AWSs 15-Quarter Record|The 1.2B Cash Floor Nobody Priced In?
The Day the Cloud Numbers Split in Two
Amazon Web Services just posted the fastest revenue growth it has seen in fifteen quarters. Twenty-eight percent year over year, $37.6 billion in a single quarter. The headline landed exactly the way bulls wanted it. Then the cash flow number arrived.
Free cash flow on a trailing twelve-month basis: $1.2 billion. Twelve months ago, the same figure was $25.9 billion. That is not a rounding error — that is a $24.7 billion hole carved out of the same business while revenue was accelerating.
The US market spent most of Thursday threading between two simultaneous narratives. The first was the AI infrastructure boom in full sprint. CoreWeave reported $2.08 billion in first-quarter revenue, topping estimates, while revealing it had raised the floor on its 2026 capital expenditure to $31 billion, keeping the ceiling at $35 billion. The company now holds $99.4 billion in contracted revenue backlog, up from $66.8 billion at year-end. Alphabet continues to close the market-cap gap with Nvidia — as of Wednesday's close, Alphabet sat at $4.81 trillion versus Nvidia's $5.05 trillion, a gap that has compressed sharply since Alphabet's blockbuster April 29 earnings report showed Google Cloud revenue growing 63% to $20.03 billion.
The second narrative ran underneath. An equity analyst at Porter and Company published a warning that the earnings powering Big Tech valuations are built on a circular cash loop. OpenAI and Anthropic account for an estimated 49% of Microsoft's $627 billion cloud backlog and 51% of Amazon's $464 billion backlog. Alphabet and Oracle show the same pattern at 43% and 54%, respectively. The logic the analyst flagged: hyperscalers extend capital to AI labs, the labs spend that capital on hyperscaler compute capacity, and the resulting cloud revenue flows back onto the hyperscalers' income statements. The "E" in the price-to-earnings ratio, on this reading, is partly self-funded.
Against that backdrop, the AWS free cash flow implosion is not incidental. Amazon spent $43.2 billion in capital expenditure in the first quarter alone, and the company has projected full-year 2026 capex near $200 billion. The cash that should be accumulating behind the revenue growth is instead being redirected, at record speed, into the same infrastructure the AI labs will rent back.
When the Revenue Engine Feeds Itself
The mechanism worth examining is not the capex figure in isolation — it is the sequence. Amazon lands a partnership with Anthropic worth up to five gigawatts of Trainium capacity. Anthropic needs that compute. Amazon books the contract into its AWS backlog. Anthropic, having received investment capital partly from Amazon and from other hyperscalers, routes spending back to AWS to fulfill the compute commitment. AWS revenue rises. The backlog swells. The earnings per share beat consensus. And Amazon's stock is described as having an "AI-driven flywheel."
The flywheel is real in mechanical terms. The question the cash flow number introduces is whether the rotation is externally powered or internally sustained. At $1.2 billion in trailing free cash flow against a market cap that reflects a very different trajectory, the gap between the income statement and the cash statement is where the risk lives — not in the revenue line.
This is where a counter-signal emerges that complicates the warning. Amazon's own custom silicon is scaling at a trajectory that has nothing circular about it. Graviton, Trainium, and Nitro crossed a $20 billion annual revenue run rate in the first quarter, growing at triple-digit percentages. Amazon landed over 2.1 million AI chips in the trailing twelve months, more than half of them Trainium. Bedrock processed more tokens in Q1 than in all prior years combined, with customer spend rising 170% sequentially. These are organic enterprise workloads — companies that are not OpenAI, not Anthropic, choosing Amazon's proprietary infrastructure because it delivers better price-performance than renting Nvidia H100s.
The reversal condition is precisely here. If Trainium adoption continues accelerating among third-party enterprise customers — the ones with no circular relationship to Amazon — then the capex being deployed now converts into durable, non-circular cash generation in 2027 and beyond. That is the Amazon-early-days argument: sacrifice margin now, own the infrastructure at cost advantage later. The trap version of the same story is that enterprise Trainium adoption plateaus, the AI lab contracts remain the dominant backlog driver, and the circular dependency deepens as capex commitments extend further into the future.
The two scenarios are not distinguishable from the Q1 revenue figure alone. They require watching what the free cash flow does as the Trainium ramp matures.
The Benchmark That Will Settle the Argument
The historical parallel is explicit. Zacks Investment Research compared CoreWeave's current posture to Amazon's early e-commerce era — accepting deep losses in pursuit of infrastructure dominance. Amazon's investors who held through the 2001 and 2008 periods when free cash flow was negligible were eventually rewarded as the cost structure compounded in their favor. The analogy is structurally coherent. It is also the same argument that was applied to every capital-intensive infrastructure bet that eventually failed to achieve the scale required to make the math work.
What distinguishes the two outcomes, historically, has been whether the infrastructure became the default substrate for the next generation of workloads — or remained a specialized layer that incumbents eventually replicated at lower cost. AWS won the first cloud cycle because enterprises could not afford the upfront capital to match it. The question for this cycle is whether Nvidia's H100 and H200 installed base, combined with the hyperscalers' own competing silicon, creates a ceiling on any single provider's compute advantage.
For AMZN specifically, the near-term verification point is the second-quarter free cash flow number, expected alongside Q2 results. Amazon guided Q2 operating income of $20 to $24 billion. If operating income comes in at the top of that range and capex growth begins to decelerate even modestly from the Q1 pace, the cash flow gap starts to close — and the circular-backlog concern becomes a theoretical footnote to a real recovery story. If capex holds at the Q1 pace and operating income lands at the floor, the gap widens and the backlog composition becomes the next question analysts pursue.
The Microsoft read is similar. Azure is doing the heavy lifting on revenue, with cloud backlog at $627 billion, but the same 49% OpenAI concentration applies. MSFT trades at a forward multiple that embeds substantial backlog conversion — the circular portion of that conversion is the variable the market has not yet fully priced.
Alphabet enters this picture from a different angle. Google Cloud's $460 billion backlog carries the same Anthropic concentration, but Alphabet's custom TPU silicon and the 63% cloud growth rate suggest organic enterprise pull that is harder to dismiss. Alphabet's forward P/E of 28 sits well below Nvidia's 42 — a valuation that reflects either more conservatism or more uncertainty about which part of the backlog is durable.
The directional lean here tilts toward the AWS-infrastructure-wins scenario, but only conditionally. The Trainium ramp is real and measurable. The circular backlog concern is also real and measurable. They are competing against each other on the same timeline. The free cash flow number in August is the benchmark — not the revenue figure, not the backlog total. If it recovers toward double-digit billions, the Amazon-early-days thesis gets its first confirmation. If it stays compressed near $1 billion while capex remains above $40 billion per quarter, the question about backlog quality moves from analyst note to market narrative.
What would prove the infrastructure thesis wrong faster than anything else is an Anthropic or OpenAI decision to diversify compute away from Amazon. That has not happened. It remains the condition to watch.