Anthropic's Filing Turns Amazon's AI Bet Into a Ten-Year Contract

When Anthropic filed its S-1 registration statement with the SEC on Oct. 26, the tech world received its first audited look inside a frontier AI laboratory. For years, investors treated generative AI like an expensive mobile game where venture funds foot the bill. Anthropic's filing changes that narrative. Tucked inside the disclosure is a ten-year guarantee that turns Amazon's massive AI wager from an open-ended bet into a legally enforceable revenue stream.

To see why that matters, look at what Amazon actually sells. Most people know Amazon as the storefront that drops packages on your porch, bills your monthly video subscription, and stocks your toothpaste. But its primary profit engine is Amazon Web Services. AWS is the digital landlord of the AI boom. It builds the specialized warehouses, packs them with servers, and rents out computing power by the hour. Anthropic, which makes the Claude chatbot, is simply one of its anchor tenants.

The relationship between the landlord and the tenant, however, runs in an uncanny loop. Anthropic distributes Claude through Amazon's cloud marketplace, handing Amazon a sales fee. Anthropic then takes its revenue and buys compute time on AWS servers to train next-generation models. Finally, Amazon marks up the value of its equity stake in Anthropic on its own corporate ledger. It is one counterparty playing three roles: customer, vendor, and investment holding. The money travels in a neat circle, and Amazon collects a toll at nearly every turn.

For the broader AI sector, the filing provides the first hard evidence that frontier labs can outgrow their infrastructure bills. The prevailing fear has been that training models consumes cash faster than software can bring it back. Anthropic flipped that equation.

In the six months to June 30, 2026, its revenue rocketed to $16.29 billion from $1.16 billion a year earlier—a 1,300% surge. Over the same span, its compute and infrastructure expenses climbed to $8.95 billion, falling from 201% of revenue down to 55%. If you wanted proof that generative AI software can outrun its raw processing costs, this is the ledger.

Yet the filing also highlights the quirks in how Amazon reports its profits. Under fair-value accounting rules, public companies must regularly re-price private equity holdings, recording paper gains directly onto their income statements before any cash is realized. That mechanism explains why Amazon looks deceptively cheap right now, trading roughly 14% below its 52-week high and about a quarter above its low.

Its trailing earnings multiple sits near 20x, while its forward multiple jumps near 24x. That rare inversion happens because trailing earnings per share were artificially lifted to $12.43 against a forward consensus of $10.43. Amazon’s private equity portfolio swelled from $16.2 billion to $122.3 billion in six months, driven by $62.8 billion in upward fair-value marks on Anthropic preferred stock.

That dynamic defines what has to happen for the equity to re-rate from here. Across the last four quarters, Amazon poured roughly $169 billion into capital spending, net of asset-sale proceeds, against roughly $161 billion of operating cash flow, leaving free cash flow slightly negative. Paper accounting gains cannot fund datacenters indefinitely. For Amazon to command a durable premium on its $2.70 trillion market valuation, AWS must prove that commercial AI revenue generates real operating margins even as heavy depreciation charges hit the books.

The bear case is that these paper profits can unwind just as quickly as they were booked. Anthropic left the offering size and price range blank in its S-1, meaning Amazon's carried valuation has no public market confirmation. If the IPO prices below those marks, tens of billions of dollars in paper gains reverse into immediate accounting charges.

Worse, that ten-year, $110 billion of non-cancelable revenue rests entirely on a counterparty that posted an $80.7 billion net loss in the half and has never delivered a full-year profit. If enterprise cloud appetite cools, Amazon remains stuck with roughly $170 billion in annual capital spending that depreciates regardless of utilization. The short thesis locks onto two clear triggers: an IPO clearing below Amazon's carried marks, and AWS quarterly revenue growth falling behind its capex trajectory.

The moment of truth arrives with Amazon's third-quarter report on Oct. 29, after the closing bell. Investors need to see AWS revenue growth holding firm while capital expenditure growth finally decelerates. Paper marks built the headline earnings, but cash flow must sustain them.

Disclosure: The Signal holds no position in AMZN. Positions may change. This is not financial advice.