Anthropic, a leading rival to OpenAI, has finally opened its books. The company's public IPO filing on October 26, 2026, provides the first audited look at its finances, confirming the massive cost of building frontier AI models.
Explosive Revenue Meets Eye-Watering Costs
The S-1 filing reveals
a company in hyper-growth. Anthropic reported revenue of $16.3 billion for the first half of 2026, a staggering 14-fold increase from the $1.2 billion in the same period a year prior. Full-year revenue for 2025 was approximately $4.6 billion. However, this explosive growth comes with monumental expenses. The headline figure that caught many off guard was a net loss of $80.7 billion for the first six months of 2026. While this number is alarming, the filing clarifies that the vast majority, about $72.4 billion, is a non-cash charge related to the accounting of convertible notes—a side effect of the company's valuation soaring. The actual operating loss, which reflects the day-to-day business, was a more manageable $693 million for that period. In fact, the company even posted a brief operating profit of nearly $1 billion in the second quarter of 2026 before swinging back to a loss in the third quarter due to rising costs.
The Billion-Dollar Compute Bill
The single greatest cost of growth for Anthropic is the unquenchable thirst for computing power. Training and running large language models like Claude requires immense computational resources, and the IPO prospectus lays these costs bare. The company has entered into massive, long-term commitments with cloud providers who are also its investors, including Amazon, Google, and Microsoft. The filing details future compute and infrastructure commitments totaling hundreds of billions of dollars. For the first half of 2026, compute and infrastructure expenses amounted to approximately $9 billion. In the third quarter of 2026 alone, these costs hit $9.9 billion as the company relied on more expensive short-term capacity to meet demand, swinging it from an operating profit back to a loss. These figures underscore that success in the AI race is not just about having the best algorithm, but also about having the deepest pockets to pay for the underlying hardware.
The War for Human Talent
While hardware is a dominant expense, the cost of human genius is another critical factor. The AI industry is defined by a fierce war for a very small pool of elite researchers and engineers. Anthropic, founded by former OpenAI research leads, has had to compete aggressively for this talent. IPO filings typically reveal significant spending on stock-based compensation and high salaries to attract and retain top minds. This is a non-negotiable growth cost, as the expertise of these individuals is what drives the innovation that gives a company like Anthropic its competitive edge against rivals and a burgeoning open-source community. The S-1 filing notes that as of September 2026, the company had over 11,900 customers paying at least $100,000 annually, demonstrating strong enterprise adoption that relies on a highly skilled workforce to support it.
A Business Model Under the Microscope
Anthropic's prospectus clarifies its business model, which is heavily reliant on enterprise customers. Approximately 80% of its revenue in the first half of 2026 came from consumption-based API usage, with the other 20% from subscriptions. This differs from OpenAI, which has a more diversified mix including a massive consumer-facing application in ChatGPT. The concentration on enterprise clients and API access creates a 'sticky' customer base, but also carries risks. The filing notes that a significant portion of revenue comes from a limited number of customers and distribution channels like the marketplaces of its cloud partners, Amazon and Google. Many of these customers are not on long term contracts, meaning they can reduce their spending quickly, while Anthropic's massive compute costs remain locked in for years.
