Anthropic Races for Record AI IPO at Up to $1.02 Trillion Valuation as $900 Billion Compute Commitment Draws Attention

In the third quarter, the company recorded an operating loss of $567 million, compared with an operating profit of $974 million in the second quarter. The return to a loss was primarily driven by $5 billion in high-cost short-term computing power and approximately $2 billion in non-cash charitable donation matching expenses. Excluding items such as stock-based compensation and non-cash donations, the company's third-quarter adjusted operating profit reached $2.5 billion, up from $2.2 billion in the second quarter; however, its adjusted operating margin fell from 19% to 15%.

For the first nine months ended September, Anthropic's cumulative operating loss was approximately $1.26 billion, narrowing significantly from a $5.22 billion loss in the same period last year. During the same period, adjusted operating profit reached $4.08 billion, showing that the Claude business has achieved a degree of profitability on a non-GAAP basis.

The prospectus showed a net loss of as much as $104 billion for the first nine months, but this figure cannot be directly viewed as an operating cash outflow. Approximately $94.3 billion of this resulted from changes in the fair value of convertible notes, derivative liabilities, and agreements for future equity. As Anthropic's valuation rose, these instruments, which will need to be converted into equity in the future, resulted in higher accounting liabilities, generating massive non-cash losses.

The latest prospectus filings show that computing, infrastructure, and lease commitments signed as of the end of September 2026 and thereafter total approximately $900.9 billion, of which about $451.6 billion is expected to occur before the end of 2030, while commitments due in the next 12 months stand at roughly $70.7 billion. These amounts are not all immediate debt obligations; some contracts will be recognized gradually as capacity is delivered, and certain arrangements include cancellation or financing conditions, but they still represent fixed expenditures that Anthropic will have to bear going forward.

Among major suppliers, Anthropic has committed to paying Microsoft at least $47.2 billion and up to $49.2 billion by 2033; paying Google at least $121.3 billion, with potential additional purchases of TPU computing power estimated at $125 billion to $150 billion; and committing to a minimum spend of $110 billion with Amazon. Multiple contracts feature 'pay regardless of actual usage' arrangements, leaving the company liable to bear high costs if future demand forecasts deviate.

Anthropic also disclosed that it has signed multiple Nvidia GPU computing power agreements with SpaceX's SpaceXAI, with a maximum contract value of approximately $120.5 billion extending through 2029. Full capacity is expected to be delivered by December 2026, at which point monthly expenses will average roughly $3.1 billion. However, after the initial non-cancelable periods end in January and March 2027, respectively, either party can terminate the agreements with 90 days' advance notice, meaning the $120.5 billion represents a contract ceiling rather than necessarily the final actual outlay.

These contracts are both a growth safeguard and a financial risk. Securing GPU and data center capacity in advance allows Anthropic to avoid missing market opportunities during computing shortages; however, if demand for Claude slows, model efficiency improves significantly, or enterprise customers turn to cheaper competing products, part of that capacity may end up underutilized.

This content was translated using AI and reviewed for clarity. It is for informational purposes only.