The Money

The AI investment boom has accounted for nearly half of S&P 500 earnings growth this year, and this tailwind should begin to fade next year even as capex spending continues to grow.

— Goldman Sachs

Ben Snider, Goldman Sachs’s chief US equity strategist, in a note to investors. His base case is not a burst bubble but a marked slowdown in index earnings growth over two years. He adds that the surge in chip-makers’ margins leaves S&P 500 earnings exposed to a fall in chip prices, and that ‘other income’ — paper gains the largest technology companies book on private investments — would, if it vanished entirely in 2027, take 8 points off earnings growth.

We still have an incredible balance sheet.

— Sarah Friar

Where necessary, we may provide modest residual value guarantees, which are contingent liabilities we view as low risk, supported by the strong profitability trajectory of these labs and the sustaining value of the underlying assets.

— Amie Thuener

In 2027, we have secured the supply to again double AI revenue to approximately $115 billion. Our demand actually exceeds this outlook, and we will work to improve supply. In 2028, we expect the trajectory of growth to continue. We have line of sight for fiscal 2028 AI semiconductor revenue growth to again double to $230 billion.

— Hock Tan

The issue is not simply that investors are borrowing more, but that leverage is interacting with high valuations and market concentration, in particular the increasing cross-investment between artificial intelligence (AI) companies and hyper scalers, in a way that could amplify a future market correction. I remain concerned therefore that a large shock or combination of shocks could concurrently trigger multiple vulnerabilities.

— Andrew Bailey

Revenue for the third quarter of fiscal 2027 is expected to be $108.0 billion, plus or minus 2 percent, with no Data Center compute revenue from China assumed.

— NVIDIA

Is this circular financing? No. OpenAI will pay the lease.

— Jensen Huang

Jensen Huang published a long defence of the Ohio guarantee on the morning it was signed, answering the charge that has followed Nvidia since it committed up to $100 billion to OpenAI in September 2025. The post concedes the premise the instrument rests on — frontier laboratories “are growing faster than their balance sheets and long-term credit profiles can support” and “lack the decades-long infrastructure contracts and investment-grade financing capacity” to secure sites on their own — and argues the answer is not circularity but supply-chain discipline: Nvidia secures land, power and shell the way it secures chips and memory, on visible demand. It also puts numbers on the upside. Each generation of systems deployed at PORTS-Pike could mean about 1.5 million GPUs and “approximately $150 billion to $200 billion in NVIDIA revenue”; OpenAI's existing and planned commitments amount to about 12 gigawatts of Nvidia compute, rising to roughly 16 if Nvidia takes up its option, or “roughly $600 billion of NVIDIA compute through 2030.” Huang describes the guarantee as “limited to defined portions of lease and power payments, along with a specified residual-value commitment — not the full cost of the site,” declining as OpenAI makes payments and capacity comes online, and argues the capacity is resellable because CUDA makes Nvidia compute fungible: “Versatility makes it fungible. Fungibility drives utilization and durability — making NVIDIA compute a productive asset: rentable and financeable.”

Anthropic told investors its annualised revenue run rate reached $65 billion at the end of July, and that preliminary second-quarter revenue exceeded $11.5 billion against $787 million a year earlier, with positive adjusted operating income.

— Anthropic

The figures reached the market the way private-company figures do: through documents shown to prospective investors and a weekend investor update, reported by Bloomberg and confirmed by CNBC through three people familiar with them. Anthropic declined to comment on the quarterly number. Note the disclosure forms, which are three different things and are routinely conflated. Quarterly revenue: $4.73 billion in the first quarter of 2026, more than $11.5 billion in the second, against $787 million in the second quarter of 2025 — a fourteenfold increase, and the company's first quarter of positive adjusted operating income. Annualised run rate: above $47 billion in May, $65 billion at the end of July. Full-year revenue: roughly $10 billion for all of 2025, as the company characterised it in May. The last two of those cannot be reconciled with the $787 million second quarter from anything public, and nothing here is audited or filed; the audited numbers become public only in an amended registration statement if the company proceeds to list. For scale, OpenAI's run rate passed $40 billion in the same month, though Bloomberg notes the two companies' accounting methods differ enough that the figures are not directly comparable.

Historical experience suggests that technological revolutions carry risks of a boom-bust cycle in asset prices, and this risk does not depend on today's valuations being rational or irrational.

— European Central Bank

Five European Central Bank economists — Malin Andersson, Johannes Breckenfelder, Stefano Corradin, Kalin Nikolov and Maria Antonietta Viola — set out the euro area's exposure to an AI-driven correction on the ECB's blog, which carries the standard disclaimer that the views are the authors' rather than the institution's. Their argument is deliberately undated: both the rational explanation for high technology valuations (uncertainty about a new technology's productivity, which becomes economy-wide and undiversifiable as adoption spreads, raising the risk premium) and the behavioural one (overconfident investors bidding past fundamentals) imply a correction at some point, and “the exact timing is unknowable in advance.” What is measured rather than argued is the exposure. The S&P 500's cyclically adjusted price-to-earnings ratio sits close to its historical peak. Euro-area households hold around €440 billion of exposure to US technology equities, mostly through mutual funds and exchange-traded funds rather than direct holdings, “without necessarily being aware of the associated concentration risk”; insurers and pension funds hold significant exposures too. The fund structure is itself a transmission channel, since redemptions force sales that push valuations down further. Euro-area equities are less richly valued and its technology sector smaller, which limits the risk of a home-grown crash — but the two markets are historically highly correlated, and, unlike in the dot-com episode, there is markedly less room to cut interest rates or use fiscal policy to cushion a fall.

OpenAI's annualised revenue run rate has surpassed $40 billion, roughly double its pace at the end of 2025, with monthly run rate growing more than 20% in July alone.

— OpenAI

Bloomberg reported the figure; president Greg Brockman gave the July growth rate to staff in an internal note. The company attributes the acceleration to subscriptions, early advertising, the Codex coding agent and enterprise ChatGPT products, and has cut prices on some models to hold share against cheaper rivals. The comparison that matters is with the audited record already here: OpenAI's chief financial officer said the company closed 2025 with an annualised run rate above $20 billion, while its audited 2025 revenue was $13.07 billion — the gap between an exit-month run rate and a year's realised revenue, which is the same gap this $40 billion figure carries. It is a run rate, not a filing.

To be clear, the $100 billion ARR in December is not a question mark. That's what we would achieve if we basically did nothing.

— Elon Musk

On SpaceX's first earnings call as a public company, chief executive Elon Musk told investors the AI buildout would pay back fast, saying $100 billion in annualised recurring revenue by December was 'not a question mark' — though his arithmetic assumes closing the $60 billion Cursor acquisition. Chief financial officer Bret Johnsen said the company was seeing 'less than a one-year payback' on its AI-compute capital spending and had already contracted $6.7 billion of additional cloud-services revenue in the first weeks of the third quarter. Musk separately said SpaceX's trillion-dollar-revenue target had moved up from 2031 to 2030, with a 'non-zero chance' of 2029 — against FactSet's roughly $207 billion estimate for 2029. Investors weighed the $18.4 billion of capital going out this quarter more heavily than the promised return, sending the stock down about 8%.

Named investors called AI-era markets a bubble: Ray Dalio saw 'classic signs' of one reminiscent of 1929 and 2000 and agreed with Jeremy Grantham that this is 'the biggest investment bubble in American history'; Goldman Sachs's Peter Oppenheimer partly dissented — 'there does not appear to be a valuation bubble … there may be an earnings bubble' — while Apollo's Torsten Slok declared 'the 60/40 portfolio is broken.'

— Ray Dalio

A cluster of prominent investors weighed in on the returns question within days of SpaceX's IPO. Ray Dalio, the Bridgewater founder, said there are 'classic signs' of a bubble reminiscent of 1929 and 2000, singling out the surge of AI equity issuance — SpaceX's record IPO, with Anthropic and OpenAI heading toward roughly trillion-dollar listings — as one of the two forces that 'prick' a bubble, alongside rising rates. He endorsed GMO co-founder Jeremy Grantham's verdict of 'the biggest investment bubble in American history,' a 'bubble within a bubble'; Grantham's January 2026 paper with Edward Chancellor put price/book and cyclically adjusted earnings multiples at extremes 'surpassed only in 1929, 1972, 1999–2000, and 2021.' The dissent is a nuance worth recording: Goldman Sachs's chief global equity strategist Peter Oppenheimer argued on August 3 that the risk is 'an earnings bubble,' not a valuation one, and the same day Apollo's Torsten Slok said 'the 60/40 portfolio is broken.' Because the four voices range from full bubble call to partial dissent, the aggregate is recorded as a single dated statement of the conversation, not a consensus.