + r). This is based on the assumption that no additional capital is required.

The result will depend on the specific details; if C is less than r multiplied by V, then the delay will reduce the value. However, if the cash generated by the development covers the return required for waiting and the original operating term remains unchanged, merely waiting will not lead to a fall in value. This discussion is solely concerned with the timing of the cash flows and not with the Jupiter contract itself.

For example, where the future net operating distributions are worth $100 at the present time and the required return is 12 percent, a further year which brings $8 in net development distributions results in a new value of $108 divided by 1.12, that is $96.43; if the additional year yields $14, the new value is $114 divided by 1.12, or $101.79. In the first case, there is a loss of $3.57 while in the second there is a gain of $1.79. In both situations, the final operating stream remains unchanged.

The calculation has not taken into account any construction delays, changes in financing, tax implications, rising prices, counterparty risk, or any spending that might be delayed; in an actual project, these factors could affect the result. Moreover, the examples of $8 or $14 are not intended to convert Reuters’s reported 9% and 11% yields into dollar amounts, since we do not have the same-base cash flows or a full breakdown to perform such a conversion.

It is here that the common defense—”the rent is still owed, so no change has taken place”—breaks down. This argument supposes that the value of a dollar does not depend on when it arrives or on what must be financed during the delay period. Similarly, the other widespread claim—”the higher rent begins later, so the equity is reduced”—omits the development-stage payment and the retained operating term. The situation cannot be assessed from the headline alone; instead, we can identify the missing variables.

The three calendars behind four tickers

Begin by looking at $ORCL. Oracle could obtain some relief regarding the timing of when it must begin paying higher operating rent, even though it will still have to pay development rent and debt costs as previously reported. This situation is relevant only when Oracle starts receiving payments from its own customers. It is not necessary for the timing of the customer contracts to coincide with the landlord’s schedule.

The cash-flow statement that Oracle produced for the first quarter of fiscal 2027 illustrates the impact of timing on the entire company. It shows operating cash flow of $23.103 billion, with this figure comprising an $11.363 billion rise in deferred revenue due to customer prepayments, which have a substantial financing component. Capital spending amounted to $28.499 billion, resulting in negative free cash flow of $5.396 billion.

If the other cash items remained unchanged and only the rise in prepayments were eliminated, operating cash would have been $11.740 billion, and, with the same level of capital spending, the difference would have been- $16.759 billion. It should be stressed that this is merely a sensitivity analysis and not the actual or normalized free cash flow. The money tied up in prepayments is real cash and is subject to future delivery obligations. The report does not state how much of this relates to Jupiter. The main issue is how much future construction can be financed by customers on terms that are still viable should one campus open later.

Look at $OWL. As Reuters’ source states, the $3 billion figure refers to the project's equity, and Bloomberg indicates that funds managed by Blue Owl are providing this equity. The shares in the publicly listed manager do not directly reflect all of the gains or losses from each of the managed funds. To incorporate a Jupiter scenario into $OWL’s earnings, investors must be aware of the breakdown of fund capital, any principal capital reported on the company’s balance sheet, fee earnings, guarantees, and performance fees. Even if the company’s earnings estimate does not change by the same amount, a longer period of lower rent can affect fund returns.

Oracle has stated that it intends to use Bloom fuel cells as its on-site power source, with an installed capacity of up to 2.45 gigawatts. It must be understood, though, that the planned capacity does not represent the amount of power actually used or the amount of computing power provided. The fuel cells still require natural gas, as is confirmed by Oracle’s permit application. Investors should be made aware of the dates when the equipment is manufactured, delivered, accepted and paid for, as well as what will happen in the event that there are delays with the gas infrastructure. Placing an order and receiving steady short-term cash receipts are not the same thing.

$CRWV should be included in the comparison rather than in the Jupiter ownership chain. If competing sites take longer to open, existing computing capacity could become scarcer. A company that has development commitments and financing exposure elsewhere can also be affected by the same bottleneck. A delay at one campus is not a clear long or short signal for the entire peer group.

What would settle it

Our understanding would be revised in light of documents that illustrate the key milestones—specifically, the exact conditions under which operating rent would begin, what Oracle is owed during the development phase, whether costs or debt service can be passed on, and the protections suppliers have if the installation is delayed. The significant document could be a payment schedule or a completion certificate, not merely a press release announcing the total gigawatts.

Once power permits have been obtained, a funded gas connection has been secured, and a commissioning plan has been prepared in advance, delivery can be considered more certain. A binding agreement specifying the development payments and the original full operating term would help establish the argument that the contract will survive. Conversely, new equity injections, alterations to the return waterfall, or the failure to meet supplier milestones would make it more costly to wait. The private loan rates given early on, ranging from 89 to 91 cents, together with a later figure below 90 from Bloomberg, should prompt further investigation, not be taken as proof of a sale, a default, or an actual loss.

The demand is not the weakest aspect of this story. In its most recent quarter, Oracle reported $664 billion in remaining performance obligations and was delivering an additional 850 megawatts of data-center capacity across its business. These figures are substantial counterarguments to the notion that the entire construction has come to a halt. However, they do not indicate which party receives what while Jupiter waits.

This is the real investment question behind the term ‘force majeure.’ We need to know the timing of cash flows, who is responsible for each obligation, and what the return on capital is during the waiting period. A contract might guarantee the total number of rent payments, but it does not tell us if those payments provide a good return.

We looked at the twenty-year auction on 16 September. We compared it with seventy-seven auctions held since that maturity was introduced, and found that its foreign portion was the lowest ever recorded.

On Wednesday, the Treasury issued $ 70 billion in five-year notes. We have compared this sale with each of the five-year auctions since April 2008, for a total of 221 auctions. We worked out all the figures using the results published by the Treasury.

The four readings

The auction achieved a high yield of 5.033 percent, the highest of the 221 auctions in this series, and, over the period covered by Treasury data, the five-year note has never reached a yield of 5 percent at auction.

The bid-to-cover ratio—the ratio of bids received to notes sold—was 2.21; it last reached that low on December 26, 2018.

The contribution from foreign and official sources—comprising central banks and overseas institutions—was 54.3 percent; it had last been at this level on March 25, 2020.

The dealers, having to take up any remaining notes, ended up at 15.8 percent, the highest figure since May 28, 2024.

Now, the part the headlines will not do

The majority of the reports will highlight foreign demand having reached its lowest level since the pandemic; although this is the case, it doesn’t give the full picture.

The 221 auctions involving a foreign share were ranked, and the auction in question came 87th. A foreign share of 54 percent has been typical since 2008. What is notable is that this is the lowest figure of the decade, since demand from foreign buyers was particularly strong in the early 2020s.

The bid-to-cover ratio paints a different picture: when ranked in the same manner, the auction on Wednesday was the 11th-weakest of the 221. The majority of the poor results occurred in 2008 and 2009 during the financial crisis, with another one in December 2018.

The measure that is genuinely historic is receiving the least amount of attention, whereas the other one is only noteworthy if you begin counting from 2020.

Why we checked the month before

On Tuesday, many people said the two-year auction had revealed a decline in foreign demand, with the figure having fallen from 66 percent to 58 percent. However, on checking, we discovered that the 66 percent figure for August was in fact the outlier, with September’s figure being nothing more than a return to the recent average.

For the five-year auction, we carried out the same check. In August, the foreign share was 61.5 percent, below the average of the previous twelve auctions, and there was no particularly strong month for Wednesday’s result to respond to.

This time, however, there is a genuine decline. The auction over a two-year period had not revealed that. It is precisely because we notice the difference that we examine the data.

What it adds to what we already had

The Treasury has now reached the second auction this month at which the highest yield in the series was obtained, and demand has been weak according to a measurable criterion.

On the 16th of September, the twenty-year bond cleared at 5.420 percent, the highest figure since that maturity was returned, while the foreign portion was the lowest of all 77 sales. On Wednesday, the five-year bond was cleared at 5.033, the highest of all the 221, with the bid to cover at its weakest level since 2018.

There are two stages and two distinct methods, but both tend in the same direction. Neither on its own would be very significant. Only when taken together are they less likely to be dismissed as merely a one-off event.

What would change our mind?

Wednesday was merely a weak session in a busy week if the bid-to-cover ratio of the next five-year sale in late October exceeds 2.35.

If the ratio drops below 2.25 again, it would indicate that the weakness lies in demand for this maturity, not merely in the timing of a single sale.

Available for checking, and the date is given.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

The index is still reaching new records, but the individual stocks make no such gains. That is the true story of this week.

On Monday, the Nasdaq climbed 2.26 percent to 27,122.09, making it the twenty-first time in 2026 that the index had set a record, the first such occasion since June 2 and also its largest single-day increase in seven weeks; on Tuesday it gained an additional 0.45 percent to reach its twenty-second record close. Meanwhile, the S&P 500 finished Tuesday without any change.

On Tuesday the stock reached $345.34 during the day, momentarily exceeding a market capitalisation of five trillion dollars, but ended the day at $339.75, giving it a market value of $4.96 trillion. It has risen 7.2 percent this month and 25 percent this year. The only company that has ever closed above five trillion dollars was not Apple.

All the previous points can be verified. The statistic which everybody keeps repeating now comes into play.

The claim, and why we did not simply publish it

It has been claimed that when the bank index is more than ten percent below its most recent peak whereas the general index is within one percent of its own peak, this situation last occurred in January 2000.

It’s an excellent line, but it entirely depends on two points which nobody mentions: which bank index and which high.

We therefore tried it in four different ways, using the daily closing prices of funds that an ordinary investor could purchase.

When compared with the bank fund’s highest intraday price over the past thirty sessions, its price is 10.34 percent lower. This lends support to the claim. Against its highest closing price over the same thirty sessions, however, it is 9.51 percent lower. The claim then fails.

It is 7.63 percent lower when compared with its highest closing price over the past thirty calendar days. This is much less than had been claimed.

When the same period’s figures for the overall bank fund, the regional bank fund, and the financials sector fund are examined, the figures show drops of 8.99, 8.65, and 6.42 percent respectively. None of these contradict the claim.

Of the six measures, only one shows a fall of over ten percent, and that is the one that compares today’s closing price with an intraday spike from five weeks ago. That is the most favourable comparison that could possibly be made.

Where the wording came from

There’s one more matter, and it concerns attribution, not arithmetic.

The initial observation is from a note written by a technical strategist and has been reported in two publications; one states that it is “more than ten percent below a recent high” while the other says that it is “more than ten percent below its thirty-day peak“.

The thirty-day version—this is the one currently being distributed—appears in just one of the articles, in the section headed ‘Summary’ by that author. We were unable to access the original institutional note.

It is possible that the exact phrasing which people are repeating was not actually what the analyst said at the time. Moreover, the comparison for January 2000 requires index data from the 1990s, which is behind a paywall and not freely available.

The version that does hold up

There is also a statistic from that same week which is better supported and so one that we should rely on.

On Monday the index increased by more than one percent and thus reached a point which was just one percent away from a record, even though thirty of its components hit new 52-week lows and only seven reached new highs. The analyst who drew this comparison identified only two instances like it: 21st December 1999 and 23rd July 1929.

Two points should be noted. The first is that the source is attributed to a particular individual who published it himself, and the dates given are specific rather than merely referring to a general month. Nevertheless, it depends on breadth data extending back to 1928, which can only be verified by firms that have a license.

We include it alongside his name and the qualification, even though that does not present it as a confirmed fact.

What we actually think this means

The fact that neither of the statistics is considered makes the basic situation obvious and fully interesting without the need for a reference to 1929.

The semiconductor index is 13.3 percent lower than its level in June. The bank fund fell by 2.33 percent on Tuesday. The difference between the two-year and ten-year Treasury rates has shrunk from 72 basis points at the beginning of February to 20 basis points on Monday, which is the smallest it has been this year. A flat yield curve tends to have a detrimental effect on banks since it lowers the amount they earn on long-term loans in comparison to the amount they pay on short-term deposits.

At the same time, Korean chip exports in the first twenty days of September were up 259.4 percent compared with the same period the previous year, and total exports rose by 78.3 percent; this is the strongest argument against the idea that we are at the peak of the market.

A narrow market, with a flattening yield curve and strong chip exports, is indeed complex. Comparing it to 1929 doesn’t make it any simpler.

The principle, since it is the reusable part

If a comparison is unexpected, the right question to ask should be which definition makes it true and whether anyone actually expressed it in that way.

Almost all statistics based on the ‘last seen in’ figure involve a flexible parameter—for example, the lookback window, the use of closing prices or intraday prices, or the selection of which index members to include. If the parameter is altered, then the comparison either changes or vanishes.

We are not saying that those who published these statistics were careless. Rather, what we mean is that by the time a statistic gets to you it has generally been repeated several times and the version that remains with people is usually the one with the most impressive phrasing, not the one that gives the most accurate details.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

This week certain reports state that Saudi oil shipments had decreased by more than 70 percent, falling from seven and a half million barrels per day at the beginning of the year to just two point one million by mid-September.

All morning we had tried to verify those figures but could find no evidence to back them.

Instead, we found a genuine and considerable decrease of 43 percent, a data error that price charts can correct, and a workaround that shows more than the headlines suggest.

Start by looking at the official report from Saudi Arabia.

Saudi Arabia sends its oil figures to an international statistics group each month, who then make them available free of charge. The figures in question are not estimates or ship tracking figures but are the country’s own data: 6.99 in January, 7.28 in February, 4.97 in March, 3.99 in April, 3.44 in May, 3.99 in June, and 4.13 in July, which is the most recent month for which figures have been reported.

Exports are down 43 percent from the February peak.

Exports never reach 2.1 million and they never begin at 7.5 million; the production figures show the same pattern, falling from 10.88 million barrels per day in February to 8.14 in July, and the energy agency gives the figure for August as 5.97 million.

A 43 percent decrease is already considerable. There is no point in exaggerating it to 70 percent, since that would draw attention away from the real issue and toward arguments about the sources.

The date, and why a price series settles it

While most reports state that the pipeline from east to west was attacked on 11 September, the statement by Saudi Arabia’s energy ministry is more precise in that it says the pipeline “was the object of a number of attacks in the morning of Thursday, 10 September 2026“ and was closed as a precautionary measure, resulting in injuries. Two days afterwards, the foreign ministry claimed the drones had been “launched from Iraq”.

The shutdown was announced on the eleventh, which is why there’s confusion about the date.

What stands out is that you don’t have to rely solely on the ministry’s statement; an independent source maintains timestamped records and has no stake in the matter.

The daily physical Brent price is published free of charge by the American energy agency. On 9 September it was $109.51, rose to $120.98 on 10 September, which was an increase of $11.47 in a single day, and then fell back to $118.06 on the eleventh.

On the tenth, the market responded to the attack, and buyers learned about it before the news was made public.

If there is a discrepancy between the official records and the news reports regarding a date, price data can be used as a reliable source to check on it.

The workaround, which is the actual news

We made it clear on 15 September that when the pipeline is blocked the alternative route is the strait which the pipeline was intended to avoid; the way to deal with a restricted strait is to use that strait itself.

This is currently happening along a definite route and at a certain volume.

Aramco has sold about sixty million barrels of oil, loaded at Ras Tanura on the Gulf coast, and transferred at sea to larger tankers at the Omani port of Sohar this month and next. The buyers are refineries in China, South Korea, India, and Japan. Gulf exports have now recovered to between one and one and a half million barrels per day, similar to the figure for August.

It is more than just a headline; Saudi Arabia has built a pipeline over 1,200 kilometers long across its own territory so its oil does not have to pass through the Gulf. Nowadays, since the pipeline is no longer in use, the oil goes through the Gulf, is loaded onto one ship, sails to Oman, and is then transferred to another ship.

All of these steps increase cost and risk; traffic data indicates fewer ships are now using the corridor, with only seventeen commodity vessels passing through the strait over the weekend, compared with thirty-seven the week before.

The terminal at the far end of the pipeline has also come under attack. On 19 September the Houthi movement stated that strikes had been carried out on facilities in Riyadh and at Yanbu belonging to Aramco. The Saudi-led coalition asserts that these attacks were averted and that a missile directed at Riyadh was intercepted, with no casualties recorded. We have included both the claim and the denial since that is what the record indicates.

What we are watching, and one of them is today.

The Brent spot price series, which we referred to earlier, ends on 15 September at $130.80, with the next update scheduled for today, 23 September.

This matters because, as we pointed out on 18 September, the actual Brent price and the frequently quoted futures contract were twenty-two dollars apart that day, and anyone can check today’s update free of charge to see whether that difference is narrowing.

If the physical price is the same as the futures price, then the gap was merely the result of one outage over two weeks, and we will state that. However, if the physical price remains considerably higher, this shows the market still lacks barrels, regardless of what the futures indicate.

Another thing to watch is the August exports figure, since it will be included in the next report. If it falls below three million barrels per day, then the situation is more serious than we have described. If it goes above five million barrels, the pipeline is performing better than the traffic data shows.

The honest other side

Exports were already falling before the pipeline was attacked, with data showing a decline from seven million and a quarter to under four million between February and May, long before these events. Since the attack hit a system that had already lost most of its volume for other reasons, it is just as incorrect to attribute the entire drop to a single strike as it is to claim the figure was seventy percent.

Starting in mid-September, the trend has turned, and exports have begun to increase. Gulf loadings have recovered, and satellite data indicates that tankers are now gathering at Gulf terminals at the highest level seen since at least June. Saudi Arabia is clearly taking action to deal with the problem.

What we want to make clear is that the figure is 43 percent, not 70 percent; the right date is the tenth, not the eleventh; and the new route is more expensive and more vulnerable than the original one.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

Even in the absence of a recession, AI could undermine the case for investing in consumer companies, since if it becomes easier and cheaper for customers to compare options and switch providers, those companies may lose the profits they earn when customers are not paying close attention.

The necessary technology is now beginning to emerge. OpenAI announced Instant Checkout in September 2025, a feature that enables users to make purchases within ChatGPT with confirmation and the merchant carrying out the transaction. In December 2025, Stripe launched the Agentic Commerce Suite, a service that connects merchants with multiple AI agents and provides payment support. OpenAI launch announcement, Stripe announcement.

The announcements provide the fundamental tools for carrying out transactions. Yet they do not indicate that AI agents have already caused a significant rise in subscription cancellations, nor do they show that many people are using AI to switch their insurance or telecom providers. The true opportunity for investors is to decide what to monitor before these changes appear in companies’ earnings.

Which switching costs protect the margin

A number of customers remain with a company because the product is better, the service is reliable, or switching would be risky. With others, however, the reason for staying is that spending the evening comparing options, reading the fine print, and completing forms would be better spent doing something else.

The two reasons might currently result in the same customer retention rate. However, if the software makes it easier to leave, these reasons won’t be as valid.

It would be helpful to picture a situation where a customer is given a skilled assistant who can compare similar offers and execute the switch if the offer is approved. Would the customer then choose their present provider at the same price?

The company could still have strong pricing power even if the reason for a yes is factors such as coverage, reliability, quality, or trust. However, if the only reason is that the customer never takes the trouble to open the renewal email, then part of that loyalty is simply due to the inconvenience of switching.

It is not necessary for AI to make the switching process completely effortless to affect the business; it only needs to persuade a sufficient number of customers that the effort involved in switching is now worthwhile. Conversely, if cancellation fees, the actual cost of integration, or trust are still theprimaryn reasonswhy peopleremainy, then easier comparisons may not make a great difference.e.

The earnings leak behind a stable customer count

Let’s consider an example: suppose there is a subscription company which has one million subscribers. Each customer pays $50 per month and generates $15 for the company before the company spends money on acquiring new customers. The cost of signing up a replacement customer is $300.

Let us assume that each customer who leaves is replaced within the same month, in which case the number of subscribers and the revenue remain unchanged. The only change in this scenario is that the monthly churn rate rises from 1% to 1.5%.

The business ends up losing 12.5 percent of its monthly contribution after spending money to replace its lost customers, even though both the number of customers and the revenue remain the same.

This is only an example and should not be regarded as a forecast for any particular company. It supposes that the cost of acquiring customers and the profit per subscriber remain unchanged, and it fails to take into account fixed costs, taxes, or accounting differences. The idea is that net additions can mask deteriorating business conditions, since a greater number of new customers might be required merely to maintain the status quo.

A company could prevent an increase in churn by offering larger discounts to retain its customers. Although the number of customers and the retention rate might then appear satisfactory, the actual prices it receives would decrease. The problem would then become evident in a different section of the financial statements.

The defense can consume the benefit

It is possible that company executives will argue that AI reduces customer service expenses or makes marketing more efficient. While such savings are significant, they do not, by themselves, provide a complete answer to the question.

In that example, the cost of replacing lost customers would be an additional $1.5 million per month, and the company would have to save that amount elsewhere to maintain the same level of profit. Even if the cost of acquiring a new customer fell from $300 to $200, the total cost of the higher number of replacements would still be $3 million, so the extra churn would be offset in this simple model.

A clear test is therefore available to investors. While AI agents can make it easier for customers to leave, the same technology can also reduce costs for companies when acquiring and serving new customers. The statement that “AI is good for margins” does not tell the whole story unless both sides are considered.

There is also the problem of who receives the value. If customers begin to use an agent to compare prices and make purchases, the company may lose some of its previous advantage and end up paying fees to a new intermediary. Consumers will not receive all the savings, as referral fees, commissions, and paid access can take a portion.

The company that controls the customer’s next purchase may end up being more important than the one that served them previously.

What to look for in reported results

Start by examining the figures on gross additions, gross departures, and acquisition spending since these are reported by companies. When net additions remain constant, but replacements increase, this situation should be explained. Moreover, care should be taken when defining churn since subscriber churn, account churn, and revenue churn have different meanings.

Now look at the real prices that customers are paying. Even if large renewal discounts are offered, fewer premium customers or more promotional months can keep the number of customers stable but reduce profits. Changes in average revenue can result from the product mix, so no single figure provides the full picture.

How long customers stay is another helpful sign. If new and long-time customers pay very different amounts, it could mean some revenue depends on people not checking their options. This is just a place to start looking, since older plans or different service levels can also explain the gap.

In the end, it is necessary to verify whether competitors can actually profit from customers who leave. A competitor offering a low initial price at a loss poses less of a threat than one that has genuine and sustained cost advantages. It is just as important to consider how long the challenger can survive as it is to assess how easy it is to spot them.

When retention earns its valuation

The case for switching is less strong when customers clearly favor their current provider after comparing available options, when switching involves risk, or when the amount of money saved by using AI is sufficient to offset the increased competition. Limits on how quickly people will switch also include trust, access to their transaction data, and clear permission to take action.

Before assigning a high valuation to a company, investors should investigate why customers remain. Although a long relationship may indicate excellent service, it might merely be because it is difficult to leave.

Loyalty that lasts even after customers compare their options is more valuable. As people get better tools, this difference might show up in company earnings before it shows up in overall spending.

On Friday, the Japanese government’s ten-year bond yielded 2.987 percent.

We checked the daily data from the Ministry of Finance, which goes back to July 1986, to see when yields were last this high. The answer is 13 September 1996, when the yield reached 2.993.

That was thirty years ago this week.

Earlier this month, on 2 September, the yield briefly reached 3.006, the first time it has hit three percent since 6 September 1996. The Bank of Japan will meet on Thursday and Friday, with an announcement expected on Friday, 18 September.

Why a Japanese bond yield is an American problem

Japan is the world’s largest creditor nation. For most of the past thirty years, Japanese pension funds, insurers, and banks had few good options at home. Government bonds paid almost nothing, and sometimes even less than nothing.

So, Japanese investors sent their money overseas. They bought American Treasuries, French and Australian government bonds, and American corporate and mortgage debt. These assets were not especially attractive on their own, but they looked better than earning nothing at home.

This flow of money has quietly supported the global bond market. For thirty years, whenever a developed country issued long-term debt, some of the buyers were Japanese investors searching for better returns than they could get at home.

This strategy only worked if the yield on foreign bonds was high enough above Japanese yields to cover currency hedging costs and still offer a profit.

Here’s how that gap has changed this year.

The gap is closing from the wrong end.

Since December, Japan’s ten-year yield has climbed from 2.066 to 2.987 percent, an increase of ninety-two basis points.

Over the same period, the American ten-year yield rose from 4.18 to 4.96 percent, up 78 basis points.

Japanese long-term rates have increased faster than American rates this year. The gap between them has narrowed from 211 to 197 basis points.

We think this part of the story is not being fully recognized. Most commentary treats Japan’s rate rise as just a Japanese or yen issue. It’s more than that—it’s about who is buying long-term government debt around the world.

Now, Japanese institutions can earn nearly three percent in yen from their own government bonds, without taking on currency risk. The incentive to invest abroad is smaller, even as the rest of the world needs more buyers for long-term debt.

It is already visible in the holdings.

This isn’t just a prediction. The shift has already started, and the U.S. Treasury’s monthly data shows it.

Japanese holdings of US Treasury securities stood at $1,239.3 billion in February. By June, the most recent month published, they were $1,116.7 billion.

That’s a drop of $122.6 billion in four months, almost ten percent of the total holdings.

Japan remains the largest foreign holder of U.S. government debt. But the trend has reversed, and it’s the direction of change that affects prices at the margin.

And the American long end is where you would expect it to be

On Friday, the twenty-year Treasury yielded 5.38 percent.

We looked at the daily yield curve data for each year. The last time the twenty-year yield was this high was on 14 June 2007. The peak of the widely discussed 2023 bond selloff was 5.30 percent. We have now surpassed that.

The ten-year yield is at 4.96 percent and the thirty-year at 5.35 percent, so the twenty-year now yields more than the thirty-year. Last week, from Tuesday to Friday, the twenty-year yield rose by twelve basis points and the ten-year by sixteen.

This updates what we wrote two weeks ago, and we want to be clear about it. On 1 September, we said that August’s bond selloff was not really about long-term bonds. The biggest moves were at the short end, where the Fed’s decisions have the most impact. That month, the thirty-year yield actually fell by two basis points and the twenty-year by four.

That was true at the time, but it’s no longer the case. In September, the biggest changes have happened at the long end of the yield curve.

The honest uncertainty

We don’t think the Bank of Japan’s decision on Friday is the main event. Whatever the bank decides, it will be reacting to changes that have already taken place in the market.

The real question is harder to spot: will Japanese institutions see three percent yields at home as a reason to bring money back, or will they view it as a temporary change and keep their foreign investments? Large institutions don’t move quickly, but Treasury data shows some have already started shifting.

There are good arguments against our view, and we want to acknowledge them. Hedging costs can change, and if it becomes cheaper to hedge dollars back into yen, investing abroad could become attractive again, even if the yield gap is smaller. An unhedged Japanese investor cares more about the yen’s value than the yield difference. Neither factor makes domestic yields unimportant, but both could slow down the shift.

What would change our view? If Japanese Treasury holdings rise for two months in a row, or if the U.S.-Japan ten-year yield gap widens past 230 basis points. We’re putting this in writing today.

Where we think this lands

If the world’s biggest exporter of savings stops being the main buyer of long-term government debt, those bonds will need new buyers, likely at different prices.

There’s nothing dramatic about this. It won’t be settled on any single day. It’s simply a shift in who is buying at the auctions.

The American twenty-year yield hitting a 2007 high, while the Japanese ten-year is at a 1996 high, shows what this transition looks like in real time.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

The Financial Times stated on September 18 that the loans associated with the $18 billion Project Jupiter—this is the project which finances a data center leased to Oracle—were being privately quoted at around 89 to 91 cents on the dollar due to difficulties in selling the debt. It should be noted that these are merely the quoted figures, not actual sales. (Financial Times reporting.)

The fact that a quote is near 90 doesn’t imply that there has been a default, that there is a 10 percent chance of default, or that there has actually been a 10 percent loss on the financing; the price may be affected by interest rates, liquidity, credit concerns, and the discounts had to be offered in order to attract buyers.

People ought to inquire into the length of time the financing remains with the institutions that agreed to it, and how this influences their readiness to take on new deals.

Capital has a holding period

A bank or arranger which decides to finance a project with a view to selling the loan to other investors does so by assuming risk while it holds the loan. In some instances the commitment is made before any money has in fact been lent, and in other cases the arranger keeps a portion of the loan as good. Regardless of the situation, if the period of exposure extends beyond what is expected, it may use up the bank’s lending limits and funding resources.

Let us look at a simple example intended to be as straightforward as possible. A lender commits $1 billion to holding the loans before selling them; there is enough eligible demand, capacity remains fully utilized, each sale frees up the full amount, and the risk per dollar remains the same.

When calculating throughput, multiply $1 billion by 365 days and then divide by the holding period; if the holding period is doubled, the number of loans processed each year is halved, yet all of the loans can still be paid on time.

It has a fixed and dedicated capacity. Real institutions have the possibility of raising capital, altering their limits, hedging their exposures or agreeing to a lower sale price. It does not constitute a regulatory capital calculation nor does it assert that Jupiter has halved any bank’s lending capacity.

This model is intended to illustrate a potential limitation that default forecasts may overlook. If the earlier commitments take longer to be distributed, the market could end up funding fewer new projects. Developers might encounter this limit in the form of smaller loan commitments, higher interest rates, greater equity requirements, or later closing dates.

Construction and financing run on different clocks

There is also a timeline concerning the physical delivery of a project. The timing of when it can begin serving customers is determined by factors such as obtaining permits, securing a fuel supply, acquiring equipment, and the commissioning process. A delay in construction does not necessarily mean the loan distribution will be delayed by the same amount of time, but it may make lenders more cautious or lead them to request different terms.

The condition set out in Oracle’s 8 September announcement regarding its expected commitments to the Jupiter community and the economic benefits was that the air permit and the pipeline be approved as originally planned. This condition has a specific scope and should not be interpreted as applying to all of Oracle’s obligations. Oracle announcement.

Even for equity investors, importance should be attached to timing even if demand remains strong, since a project may eventually function as intended but could still end up being considerably less valuable than expected if it is delayed.

Here’s another example: A project requires an immediate investment of $1 billion. The present value of the future net operating cash flows is $1.1 billion when a 10 percent annual discount rate is applied. Therefore, the net present value of the project is $100 million.

If you delay all the operating cash flows by seven months but leave the original outlay unchanged, the present value of the cash flows will be about $1.041 billion; that is, $1.1 billion divided by 1.10 raised to the power of 7/12. As a result, the project’s net present value then drops to approximately $41 million.

The operating cash flow’s value decreases by only 5.4 percent, but the excess value over the original investment declines by about 59.5 percent. Indeed, a small change in the total value is enough to eliminate most of the meager investment buffer.

It is not a valuation of Project Jupiter, nor is it a prediction of equity returns; it assumes that all operating cash flows are merely postponed and fails to account for any additional financing costs, penalties, lost contract years, or the possibility of delaying construction spending. The key issue is that value creation is highly sensitive to timing when the initial cushion is small.

The same delay can hurt one asset and help another

An investor who assumes all AI infrastructure will be affected in the same way might overlook a key issue. When demand remains strong but new supply is delayed, the current operational capacity may become more valuable. A data centre that is already running and has paying customers has to deal with different problems than land which still requires permits, power and financing.

There is therefore a need to compare assets on the basis of their remaining delivery risks. Simply looking at announced capacity is an inadequate basis for such a comparison, and so is a backlog figure without details on the spending and milestones required to convert it into cash.

Evidence shows that a financing problem does not mean the entire construction program has come to a halt. In the first quarter of fiscal 2027, Oracle announced that it had delivered an additional 850 megawatts of data center capacity; however, this figure represents the company's total and does not confirm that Project Jupiter is operational. Oracle quarterly results.

Investors have to be able to distinguish between assets that are already in operation, projects that can be carried out, and commitments that may or may not result in revenue.A company could have all three kinds.

What would change our view

Once financing has been completed, we will examine the actual terms. In the case where the loans are fully distributed, the banks will have less exposure and, as a result, new arrangements can close without terms having to be much worse, weakening the case for a capacity constraint. Discounted sales can also be of help—by accepting a loss or lower profit, it is possible to free up capacity for new commitments.

We should also consider who is financing the delay. Although customer prepayments, extra sponsor equity, and construction contingencies can absorb this pressure, the costs involved and the contractual obligations remain important. Shifting the funding responsibility does not mean it is automatically removed from the economic system.

The questions which are most useful to managers are concrete; for example, how much of the committed financing still needs to be distributed? What approvals lie between the current spending and the first customer payment? How much spending could be affected if delivery is moved? And what effect would a six-month delay have on the return?

A good answer would link these questions to cash commitments and project milestones. Until that happens, the fact that there are no defaults does not say much about how fast new construction is moving. The AI build can slow down even if all borrowers keep paying their bills.

We think that Micron might become more predictable, but it will not derive unlimited benefits from AI spending. Although customers may be willing to pay for a reliable supply, they usually manage to secure some of Micron’s future profits as well. This kind of trade-off must be carefully evaluated.

It is important to make this distinction since Micron is achieving exceptional results. In the third quarter its revenue was $41.46 billion and its GAAP gross margin was 84.6%. The company anticipates a margin of about 86% in the fourth quarter. Although these figures are impressive, they do not indicate what margins Micron can achieve over the course of a full investment cycle.

The shareholder trade inside a supply agreement

Micron has announced 16 strategic agreements with customers, these involving take-or-pay commitments and in the case of the largest ones both minimum and maximum prices have been set. Micron Q3 Form 10-Q, page 27.

The agreements include around 20 per cent of the DRAM volume and one third of the NAND volume during the period in question. Although management aims to eventually achieve coverage of half or more of revenue, this is a separate objective for the future. Negotiations regarding the premiums for new products are still ongoing. Micron have prepared comments, on pages 3 to 4.

When a supplier enters into this type of contract it is making a real economic trade-off; the price floor serves to guard against a fall in market prices and the ceiling ensures that customers receive some of the benefits if a shortage occurs. The two features should be considered together, not one at a time.

Picture a memory contract having a minimum price of 70 and a maximum price of 100; if the market price falls to 50, the floor ensures a protection of 20 in value, and if the price rises to 140, the ceiling means that the customer retains 40 of what could have been earned in revenue. These figures are merely example values and not the actual contract prices set by Micron. The example illustrates that the value of a contract is determined by the future prices, the amount committed, and the cost of delivering the product.

The agreement can still be very favourable to shareholders even if there are these compromises, since by preventing a large cash shortfall the company will be able to avoid the need for expensive borrowing, cutting back on investments, or having to dilute its shares. It is valuable to have predictable income when it comes to factories that require large investments before demand has been established.

The valuation should take this trade-off into account. It would be wrong for investors to grant the company extra value in order to provide it with protection against losses and at the same time assume that it will receive full shortage pricing on those sales; that would amount to double-counting the benefits.

A price floor leaves production costs exposed.

The next point is the amount of profit that is left when prices are at the floor.

In the case described, when the floor price is 70 and the unit cost is 20, the gross profit is 50; if the unit cost rises to 35, the gross profit falls to 35, representing a 30% decrease, even though the customer continues to pay the protected price. Committed volume is only of any benefit if each additional unit still provides a satisfactory return after the cost of supplying it has been taken into account.

That is the reason why the contract should be of the same length as the factory and product plans. What products are covered? How does the agreement deal with changes in the product specifications? Who is responsible if there is a difficult transition? A supplier may have considerable rights regarding customer orders but can still encounter major manufacturing risks.

It by no means implies that all new products will have a negative effect on margins; rather, it points out the difference between a sales contract and a forecast of cash flow for shareholders. Higher yields, an improved product mix, and agreed-upon premiums could more than make up for the increased costs. It is important that analysts make these assumptions clear.

The cash balance needs an ownership test.

Micron has projected customer deposits and related commitments totaling $22 billion, of which about $18 billion consists of cash deposits. Management stated that deposit flows constitute financing cash flows, which are excluded from free cash flow and are returned later under the agreements. These are projected inflows and should not be deducted wholesale from the previous quarter’s cash balance. These comments appear on page 7 of Micron’s remarks.

The fact is that this financing is valuable since it reduces the requirement for external capital in order to grow; the true advantage lies in the terms of the funding and the returns obtained during the time the money is held.

That is not equivalent to customers providing shareholders with a permanent cash payment. If you include the entire deposit in the equity value without taking into account the obligation to return it, you inflate the benefit. Conversely, viewing these deposits as ordinary and costly debt overlooks the potential value of customer financing.

You need to consider both receipts and repayments at the same time, match them up with the balance sheet, and make sure that the same benefit is not included in both the net cash flows and the future cash flows. Greater liquidity does not necessarily mean greater wealth for shareholders.

The margin evidence extends beyond HBM.

A further interesting point in the results is that the gross margin of Micron’s Mobile and Client business was 87 per cent, as compared with 83 per cent for Cloud Memory. Since these businesses are not direct product comparisons, this does not demonstrate that regular memory is more profitable than HBM. It does, however, indicate that high profits are not confined to the AI sector.

We believe investors should consider the profit from unique products and the profit from scarcity in the wider memory market; even if profits from unique products remain strong, those from scarcity may decline.

A likely result—which is frequently overlooked in discussions that present only two options—is that demand for HBM remains strong, Micron’s strategic position improves, but a portion of the extra profits in the other areas disappears. Although the company succeeds on the technology front, its earnings may not reach the highest levels seen previously.

The other side of the argument is also possible. Contracts and improved products could keep earnings higher than in previous cycles for longer than history suggests. Any strong bearish argument should account for this possibility.

Value the protected and exposed cash flows separately.

We would carry out the assessment in three stages. In the first place, we would estimate the cash flow from the committed volumes at various contract prices, taking into account the cost of delivery. Secondly, we would examine how the business performs when prices and usage move outside the scope of the protections. Thirdly, we would model future products and new contracts by making clear assumptions regarding bargaining power.

The only thing this method does is provide a forecast, not assert that the factory can in fact be divided into three separate assets. The shared capacity, costs, and investments still have to be taken into account at the company level.

In each case the important question is this: how much cash will the shareholders receive after the company has paid for the technology, increased its capacity, and met its financing requirements?

The bullish case would be stronger if disclosures showed clear cash returns at the contract floors, if premiums from new products covered the additional costs, and if growth was achieved without reducing expected returns to investors. Conversely, the case would weaken if investment commitments increased while profits from protected sales declined, or if the unprotected portion of the business contracted sharply.

We believe our conclusion goes beyond merely forecasting a repeat of the old cycle. Although Micron may merit recognition for having altered its risk profile, the return investors receive should still be supported by the cash flows and upside left after customers obtain protection.

MoatPeak Independent Research | www.moatpeak.com

Our thesis in one sentence. The Fed delivered the expected hike, but the more important investment signal was a higher real cost of capital and a market recovery that did not lift all equity groups equally.

Four findings carry it.

First, the policy debate moved beyond September. The Fed raised its target range by 25 basis points to 3.75%–4.00%, unanimously. The September projections put the median year-end policy rate at 4.1% in both 2026 and 2027. That is a materially different backdrop from assuming this hike will be quickly reversed.

Second, the bond arithmetic strengthened the prior edition’s real-yield thesis. At five years, the nominal yield rose 8 basis points while the real yield rose 17, leaving inflation compensation 9 basis points lower. The ten-year nominal yield finished at 5.01%. Higher real yields can squeeze valuations even when inflation compensation is contained.

Third, the consumer was still spending. August retail sales rose 1.2%, including a 1.2% increase excluding autos and gasoline. Initial claims fell to 196,000. But manufacturing output declined 0.3%. The economy was resilient without being uniformly strong. These releases describe activity before the September hike could materially affect behaviour.

Fourth, the recovery was concentrated in time. The S&P 500 lost 105.17 points over Monday to Wednesday, then recovered 98.69 on Thursday and Friday. That repaired almost 94% of the decline, but it did not establish a broad improvement in financing-sensitive businesses.

The strongest argument against our cautious reading is that earnings growth and productivity can offset higher discount rates. A rising real yield can partly reflect confidence in future growth, rather than only tighter policy. If that confidence spreads into smaller companies and new orders, the market may be absorbing the hike more successfully than the weekly index split suggests.

We would become more constructive if real yields eased, oil supply improved and small caps began outperforming alongside steady employment. We would become more defensive if real yields and inflation compensation rose together while activity weakened. Neither outcome is established by one volatile week.

The Hike Was Only the Beginning

The decision removed uncertainty about the first step. The projections raised a different question: how long must policy stay restrictive before inflation returns to target? That horizon matters to asset values because a temporary increase in borrowing costs is easier to absorb than a persistent one.

A higher path does not guarantee a long hiking cycle

At the new target midpoint of 3.875%, a year-end median of 4.1% is consistent with roughly one further quarter-point increase. The distribution is informative: 12 of 18 participants projected a 4.125% midpoint for 2026, four projected 4.375% and two projected 3.875%. The median is therefore a useful centre of gravity, but it hides disagreement over the next step.

Our interpretation is that markets should test the persistence of policy, rather than trade only the next meeting. The combination of resilient demand and unresolved energy costs makes an immediate reversal harder to justify. Equally, a sustained improvement in supply or a deterioration in employment could change the path. Forecasts are conditional on an economy that will continue to evolve.

What this means for valuation

A business whose cash flows arrive far in the future is sensitive to the discount rate applied over many years. A higher policy path can also raise refinancing costs before revenues respond. The useful distinction is between companies that can fund investment internally and those whose plans depend on repeatedly accessing cheap capital. The same macro rate can have very different consequences for each.

Real Yields Did More Than Nominal Yields

The rise in the headline Treasury yield understates the change in the real return investors demanded. Inflation compensation moved in the opposite direction, partly cushioning the nominal increase. This is consistent with a stronger policy response restraining inflation risk, although the yield decomposition alone cannot prove that explanation.

Why the curve matters more than the round number

The two-to-ten-year slope narrowed from 33 to 25 basis points; the two-to-thirty-year slope narrowed from 72 to 58. These are calculated from the official Treasury par curves. The front end repriced more than the long end. That pattern fits concern about tighter near-term policy better than a simple claim that every maturity was selling off equally.

Five percent is a useful reference point, but it is not a mechanical trigger for an equity crash. Stocks respond to the combination of earnings, discount rates and risk appetite. A gradual yield rise alongside stronger profits can be manageable; a fast rise accompanied by falling estimates is a different proposition. Credit spreads and refinancing terms help distinguish the two.

The limits of the inflation signal

Nominal minus real par yields is an approximation to inflation compensation, not a pure forecast of future inflation. Risk premia, liquidity and differences between the bond curves also matter. Falling compensation is therefore evidence against an uncontrolled inflation repricing this week, but it is not proof that households or firms have stopped worrying about prices.

What the Market May Be Missing

Oil can stop rising and still remain a problem

WSJ’s settlement summary put WTI at $100.30 and Brent at $103.87, with little net weekly movement. The retreat from midweek pressure followed reports that partial pipeline flows could resume. These are contract-specific settlements; they should not be substituted with earlier intraday quotes.

The investment inference is about duration. A flat weekly oil return does not mean the cost shock has disappeared. Businesses may still face higher freight, fuel and working-capital costs than their budgets assumed. What matters for margins is the cumulative cost burden and the ability to pass it on, not just the latest five-day price change.

The Bank of England supplied a useful counterpoint

The BoE held Bank Rate at 3.75% by a 6–3 vote, with the minority preferring a hike. It reported little evidence so far of material second-round wage and price effects, while judging inflation risks to have increased. The contrast with the Fed illustrates how the same energy shock can meet different domestic conditions. [11]

For a UK-based investor, the implication is to separate global shocks from local policy responses. Domestic activity, wage growth and currency exposure can matter as much as the global commodity headline. Sterling returns on overseas assets also depend on exchange-rate movements, so the dollar-based equity scoreboard is not a UK investor’s realised portfolio return.

A credible central bank does not guarantee rising asset prices

Policy can reduce inflation risk while increasing the real return required on investments. Both outcomes can occur together. It is therefore possible for monetary credibility to improve while equity multiples contract. Investors should ask whether prospective cash-flow growth adequately compensates for that higher hurdle.

What We Are Watching

The real cost of capital

Five-year real yield: 2.55%. A sustained move above 2.65% would reinforce valuation and refinancing pressure. A move below 2.40% with stable activity would offer more convincing relief. The point is the real financing environment, not a single intraday print. [5]

The inflation component

Five-year compensation: 2.31%. A rise above roughly 2.45% while real yields also climb would be more concerning than either move alone. It would suggest investors require both more real return and more compensation for inflation risk. The measure includes risk and liquidity effects, so it needs corroboration. [4–5]

The long bond benchmark

Ten-year par yield: 5.01%. A persistent move above 5.10% alongside weaker earnings expectations would increase pressure on equity valuations. A fall below 4.85% would need interpretation: supply relief is helpful, while recession-driven demand for bonds may be much less so. [4]

The breadth of the equity response

The Russell 2000 versus the S&P 500. A durable period of small-cap outperformance, supported by improving earnings and stable credit, would challenge the view that tighter financing remains the dominant constraint. Relative gains caused only by large-cap losses would be a weaker signal.

Energy supply rather than headlines alone

Oil below $95 would be a meaningful change only if it reflected sustainable supply relief rather than collapsing demand. Conversely, prolonged triple-digit prices can sustain cost pressure even without another weekly surge. Track physical flows, refined products and company commentary together.

Employment and investment

Claims below 200,000 and stronger orders would support resilience; a sustained four-week claims average above 225,000 accompanied by weaker investment would make the growth cushion less convincing. These are our monitoring bands, not official recession criteria. Initial claims must also be read alongside hiring and continuing claims.

The research question for the next edition

Can better cash-flow expectations broaden the market while the real cost of capital remains high? That is the link between the macro data and investment outcomes. The Fed has acted; the next evidence must come from households, businesses and the prices investors are prepared to pay for their future earnings.

On Wednesday, the Federal Open Market Committee raised its target range by a quarter point to 3.75 to 4 percent. The vote was unanimous at 12 to 0. This is the first increase since July 27, 2023.

The unanimous vote is notable, especially since in July the committee was split 9 to 3, with three members wanting a rate hike that didn’t happen. In just eight weeks, the group went from divided to fully united.

But the rate hike was already expected by about ninety-two percent of the market, so it wasn’t really news. The projections, however, were.

The key line that stands out

We looked at the Summary of Economic Projections and focused on Table 1.

The committee’s median forecast for the unemployment rate is 4.1 percent in 2026, 2027, 2028, and 2029.

Four consecutive years, the same number, to one decimal place.

Back in June, the same table showed 4.3, 4.3, 4.2, and 4.2. So, they didn’t just flatten the forecast – they lowered each year’s number and then made the path flat.

When a forecast shows the same number for four years in a row, it’s not really a forecast – it’s more of an assumption. The rest of the document relies on this idea. If unemployment stays flat for four years, there’s no job loss from tightening policy, so the only thing left to focus on is inflation.

And they removed the hedges.

The projections also include a section most people don’t look at. Each member is asked if the risks to their forecast are tilted higher, lower, or are balanced.

In June, seven of eighteen thought unemployment could end up higher than they forecast, which would mean a weak expected drop to zero.

Now, none of the eighteen officials think the labor market could turn out worse than expected. In the same three months, seventeen out of eighteen expect inflation to be higher than forecast, and none expect it to be lower.

That’s about as one-sided as this document gets. The committee went from seven members hedging on the labor market to none, all while raising rates. They haven’t just changed their forecast – they’ve stopped protecting against surprises.

Why we’re not so sure

Two new numbers came out this week that show a single rate can’t really sum up the labor market.

Initial jobless claims dropped to 196,000, and continuing claims fell to 1,730,000 – the lowest since early January 2024. By itself, this suggests a very tight job market where almost no one is being laid off.

But housing starts fell 2.6 percent to an annual rate of 1,275,000, and building permits dropped 2.7 percent to 1,394,000. Construction doesn’t look like a sector that’s about to start hiring.

The way to explain both numbers is that almost no one is being fired, but almost no one is being hired either. In this situation, the unemployment rate stays steady, but not for a good reason. It remains flat because nothing is changing, and it will stay that way until layoffs begin. When that happens, the rate could jump quickly since there’s no hiring to offset it.

A flat unemployment rate can mean either a strong job market or one that’s stuck. The projections have picked one interpretation, removed any disagreement, and set policy based on it for the next four years.

Here’s the rest of the table for reference

Growth for 2026 was revised up to 2.3 percent from 2.2. Headline inflation is now 3.7 instead of 3.6, and core inflation is 3.4 instead of 3.3.

The policy path also changed a lot. The median federal funds rate for 2026 went from 3.8 percent in June to 4.1 now. For 2027, it moved from 3.6 to 4.1, and for 2028, from 3.4 to 3.9.

The 2027 change is the biggest and has gotten the least attention. It’s a half-point increase, which means two rate cuts have been taken off the table for next year in just three months.

A quick correction to a widely shared figure: the dot plot shows twelve officials expect exactly one more rate hike this year, four expect two, and two expect none. So, sixteen expecting at least one more is correct. But the rest of the breakdown being quoted has every label off by one hike.

What could make us rethink

If the December projections add back any rate cuts for 2027, then Wednesday’s move was probably an overshoot, not a new trend.

If more officials start hedging on the labor market again in December, then the unanimous vote was just a one-time event, not a lasting position.

Both sets of projections are published, free to access, and clearly dated.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

There are two prices for Brent.

The front-month futures contract, which is traded on ICE, is the one that appears in the headlines, on trading terminals, and in market summaries. This week it fell from around $105 to $98.53.

The other is the physical spot price, the price at which actual shipments of crude oil are bought and sold. The US Energy Information Administration releases this price daily at no cost and has done so for many decades.

On the latest day for which it published figures, 15 September, it reported $130.80.

The gap, session by session

The two prices are generally only a few dollars different. That is essentially what a futures contract is for; when they diverge, it indicates that the paper market and the physical market no longer agree about the value of a barrel of oil.

The situation regarding this month's gap was as follows.

On the 4th of September, the physical price was $6 higher than the futures price; by the 9th, the difference had become $8. After the attack on the Saudi east-to-west pipeline on the 10th, it had risen to $13 and remained at that level on the 11th before reaching $16 on the 14th.

On the fifteenth, the amount was $22.05.

The physical price rose from $104.47 on September 7 to $130.80 on September 15, an increase of 25.2 percent over six sessions.

On the other hand, the futures contract did not change much over the same week and then fell.

What this explains

This week, there were two things that could not be explained on the basis of future prices alone.

On Wednesday morning the diesel margin—the margin between the price of crude oil and that of finished diesel—reached approximately $117 per barrel. Diesel futures reached record levels in both the United States and in Europe, and retail diesel also set a new record at $6.3956 per gallon on Thursday. This occurred in a week when crude oil prices were falling. At first glance, this seems inconsistent when looking only at futures prices, but it becomes sensible when the physical crude price of $130.80 is taken into account.

Second, the prices of shares did not respond to the oil, which was apparently cheaper. Several of the reports referred to Wednesday’s session as the market taking a downward turn on a gift, indicating that policy had become more important than energy as the main influencing factor.

There wasn’t any gift; during the week the cost of oil for the real economy went up and it was only a financial product that decreased.

We must avoid going too far with this point. On Wednesday, the Federal Reserve’s forecasts still had an impact on the market since they clearly did so. However, the explanation relating to energy was rejected on the basis of a price which in fact did not reflect the actual situation.

Why this happens, in plain terms

A futures contract is settled at a future date and can be sold by someone who does not own oil and bought by someone who has no intention of taking delivery; the contract's price is influenced by expectations, trading positions, and events such as pipeline repairs.

A physical cargo has to be produced, shipped and delivered to a refinery which needs it quickly. The price will depend on whether the barrels are actually available.

Usually, when the two prices diverge it means that the market expects the shortage to finish even though the shortage is still going on. That is exactly what took place this week: the time needed to carry out repairs on the Saudi pipeline was reduced, production in Libya picked up again, and the futures market had already priced in a recovery. Yet the market still does not have enough barrels at present.

There is also further evidence concerning a price gap. On September 15 the gap between physical Brent and physical West Texas amounted to $23.78. When we examined this spread over 8,395 trading days, it was the 55th-widest gap on record, with the average being $4.18 from 2015 to 2019. Physical crude prices in the Atlantic basin are behaving very differently from what the screens indicate.

What it means for a portfolio

Whenever oil prices are used for any purpose—and since most people do use them—it is important that you know which oil price you are referring to.

The arguments concerning discount rates or inflation which are based on falling crude prices are currently grounded in the futures curve. Yet the cost of fuel for freight and refining margins is at record levels, and the oil that refineries purchase has never been more expensive than the futures contract meant to represent it.

We are not attempting to predict what direction either price will take. Instead, we are highlighting that the two prices have diverged so much that one can no longer serve as a stand-in for the other. Nevertheless, most of the commentary this week did exactly that.

What would tell us we are wrong

If the spot price for physical goods falls within $10 of the price of the front-month contract within a month, then the problem had been merely a temporary disturbance caused by a single pipeline failure and was not as serious as we have said.

When the diesel margin decreases rapidly even though the actual prices of crude oil stay high, the refining and crude oil situations are in fact independent and our link between them is overly simple.

We have recorded both points with today's date, and both of them can be verified on the EIA’s free website.

One honest limitation

The EIA releases its spot price series with a delay, the most recent figure being from September 15; the futures prices, on the other hand, are updated in real time. As a result, the largest discrepancy occurs when a physical price from Tuesday is compared with a futures price from Friday, and part of this difference will decrease as more data becomes available. Where possible, we have matched the dates, and the $22.05 figure is from the same day. Although the direction of the gap is obvious, its exact amount can vary from day to day.

The International Energy Agency’s September report also points out this discrepancy, stating that the futures were at $105 whereas the physical benchmarks were considerably higher. This indicates that the problem is not limited to a single data source.

Educational research only. Not personalized investment advice. MB “MoatPeak Group”.

On Tuesday the U.S. Treasury issued thirteen billion dollars’ worth of twenty-year bonds. This is a regular occurrence which takes place every month, is made public in advance and generally doesn’t become a news item.

This auction established three records at the same time.

We obtained the results for all of the 20-year auctions that the Treasury has conducted since it resumed offering this maturity in May 2020; there were seventy-seven such public sales. Rather than depend on other people, we worked out the bidder shares ourselves.

What the results actually say

The auction concluded with a yield of 5.420 percent, which was the highest of all the seventy-seven auctions; the next highest yield was 5.245 percent in October 2023.

Bids can be divided into three categories: among the indirect bidders are foreign central banks and overseas institutions; the direct bidders are mainly domestic institutions that are purchasing for their own account; the rest are primary dealers since they are obliged to.

In the auction in question, indirect bidders obtained 52.5 percent, which was the lowest figure among the seventy-seven auctions; in August they got 62.9 percent, and the average of the prior six sales had been 68.0 percent.

The percentage obtained by direct bidders was 30.7 percent, the highest of all the seventy-seven auctions; in August it was 24.6 percent.

The dealers received 16.9 percent, the bid-to-cover ratio being 2.57 as compared with the average of 2.65 from the earlier six auctions.

The three measures in this auction all set records in the same direction.

Why this bond auction matters

All through the year people have been discussing whether foreigners are pulling back from American debt, mainly relying on holding data which comes two or three months late and is frequently revised.

We too have made use of that data. Two days ago we stated that Japanese holdings of Treasuries had decreased from 1,239.3 billion dollars in February to 1,116.7 billion in June, and that China’s holdings had fallen from 731.4 billion in June 2025 to 633.4 billion one year later. Although these are actual figures, they illustrate the past and refer to stocks rather than flows.

With an auction it is different; the government has to secure buyers at a fixed price on a particular day and there is no possibility of making revisions or delays. Anyone who takes part is included.

On Tuesday, the foreign share reached its lowest level ever, the domestic share was at its highest, and the price had to be adjusted in order to complete the sale.

A fair objection, and why we disagree

That morning, when the Treasury Secretary was questioned in Congress about the rising yields, he mentioned that the ten-year and thirty-year bond sales the previous week had been a success and gave them a positive description. That is correct and does require an appropriate response.

The answer is this: the sales took place on mornings when yields had already gone up sharply before the auction; the buyers at one o’clock got a discount which was announced earlier that day; that is to say, it was a concession, and concessions draw in buyers.

The auction on Tuesday made no concessions. Yields remained close to five per cent for the majority of the session, and it had to be raised by two basis points over the price at which the market had previously priced it for there to be enough bids. Since a sale that clears without any concessions indicates real demand, this outcome was weak.

We would also like to make it clear what we are not saying. A single auction doesn’t constitute a trend; if the October sale causes the foreign share to return to sixty per cent, that should be regarded as nothing more than a bad day, not as a change in the market, and we will state this plainly.

What a regime change would mean

All bonds issued are purchased; the only thing that varies is whether someone buys them and at what price.

For twenty years foreign official institutions purchased a major portion of long-term American debt. They are not sensitive to price in the ordinary way. A central bank which is in charge of reserves is not attempting to beat a benchmark; it acquires securities because it holds dollars and has to invest them.

Private institutions are different; when a pension fund or an insurer purchases a twenty-year bond it is making a value judgment and will only do so if the yield is appropriate. Substituting price-insensitive buyers with price-sensitive ones doesn’t eliminate demand but does make borrowing more costly.

That is why recent headlines have appeared, and we think this situation will continue longer than any individual story. The fact that a government has a deficit of nearly two trillion dollars, as stated in its own eleven-month report, means that it has to sell a large amount of debt to a market that is changing.

What could change our view

If the indirect share at the sale held in October over a twenty-year period exceeds 62 percent, Tuesday’s result was nothing but noise and the idea that the buyers have changed is wrong.

When the figure remains under 58 percent for two consecutive sales, this means that the buyer mix has changed and the yield required to clear these auctions will be higher than most people expect.

For each date there is one figure and anybody can check these figures for free on the Treasury’s website.

The sources

All the figures in this piece have been calculated here on the basis of the Treasury’s own auction results, which are free and publicly available. The period covered is seventy-seven public twenty-year auctions from 20 May 2020 to 15 September 2026. The shares refer to accepted competitive bids, this being the Treasury’s own practice. The one auction involving an additional twenty-five million dollars in December 2021 has been omitted since it was not a public sale and would otherwise appear falsely at the bottom of the table.

⁠Educational research only. Not personalized investment advice. MB “MoatPeak Group”.