Every day, one story that matters, broken down into three numbered scenarios, each with its own probability. Never fixed — it shifts as the situation changes.
Scénario
Thursday, global economy & finance
Wall Street: Can the Rally Last?
Published on September 3, 2026
Share:
The question
Despite a CAPE ratio at its highest since 1999 and record margin debt, Wall Street keeps hitting new records: is the rally standing on solid ground?
The facts
On September 2, 2026, the S&P 500 closed at 7,668 points, within 2% of its summer record. Since January 1, the index has already notched 25 record closes and gained 12.7%. On paper, Wall Street has never run this well. But one number is worrying a growing number of analysts: the US stock market has only been this expensive once in 125 years of history — right before the dot-com bubble burst.
This rally doesn't rest on the whole market. Two companies alone, Nvidia and Micron, account for nearly a third of the S&P 500's earnings growth in 2026. More broadly, the index's ten biggest stocks now make up 35% of its total market value — up from 25% at the peak of the dot-com bubble in 2000. The seven tech giants nicknamed the "Magnificent Seven*," led by Nvidia at 7.5% of the index on its own, together account for more than a third of Wall Street.
To judge whether a market is expensive, economists use the CAPE ratio*, devised by economist Robert Shiller. In early September 2026, it stands at 41.9 — the second-highest level ever recorded since 1881, behind the 44.2 reading of December 1999, just before the dot-com bubble burst. One caveat: compared with bonds, stocks are, for now, slightly less overvalued than they were back then.
Comprendre
The CAPE ratio doesn't look at this year's earnings, but their ten-year average — so it isn't caught off guard right before a downturn.
A standard P/E divides the price by this year's earnings — a figure that can itself be inflated in the middle of euphoria, making the market look falsely reasonable at the worst possible moment. By averaging ten years of real, inflation-adjusted earnings, the CAPE ratio smooths out these cyclical swings and measures whether today's price is justified by durable earning power, not a passing peak.
Part of this surge is explained by a more recent phenomenon: circular financing* in artificial intelligence. Nvidia, Oracle, OpenAI, SoftBank and Amazon are funding, supplying and investing in one another, in a tangle of contracts worth staggering sums. Concretely, here's who has pledged what to whom, from the largest amount to the smallest:
Comprendre
In circular financing, the seller becomes the investor in its own customer.
Nvidia sells chips to OpenAI. But Nvidia also invests in OpenAI, and guarantees part of its data centers. So a good share of the money injected flows back to Nvidia itself, in the form of chip purchases. This arrangement inflates the reported revenue on both sides, without proving that a customer outside the loop is willing to pay as much.
Who's funding whom: AI's great tangle around OpenAI
1Oracle → OpenAI: up to $300BConstruction of the Stargate project's data centers, paid for by OpenAI under long-term contracts.
2Nvidia → OpenAI: up to $105BResidual value guarantees on a data center in Ohio, where Nvidia will also be the exclusive chip supplier.
3Amazon → OpenAI: $50BInvestment commitment made during the February 2026 funding round.
4Nvidia → OpenAI: $30BDirect investment in the same funding round, on top of the guarantees above.
5SoftBank → OpenAI: $30BSame funding round, which valued OpenAI at $730B before this capital was added.
In total, analysts tracking these arrangements estimate the web of cross-commitments around AI at more than $800B — far beyond the $25B in annual revenue OpenAI is expected to generate in 2026.
OpenAI illustrates the scale of the bet: the company is targeting roughly $25 billion in revenue in 2026, but is also expected to lose close to $14 billion, against cumulative infrastructure commitments nearing $1.4 trillion over eight years. Until end-user demand catches up with these investments, the profitability of the whole structure remains unproven. Another signal to watch: margin debt*, the money investors borrow to buy even more stock. It hit a record $1.502 trillion in June 2026, after jumping 77% in fourteen months — a surge of a scale comparable to the ones that preceded the last three major market turning points, in 2000, in 2008 and in 2022. By July, it had already pulled back to $1.417 trillion.
Michael Burry, the investor made famous for his winning bet against subprime mortgages (told in "The Big Short"), sees in AI several of the traits he spotted before previous bubbles: valuations that assume smooth, uninterrupted growth, capital piling up in cascading rounds, and a growing reliance on debt financing. Since summer 2026, he has built bearish positions on Nvidia, Micron and the SOXX semiconductor ETF. He isn't setting an immediate deadline: he points instead to 2028 as the horizon for a possible break, when installed computing capacity could outstrip real demand — a nuance that matters, since spotting a bubble never tells you when it will burst.
The Financial Times draws a useful distinction between two overlapping bubbles behind this rally: an industrial bubble, in data center and chip investment, and a financial bubble, in stock valuations. The most indebted players funding their expansion are the most exposed to forced consolidation if the second one bursts — the "neocloud" CoreWeave already carried $35.6 billion in debt in mid-2026, against $2.6 billion in quarterly revenue (up 112% year over year). But a financial bubble bursting doesn't erase the infrastructure already built, which keeps on running.
The climate looks no calmer on the rates front. The yield on 10-year US Treasuries climbed back to 4.79% in early September, its highest level since late 2023, as markets start pricing in a possible Fed rate hike on September 16 rather than the pause hoped for until now. The US midterm elections in November are shaping up as an additional source of jitters. We already looked at this uncertainty around central bank policy, see our article for more. That's why this topic lends itself to three distinct scenarios: either earnings catch up with prices and the rally continues without a hitch; or the market stays expensive and volatile, without a crash or a clear acceleration; or one of these warning signs triggers a severe correction.
S&P 5007,668 pts ▲ +12.7% since end of 2025 (6,845 pts), 25th record of the year
Valuation (Shiller CAPE ratio)41.9x ▲ 2nd-highest level in 125 years (vs 37.1x end of 2025), behind Dec. 1999 (44.2x)
S&P 500 valuation (Shiller CAPE ratio), 2009-2026
The market has only been this expensive once in 125 years: at the peak of the dot-com bubble.
Each year reflects the CAPE ratio's year-end level (inflation-adjusted earnings over 10 years) — no continuous movement between points. Source: Robert Shiller (Yale), aggregated via multpl.com — see Sources at the bottom of the edition.
Favorable, stable or degraded
Can the Rally Continue Without a Correction?
What we're assessing
FavorableAI giants' earnings catch up with their valuations, without a major correction.
StableThe market stays expensive and concentrated, but holds without a crash or a clear acceleration.
DegradedA loss of confidence in AI or debt triggers a severe Wall Street correction.
Favorable
20%
Unlikely
AI Earnings Eventually Catch Up With Prices
The AI giants' quarterly results, due in late October, beat expectations: corporate demand for AI tools finally broadens beyond the tech giants funding each other. S&P 500 earnings grow fast enough to catch up with already-high prices, without anyone needing to sell in a panic. The Fed holds rates steady on September 16, which reassures markets. The CAPE ratio stabilizes around 38 to 40 — still very high by historical standards, but without a fresh excess.
This is the least likely of the three scenarios: it assumes end-user demand catches up, within a few months, with a circular-financing structure built on more than $800 billion in cross-commitments. Less likely than the stable scenario, where the market simply holds without justifying itself any further; and far less likely than the degraded scenario, for which the warning signs — margin debt, long-term rates — are already in place.
Indicators affected
S&P 500 (projected year-end)~8,100 pts↑(vs 6,845 pts, end of 2025)
Valuation (CAPE ratio)~39x↑(vs 37.1x, end of 2025)
The France angle
French investors' stock portfolios exposed to US equities, through life insurance policies or employee savings plans, keep growing — but a Fed pause, just as the market starts pricing in a hike instead, tends to weaken the dollar: part of the equity gain erodes once converted back into euros. ↑ Favorable for France, tempered by currency exposure.
Stable
45%
Likely
The Market Stays Expensive, But Holds
No shock comes along to pop the bubble, but no solid proof emerges either to justify such high prices. The S&P 500 grinds higher in fits and starts, pulled up by a handful of tech stocks while the rest of the market stagnates. The Fed wavers between holding and hiking, bond yields stay elevated, and the November midterms add jitters without derailing the indices. The CAPE ratio holds around 40 to 44, as high as today, neither pulling back nor climbing further.
This is the most likely of the three scenarios: historically, bull markets last on average three times longer than bear markets, and most analysts describe elevated risk rather than an imminent crash. More likely than the favorable scenario, which assumes a smooth resolution rarely seen in history; also more likely than the degraded scenario, which requires an actual trigger — not just warning signs — to materialize.
Indicators affected
S&P 500 (projected year-end)~7,550 pts↑(vs 6,845 pts, end of 2025)
Valuation (CAPE ratio)~42x↑(vs 37.1x, end of 2025)
The France angle
French savers in unit-linked accounts keep the benefit of gains already banked, even if the market stays expensive and could remain jittery for months. ↑ Rather favorable for France.
Degraded
35%
Likely
A Severe Correction Wipes Out Part of the Gains
One of the warning signs already visible ends up triggering the correction: a pullback in margin debt, disappointing results from an AI giant, or a Fed rate hike that pushes the 10-year yield above 5%. Or, more unpredictably, a black swan* hits without warning, unrelated to these already-watched signals. The S&P 500 falls 10 to 20% from its summer record, dragging the most concentrated stocks down with it. The CAPE ratio drops back to 32 to 35, still high historically but far from its September peak.
This scenario remains less likely than the stable one, since it requires an actual trigger — not just warning signs already known to markets — to truly materialize. But it's markedly more likely than the favorable scenario: margin debt has already started pulling back in July, the 10-year yield is at its highest level since 2023, and several Wall Street strategists are already flagging a possible 10 to 20% correction, or more.
Indicators affected
S&P 500 (projected year-end)~6,350 pts↓(vs 6,845 pts, end of 2025)
Valuation (CAPE ratio)~33x↓(vs 37.1x, end of 2025)
The France angle
A 10-to-20% Wall Street correction drags down French portfolios exposed to US equities, through life insurance policies or employee savings plans, with a possible knock-on effect on the CAC 40. ↓ Rather unfavorable for France.
Rough estimates for the 3 scenarios above, based on the information available at publication and reassessed as the situation changes — never guaranteed forecasts. Learn more about our method →
The essentials
Carried by a handful of AI giants, can Wall Street's rally continue without a sharp correction?
Wall Street's CAPE ratio, which measures how expensive the market is relative to the past ten years of earnings, reaches 41.9 in early September 2026 — its second-highest level in 125 years, behind the dot-com bubble peak in 1999.
The most likely scenario (45%): the market stays expensive and volatile, without a crash or a clear acceleration, carried by a handful of tech stocks while the rest of the index stagnates.
Signal to watch: the Fed's rate decision on September 16, 2026, and the AI giants' quarterly results in late October.
Slightly positive
Our assessment of the impact on France: slightly positive. With a 65% chance that the favorable and stable scenarios (20% and 45%) keep French savers' unit-linked holdings at their current level or above, against a 35% risk of a correction that would push them down.
An indicator devised by economist Robert Shiller, which compares a stock index's price to its average earnings over the past ten years, adjusted for inflation — the higher it is, the more expensive the market is judged to be relative to its real earnings.
Circular financing
An arrangement where a company invests in one of its own customers or suppliers, so that the money it injects eventually flows back to it in the form of purchases — which can artificially inflate the reported revenue on both sides.
Margin debt
Money an investor borrows from their broker to buy more stock than their own capital would allow, which amplifies gains when prices rise — but also losses when they fall.
Magnificent Seven
The nickname for seven US tech giants (Nvidia, Apple, Microsoft, Alphabet, Amazon, Meta, Tesla), whose weight in stock indices has become so large that they alone drive a significant share of market performance.
Black swan
A rare, unpredictable event with enormous consequences, explainable only after the fact — the term, popularized by essayist Nassim Nicholas Taleb, describes an external shock unrelated to the risks markets already track and watch.
