Competing with OpenAI is a major challenge for Anthropic. But the company behind Claude faces another problem that could prove just as important: the enormous cost of building and running artificial intelligence.
Anthropic’s latest IPO filing puts that challenge into perspective. The company has entered into purchase commitments and lease agreements worth approximately $900.9 billion, substantially all related to computing power and infrastructure. These commitments stretch across several years, with much of the spending expected between 2026 and 2045.
The figure is not a bill due immediately. But it shows how much infrastructure Anthropic believes it needs to keep developing models and serving customers.
Why does running AI cost so much?
Every time someone asks Claude a question, the company needs computing resources to process it. This is known as inference. Training more powerful models requires even more computing capacity. That means spending on advanced chips, servers, data centres, networking, electricity and cooling. Anthropic uses a mix of Amazon's Trainium chips, Google's TPUs, and GPUs from Nvidia and AMD. It says this multi-chip approach helps it balance performance and cost.
The company expects around $70.7 billion of its disclosed commitments to fall due in the 12 months after September 30, 2026. Another $451.6 billion is expected before the end of 2030.
Trending Stories
Revenue is rising, but profitability remains a test
Anthropic's growth is substantial. It reported $17 billion in revenue for the third quarter of 2026, up from $1.4 billion a year earlier.
However, it also recorded an operating loss of $567 million in the same quarter. The company said $5 billion of expensive, short-term compute and $2 billion in non-cash donation-matching charges contributed to the result. Its adjusted operating income was $2.5 billion, a separate non-GAAP measure.
Reuters reported that Anthropic's annualised revenue run rate reached $81 billion in October. That is a projection based on a recent revenue pace, not the same as revenue already earned over a full year.
How Anthropic compares with OpenAI and Google
OpenAI faces similar infrastructure pressure. Reuters reported its September annualised revenue run rate was nearly $50 billion, although comparisons with Anthropic are complicated by differences in how partner-cloud revenue is counted.
Google has a different advantage: it owns a large technology and cloud business, develops its own TPUs and earns money from advertising, subscriptions and enterprise services. Alphabet forecast capital expenditure of $195 billion to $205 billion for 2026, largely driven by infrastructure investment.
2026 Q3 Earnings Call
The Federal Reserve has also highlighted the economics of AI inference. It cited estimates from the companies that gross margins on inference are around 30-40 per cent, leaving a large share of revenue to cover computing costs alone.
Anthropic's central test is therefore clear: can it turn fast-growing demand for Claude into enough recurring revenue to pay for chips, data centres and power over decades? The competition with ChatGPT matters. But the cost of keeping AI running may be the bigger long-term business challenge.

&imwidth=896&imheight=504&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&im=FitAndFill=(700,400)&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)
&imwidth=160&imheight=90&format=webp&quality=medium)