Anthropic committed $200 billion to Google Cloud last week — five gigawatts of TPU capacity over five years, beginning in 2027 — first reported by The Information on May 5. The contract represents roughly 40% of Alphabet’s $462 billion cloud backlog disclosed at Q1 earnings. Alphabet shares are up about 160% over the past twelve months and briefly passed Nvidia in after-hours market cap the day the report ran. Wall Street read the news as confirmation that “owning the stack” — chips, models, distribution — is the durable AI position.
The number underneath the rally tells a more complicated story. Anthropic’s annualised revenue run rate is $30 billion as of April 6. The $200 billion commitment averages $40 billion per year of cloud spend — more than the company’s entire current run rate, every year, for five years beginning 2027. To honour the contract, Anthropic needs the revenue trajectory it disclosed to investors: a roughly 30x scale-up by 2029. The math closes only if the curve holds.
This is not an isolated arrangement. According to The Information, Anthropic and OpenAI’s combined cloud contracts account for nearly half of the roughly $2 trillion of long-term backlog disclosed by the four largest hyperscalers (AWS, Azure, Google Cloud, Oracle). OpenAI alone has committed $250 billion to Microsoft Azure, which by Microsoft’s Q3 FY26 commercial RPO disclosure represents roughly 45% of the company’s $625 billion backlog. OpenAI is also locked into a reported $300 billion deal with Oracle over five years from 2027, a $38 billion seven-year agreement with AWS signed November 2025, and approximately $22.4 billion with CoreWeave. The cloud industry’s two largest forward customers are two private, cash-burning model labs.
The structure is more fragile because of its circularity. Alphabet has agreed to invest at least $10 billion in Anthropic, scaling to $40 billion if performance targets are hit. Amazon has committed up to $25 billion. The hyperscalers are simultaneously the suppliers Anthropic and OpenAI commit hundreds of billions to, and the investors funding the labs’ ability to make those commitments in the first place. Anthropic is now reviewing preemptive offers at valuations up to $900 billion; OpenAI is targeting a $1 trillion IPO. Both valuations rest on growth projections the hyperscalers themselves have priced in by underwriting the backlog. The money loops.
This complicates the central hypothesis on a layer the blog has not directly tested. The argument so far has been that AI commoditises the software layer — value compresses where AI can re-derive the product cheaply, value expands one layer up where customer-specific context cannot be re-derived externally (Above the Model Layer). The cloud-AI infrastructure layer looked like the second category — high capex, hard to replicate, defended by the same gravity that protects Palantir’s ontology. The $200 billion Anthropic-Google commitment appears, on the headline, to confirm that thesis.
The backlog telling that story is not durable revenue in the usual sense. It is contractual demand from two firms whose ability to honour the contracts requires a 20-30x revenue scale-up over five years and continued investor patience for cash burn. When Oracle lost roughly half its market cap over five months earlier this year after investors absorbed how much of its backlog was OpenAI-linked, that was the market refusing to extend the assumption indefinitely. Microsoft’s April 27 restructuring of the OpenAI partnership — capping revenue share, dropping model exclusivity — was the most exposed counterparty starting to de-risk. The Six-Year Moat treated those moves as commoditisation of model distribution. Read alongside the backlog data, they are also a re-pricing of customer concentration.
Hypothesis update: 55% → 54%. Complicates rather than falsifies. The mechanism — AI commoditising the software layer — remains supported by margin compression at the application tier (SAP, Atlassian, ServiceNow) and by the fortress-quadrant outliers above it (Palantir). The new evidence is that the apparent fortress at the infrastructure layer beneath the models is held up by two customers whose financial commitments are larger than their current revenue. The cloud-AI stack moat is real conditional on Anthropic and OpenAI continuing to scale into their contracts. It is not unconditional.
What to watch: whether Anthropic’s annualised revenue keeps pace with the curve implied by its $200 billion commitment; whether any hyperscaler books a non-AI-lab customer of comparable scale before the model-lab IPOs; whether OpenAI’s anticipated Q4 2026 IPO at a $1 trillion valuation gets the public market test required to validate the backlog; and whether the revenue-share cap and non-exclusivity terms of Microsoft’s restructured OpenAI deal show up at Google or Oracle next.
Sources: The Information, May 5 2026; CNBC, May 10 2026; CNBC, April 27 2026; TechCrunch, September 10 2025; CNBC, November 3 2025; Alphabet and Microsoft FY26 earnings filings.