Newsletter — week of October 5, 2026

Market performance

The S&P 500 (the index of the 500 largest U.S. companies) ends the week up 1.06%, after a choppy stretch. The index set a record on Tuesday, at 7,844.52 points, before pulling back when strain in the bond market pushed the 10-year yield toward its highest level since 2002.

Beneath this apparent rise in the index, the rotation is pronounced. Real estate gains 1.85% and healthcare 1.48%, while technology sheds 0.52% and the semiconductor index falls 4.30%. In other words, the market is rising without its usual engines, which deserves attention and shows that, contrary to what one often hears, the U.S. economy doesn’t run on AI alone.

Market news

Hormuz: nine oil tankers attacked in one week

We’ve been following this file for several issues, and the situation has hardened further. Nine ships were attacked in and around the Strait of Hormuz over the past week. The volumes transiting the strait have fallen to 9.5 million barrels per day, about 30% below the pre-war level. (source: CNBC)

This week Brent stayed above $100, ending the week up 3.45%. A hurricane threatening Gulf of Mexico production added its share of strain to the market.

In our view, the G7 reserve release mentioned last week has not changed the picture. Worse, it reinforces it: those barrels will have to be bought back. Our read is therefore confirmed—it’s a bandage, not a solution. As long as the strait operates at two-thirds of its capacity, the price of the barrel remains hostage to military decisions rather than to the balance of supply and demand. We also want to remind you that, even once the strait reopens for good, there would still be the strategic reserves to refill, on top of the daily consumption to supply. A simple calculation gives the scale of this demand.

The U.S. strategic reserve is currently short 431 million barrels. A VLCC, for very large crude carrier, is the largest oil tanker in common service: a ship over 330 meters long that carries about 2 million barrels. Filling this gap would require 216 full supertankers—more than a quarter of the global fleet, which numbers about 800 ships of this size. Even just to return to the end-of-February level, before the March releases, 66 more would be needed. For comparison, the United States consumes 20.1 million barrels a day, the equivalent of ten of these ships daily.

In this logic of constrained supply, prices should therefore stay high over the medium term. In this context, our exposure to energy keeps all its relevance.

Long-term rates, once again

The U.S. 10-year yield touched 5.3645% intraday, its highest since 2002, before easing toward 5.23% at week’s end. This lull owes much to a solid auction on Wednesday—that is, an auction of new debt that found buyers without difficulty, reassuring on investors’ appetite for U.S. bonds.

On the central-bank side, the market now anticipates a hold in October, with about 78% probability, then a hike in December, around 69%. The minutes of the September meeting confirm that a majority of members still expect a hike by the end of the year.

In our view, nothing contradicts what we wrote in the monthly report. The pressure on long-term rates comes from an imbalance between a massive supply of debt and demand that struggles to keep up, not from a fear of recession. This week’s reprieve rests on a single successful auction, which is precisely the sign of a market that has become dependent on each issuance going well.

The misread figure

On Thursday, the semiconductor index lost 3.7% in a single session. Oracle shed more than 5%, AMD nearly 4%, Nvidia nearly 3%.

This decline followed a Financial Times article that revealed on Thursday that OpenAI’s annualized revenue stood at about $50 billion at the end of September. The market had been expecting $70 billion for weeks. It’s this $20 billion gap that triggered Thursday’s sharp fall.

Why does this $20 billion matter?

Because OpenAI is not just any company: it’s the supposed end customer of an entire chain of investments. The reasoning that underpins the sector works like this. OpenAI orders compute. Infrastructure providers borrow to build the data centers that will house it. These centers buy processors from Nvidia, memory from Micron, optical components from Applied Optoelectronics. Each link sizes its capacity—and often its debt—on the expected demand from the previous link.

This whole edifice rests on one assumption: that OpenAI will have the revenue to pay. When that revenue goes from $70 billion to $50 billion, it isn’t a spreadsheet line that moves—it’s the basis for calculating billions of investments already committed. Payback timelines lengthen, and every link in the chain sees its equation deteriorate at the same time. Oracle, which has committed considerable capacity for OpenAI, lost more than 5% on the day, more than the chip makers themselves.

A « false » gap!

This panic was actually caused by nothing more than different accounting methods between Anthropic and OpenAI!

These two companies are closely watched today, and many analysts try to compare them. The problem is that they don’t use the same method for recognizing revenue. Anthropic includes in its revenue the sales going through its partners, Amazon and Google, whereas OpenAI counts only its own share, making the comparison impossible without adjustments.

To get a comparable basis, some investors therefore applied Anthropic’s method to OpenAI’s revenue and arrived at $70 billion. This figure, which OpenAI never disclosed, then circulated from report to report, losing along the way the caveat that explained the adjustment.

Investors eventually understood the source of the error and were able to recover OpenAI’s « comparable » revenue of $70 billion. The panic was thus created by a simple misinterpretation.

The takeaway

This kind of event happens regularly in the markets and can push ill-prepared investors into serious mistakes: selling on Thursday on this news would have made the overly emotional investor miss Friday’s rebound—they’d have sold at the low and bought back higher. That’s several points lost that count toward year-end performance, especially if it happens more than once.

The only way not to fall into these traps is to have enough conviction in your companies to let you hold on even on the bad days. This conviction comes only from experience and from knowledge of the companies you invest in. That’s why you’ll find in your member area the complete analysis reports on the companies we invest in.