Anthropic expects to have more than 3.5 gigawatts of computing capacity by the end of 2026, a sign that infrastructure is becoming a critical front in the AI race.

The added capacity is meant to support more advanced Claude models and keep pace with rising demand from businesses and individual users, including for Claude Code.

To secure that capacity, Anthropic is pursuing agreements with Amazon, Google, Microsoft and SpaceX, while exploring smaller providers and alternatives to Nvidia chips to diversify supply.

The expansion addresses capacity constraints Anthropic has faced as demand has grown, but scaling data centres also requires land, electricity, cooling systems and specialist workers—not just chips.

Anthropic’s prospectus reportedly lists about $895.4 billion in commitments through 2045 for computing power and chip-related agreements, highlighting the scale of its long-term infrastructure costs.

Ahead of an expected IPO, Anthropic’s disclosures show $17 billion in revenue and a $23.3 billion net loss in the third quarter of 2026, leaving investors to assess whether future revenue can justify those commitments.

The broader takeaway is that the AI race increasingly hinges not only on model performance but on dependable access to chips, power and data-centre capacity.