Anthropic has made public a contrast that captures today’s AI economy: extraordinary commercial growth paired with infrastructure commitments measured in hundreds of billions. In an amended filing with the Securities and Exchange Commission (SEC), the company behind Claude reports $17 billion in revenue for the quarter ended September 30, 2026. The same document lists approximately $895.4 billion in purchase and lease commitments, primarily tied to computing power and infrastructure.
The figure is striking, but it must be read carefully: it is not a bill due today, nor a cost entirely recorded in the quarter. Anthropic says approximately $69.8 billion is due in the 12 months following September 2026, nearly $447.5 billion before the end of 2030, and substantially all of the remaining balance is expected between 2031 and 2045. This is a long-term capacity commitment, not a simple measure of cash immediately spent.
Growth at a Dizzying Pace
According to the filing, revenue for the first nine months of 2026 reached $33.2 billion, compared with $2.5 billion over the same period in 2025. The number of customers generating at least $100,000 in trailing annual revenue exceeded 11,900 as of September 30, and those customers represented about 74% of revenue over nine months. Demand is no longer coming only from Claude’s early adopters: it is spreading to businesses integrating the model into operations and commercial products.
But selling more AI services requires the ability to run them. Advanced models rely on clusters of processors, networks, storage and electricity. Computing capacity often has to be reserved well before it is delivered. In its filing, Anthropic explains that multi-year commitments are typically made 12 to 18 months before resources become available. Expenses then appear as that capacity is brought online, while the revenue it enables may arrive later.
Three Loss Figures, Three Different Readings
The headline “$17 billion in revenue, $23.3 billion net loss” is accurate, but it does not fully describe the business. The Associated Press reports a quarterly net loss of $23.3 billion. The SEC filing, meanwhile, specifies an operating loss of $567 million for the quarter, after Anthropic posted $974 million in operating income in the previous quarter. The company primarily attributes the decline to $5 billion in short-term computing capacity, which costs more than its multi-year contracts, and $2 billion in non-cash charges related to its donation matching program.
The gap between operating and net loss also reflects financial items and fair-value changes in certain financing instruments. For the first nine months, AP reports a $104 billion net loss, including $94 billion in accounting adjustments associated with financing agreements. Those amounts should not be confused with cash outflows of the same size. They do not make the risks insignificant: they show how financing structure and capital requirements weigh on the financial profile of a hypergrowth AI company.
The Real Bet: Reserving Tomorrow’s Capacity Today
Anthropic says computing power is essential to its expansion and that it is diversifying how it obtains capacity: cloud contracts, direct chip leases and capacity in dedicated data centers. This strategy can secure scarce resources and potentially offer more control over long-term costs. It also carries commitment risk: if demand slows, models become more efficient or new hardware changes the economics of computing, some reserved capacity could prove less useful or too expensive.
The company itself acknowledges that periods when new computing capacity is delivered can compress margins, because the costs are recognized before all related revenue is realized. In the third quarter, Anthropic reported $2.5 billion in adjusted operating income, but an adjusted margin of 15%, down from 19% in the previous quarter, according to the filing. This is more nuanced than a simple contrast between “success” and “losses”: commercial activity is expanding, but profitability remains sensitive to timing, price and actual use of computing resources.
What Investors Will Need to Examine
The IPO filing amendment provides new financial information without announcing how much the company plans to raise or the price of its shares. It also reveals how difficult it is to value a company whose scale depends on very long contracts with computing providers. Investors will need to distinguish recurring revenue from one-off growth, operating losses from accounting effects, and contractual commitments from expenses already paid.
Another question concerns infrastructure concentration. These commitments support the capacity needed for research and commercial services, but they also increase the importance of partners supplying chips, cloud services, energy and buildings. Bargaining power, delivery schedules and the ability to switch providers thus become central economic factors, alongside model quality.
The Test After the Euphoria
Publishing these figures proves neither that Anthropic’s bet will fail nor that demand will automatically justify every contract. It does, however, make the debate more concrete. The question is not only whether Claude attracts customers, but whether billable usage can grow fast enough, at sufficient margins, to finance a physical expansion committed over decades.
Anthropic has shown that an AI research lab can become a company with substantial revenue within a few years. Its filing also shows that the industry remains subject to familiar economic constraints: costs, contracts, financing, asset utilization and the risk of excess capacity. For prospective shareholders, the $17 billion figure is evidence of traction. The $895.4 billion in commitments is a question to watch: how much of this capacity will be used, at what price, and with what lasting value for customers?