Anthropic's S-1 shows the price of the frontier: over $100 billion lost on $33 billion in revenue

Anthropic lost more than $100 billion in the first nine months of 2026, according to the S-1 registration statement it filed with the US Securities and Exchange Commission (SEC). Over the same period, which ended September 30, the company reported $33.24 billion in revenue. Reported losses were roughly three times revenue.

The filing also says Anthropic has no plans to slow down. The company says it has committed $895.4 billion to AI spending for the foreseeable future. That is about 27 times its nine-month revenue. The financial disclosure follows an earlier S-1 filed last week, which signalled a NASDAQ listing under the ticker ANT. Shacknews, which first reported the figures, says trading is expected to begin later in November.

The exact loss figure needs care. Shacknews gives net losses of $104.04 billion in its text and $101.04 billion in the chart taken from the SEC filing. It also cites a combined figure of $137.28 billion, which is not a standard accounting measure. The safe reading is "over $100 billion" until the S-1 itself is checked line by line. Shacknews also says net losses rose by almost $75 billion compared with 2025. The report does not say how much of the loss is cash spending and how much is accounting items such as depreciation, stock-based pay or contract commitments. That split matters a lot when judging how sustainable the business is.

The context explains the numbers. On our own Zendoric quality index, Anthropic leads the frontier with Fable 5, ahead of OpenAI's GPT-5.6. The S-1 shows what that lead costs. In this industry, staying at the frontier means training ever-larger models and buying compute years ahead. Revenue is real and large: $33 billion in nine months is a company-scale figure, not a demo. But costs are growing faster. The pattern is familiar from earlier infrastructure booms such as railways, fibre and cloud, where builders spent ahead of demand. Some created lasting value. Others left the infrastructure behind for whoever came next.

Our reading: this is the moment the AI race stops being a story about benchmarks and becomes a story about capital. Our thesis is that Anthropic's losses are not yet evidence of failure. They are the explicit price of a strategy that assumes compute becomes cheaper and demand keeps compounding. Both assumptions are plausible, but the S-1 does not prove either. In the short term, the risks are real. Becoming a public company brings pressure every quarter, a stock price that swings with sentiment about the "AI bubble", and a temptation to put revenue ahead of the safety caution that set Anthropic apart. Spending on this scale also concentrates power in very few companies, because only a handful can raise this much money. Meanwhile, open-weight models from China and elsewhere keep closing the gap at a fraction of the cost.

In the long term, we stay optimistic, with caveats. Much of this spending goes into compute capacity that will remain after the current model generation and will likely get cheaper per task. That capacity is what can eventually speed up drug discovery and disease research and lower the cost of knowledge work for everyone. The question public investors are really being asked is not whether AI is valuable. It is whether Anthropic can turn a three-to-one loss ratio into margins before patient capital runs out. What to watch next: how cash burn is broken down, the terms of the $895.4 billion commitments, and whether the first earnings reports as a public company show costs per unit of revenue falling.