Anthropic Can Pay for Compute With Revenue Alone: $68B a Year Even With Zero Growth

The most common skeptical argument about Anthropic is delivered as a taunt: you are simply extending the line on the chart upward. But let us draw the line horizontally. If revenue stops growing altogether, will it still be enough to pay for compute capacity? According to an analysis published by the Sioloshnaya channel and based on the updated S-1, yes, it will.

What the S-1 tells us

The updated S-1 revealed that Anthropic has signed $447B in contracts through the end of 2030—an average of about $105B per year. That figure includes obligations the company can exit. There is no exact number for the non-cancelable portion; ChatGPT estimates its maximum at $350B, or about $82B per year. The picture for the next 12 months is more modest: approximately $65.5B in non-cancelable obligations and $70B in total. Importantly, this is a language model’s estimate rather than a company disclosure, so it should be treated as an order-of-magnitude approximation.

The zero-growth arithmetic

Revenue in the latest quarter was $17B. Assume the worst: growth stops, but revenue does not decline. That produces $68B over a year—enough to cover the non-cancelable $65.5B and nearly all of the $70B total.

There are other operating expenses as well: at the third-quarter run rate, roughly $30B a year, of which about $18.5B is cash spending. Those costs would have to be covered from the balance sheet. But the company has more than $120B there, and the IPO is expected to add another $80–100B.

For context, the company earned $32B in the first nine months of 2026. The internal forecast cited by the author assumes a slowdown, not an acceleration. Quarters with growth of 100% or more are unlikely to recur. Growth is still expected, however.

What the bears are saying

The author asked GPT to compile pessimists’ forecasts. The result: even they assume some growth. One forecast expects growth to fall to 0.5% four quarters from now, but the preceding three quarters would still be positive. Others project modest growth in the tens of percent.

FutureSearch also deserves mention—an AI forecasting system that recently beat leading human forecasters in a competition. Its estimate: by May, Anthropic’s revenue will have grown 65%, while the pessimistic 10th percentile projects 25% growth. The 10th percentile means that 90% of the model’s scenarios produce a higher result.

What to do with this

For developers and founders building products on Anthropic’s API, this is an argument for the supplier’s resilience. Under this arithmetic, the risk that the vendor runs out of money for GPUs within the next year is low.

There are caveats. First, the figures for non-cancelable contracts are ChatGPT estimates. Second, the calculation covers only compute and operating expenses, not any future investments. Third, it addresses a 12-month horizon, not the period through 2030.

The practical takeaway: when choosing a vendor, look past slogans and run a simple stress test—“what if growth is zero?” The same exercise is useful for your own inference budget: calculate whether the unit economics can survive a flat line.