Anthropic discloses $895.4B in commitments. That is not $895B already spent.

Anthropic filed Amendment No. 1 to its S-1 registration statement with the US Securities and Exchange Commission on 2 November 2026. It is a materially new disclosure, not a reprint of its earlier prospectus. The amended filing includes the third-quarter 2026 results and a substantially fuller view of contracted computing and infrastructure exposure. Cortist had previously covered first-half 2026 revenue and compute expenses from the original S-1; the newer filing changes the financial evidence available for evaluating the company’s expansion.

The filing reports $16.956 billion of third-quarter revenue, up from $11.555 billion in the second quarter. Compute and infrastructure expenses reached $9.863 billion in the third quarter, compared with $5.558 billion in the second. Anthropic recorded a $567 million operating loss after $974 million in operating income in Q2. Those are reported, unaudited quarterly figures within a registration statement, not an independently certified forecast of future profitability.

The figure that changes the longer-term debate is approximately $895.4 billion of aggregate committed spend under purchase agreements and leases, including contracts entered into as of and subsequently to September 30. Anthropic says substantially all relates to compute and infrastructure; it expects approximately $447.5 billion of these commitments to fall before the end of 2030 and $69.8 billion within the following twelve months.

There are three different concepts that AI infrastructure coverage frequently mixes together: an expense recognized during a quarter, cash invested in equipment or facilities, and a contractual commitment to future payments. They are not interchangeable. The $9.863 billion number is an expense category in quarterly operations; the $895.4 billion figure is a multi-year schedule of purchase commitments and lease arrangements. Describing the latter as money already spent would be a serious accounting error.

The agreement structure also matters. Anthropic says approximately $213.8 billion of the aggregate total concerns arrangements subject to exercisable termination rights or contingent conditions, including partner financing, facilities and regulatory approval. The company warns that take-or-pay and other minimum commitments still create significant fixed obligations. Neither unconditional certainty nor the ability to cancel everything is a fair reading of the filing.

The statement of operations shows a $23.291 billion third-quarter net loss, but that should not be mistaken for the direct cash cost of running Claude. The filing includes large fair-value movements on convertible notes and related instruments, which complicate a simple comparison between net loss, operating loss and cash burn. For assessing operations, readers should keep those accounting measures distinct and read the corresponding notes.

Financial risks are intertwined with a technical delivery question: Anthropic needs the promised accelerators, power capacity and buildings to arrive on schedule and at expected cost. Delays, higher funding costs, underutilization or a mismatch between contracted capacity and demand would change the economics even if model performance improves.

Cortist follows claims about AI capabilities, adoption and infrastructure as future events that require dated evidence. The amended S-1 offers unusually concrete evidence of how much a frontier model company is willing to contract for future capacity. It does not prove that its future models will achieve certain benchmarks, that the entire contracted capacity will be delivered, or that the investment produces a predictable return.

The original October filing supported an article about first-half revenue and compute intensity. This new report focuses only on the amendment’s incremental disclosures: the third-quarter result, the multi-year commitment structure, and the conditional portion of those contracts. Readers should not conflate the two reports or sum overlapping values from different periods.

An independent accountability test will compare subsequent SEC filings with this schedule. Do committed amounts turn into delivered, utilized capacity? Are computing costs per unit of useful work improving after temporary high-cost purchases roll off? Does operating performance improve when revenue grows, or do new commitments outrun cash generation? The outcome should be measured against the dates and definitions in the filing, not a market valuation headline.