Anthropic’s revenue surged as demand for Claude and AI services expanded rapidly.
Compute and infrastructure expenses account for a significant share of operating costs.
Massive infrastructure commitments highlight the capital-intensive economics of frontier artificial intelligence development.
Anthropic’s rapid growth offers a closer look at the economics of building and operating frontier artificial intelligence systems. The company is generating billions of dollars in revenue, but its expansion also requires enormous spending on computing infrastructure, model development and cloud capacity.
The numbers disclosed in its latest financial filings underline the central challenge facing AI companies: revenue is rising quickly, but the cost of staying at the technological frontier is rising alongside it.
Anthropic generated USD 4.59 billion in revenue in 2025, up from USD 386 million a year earlier. The company also reported an operating loss of USD 8.06 billion for the year.
Its reported GAAP net loss was about USD 41.97 billion, compared with USD 8.31 billion in 2024. However, around USD 34 billion of the 2025 net loss came from an accounting charge largely linked to financing instruments that could convert into Anthropic shares. The operating loss therefore provides a clearer picture of the company’s underlying business performance.
Anthropic's annualized revenue run rate has continued to rise in 2026. In April, the company said its annualized revenue had crossed USD 30 billion, up from about USD 9 billion at the end of 2025. In May, Anthropic said its run-rate revenue had crossed USD 47 billion
Building frontier AI models requires vast amounts of computing power. According to figures cited from Anthropic’s disclosures, the company spent USD 7.33 billion on compute and infrastructure in 2025, nearly three times the previous year’s spending.
That spending represented about 58% of Anthropic’s USD 12.65 billion in total operating expenses. In simple terms, the company spent roughly USD 1.60 on compute and infrastructure for every USD 1 of revenue generated during the year.
The economics are different from traditional software businesses. AI models require substantial computing resources not only during training but also when customers deploy and use them. As usage increases, companies need additional infrastructure to process requests and maintain performance.
Research into frontier AI costs also points to the same trend. A study on the rising costs of training frontier models found that the amortized cost of the most compute-intensive training runs increased by about 2.4 times per year from 2016 onwards.
Also Read: OpenAI and Anthropic Under FTC Lens as AI Agent Risks Raise Questions
Anthropic has continued securing computing capacity as demand for Claude grows. In April, the company announced an agreement with Amazon covering up to 5 gigawatts of new capacity and said it would commit more than USD 100 billion over 10 years to AWS technologies. Amazon also announced a USD 5 billion investment in Anthropic, with the potential for another USD 20 billion.
Anthropic has also expanded its relationships with Google and Broadcom for next-generation TPU capacity. The company said it had agreements covering five gigawatts of capacity with Google and Broadcom, and secured access to GPU capacity from SpaceX.
Its long-term infrastructure commitments show how capital-intensive the frontier AI business has become. Reuters reported that Anthropic’s commitments had exceeded USD 518 billion by early 2026, covering 3.5 gigawatts of dedicated computing capacity.
Anthropic’s business model centers on Claude, with revenue coming from customers paying to use the company’s AI systems. In 2025, about USD 3.8 billion of its revenue came from consumption-based payments, while subscription revenue stood at USD 789 million.
The company expects consumption-based revenue to remain the substantial majority of its revenue for the foreseeable future.
Claude is also distributed through major cloud platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic says this approach lets it reach businesses already using those platforms without building an equivalent sales and distribution network.
Also Read: Anthropic IPO Filing Reveals USD 518 Billion AI Spending Plan
Anthropic’s financial picture highlights a fundamental feature of frontier AI. A model can generate significant revenue once deployed, while the company simultaneously spends heavily to develop and train the next generation.
The result can be substantial operating losses even as individual products generate growing commercial demand.
For Anthropic, the long-term business question is whether revenue growth and increasing AI adoption can eventually outpace the cost of training models, serving users, and securing the infrastructure required to remain at the frontier.
What is Anthropic’s AI business model?
Anthropic primarily earns revenue through Claude, with customers paying subscription and consumption-based fees for AI services.
How much revenue did Anthropic generate in 2025?
Anthropic generated USD 4.59 billion in revenue in 2025, up significantly from USD 386 million recorded during 2024.
Why is frontier AI so expensive to build?
Frontier AI requires enormous computing resources, infrastructure, model development spending, and ongoing capacity to serve growing user demand.
How much did Anthropic spend on compute and infrastructure?
Anthropic spent USD 7.33 billion on compute and infrastructure in 2025, representing approximately 58% of its operating expenses.
How does Anthropic distribute Claude to customers?
Anthropic distributes Claude through platforms including Amazon Web Services, Google Cloud, and Microsoft Azure, expanding enterprise access globally.