The commitments are the strategy

Anthropic has contracted capacity from four providers who compete with each other and, in several cases, invest in it. Depending on all of them is the point.

More than $100bn to AWS over ten years for up to 5GW, a further 5GW with Google and Broadcom, $30bn of Azure capacity, $50bn of US infrastructure via Fluidstack — plus the whole of SpaceX's Colossus 1 in Memphis.

Exclusivity is the thing being avoided

A single provider holding all your capacity holds your roadmap. Four providers holding pieces of it hold nothing. The redundancy costs negotiating overhead and buys the ability to walk, which is the only leverage a compute-constrained lab has.

The mix is deliberate too: rented cloud, a custom-silicon partnership, and $50bn of buildout that is ownership rather than rental. Different failure modes, different time horizons.

The number that makes it a bet

Run-rate revenue around $30bn, up from roughly $9bn at the end of 2025. Tripling in eight months is extraordinary, and these commitments assume it continues at a rate essentially no company has sustained.

That is the risk, and it is not hidden. It is the same bet the whole sector is making, sized to one balance sheet.

Which is why efficiency is worth $6bn

Buying a team that improves chip utilisation reads differently once you know capacity is contracted years ahead. Efficiency gains do not reduce the bill — they turn capacity you have already bought into work you can sell.

Meanwhile approved H200 shipments to China are running at trivial volumes. Capacity is being locked up in ten-year contracts on one side of the world and substituted around on the other. Both are the same story: compute is the constraint, and everyone is solving for it structurally rather than by buying chips.

Anthropic — Anthropic and Amazon expand collaboration for up to 5 gigawatts of new compute → · Anthropic — Anthropic expands partnership with Google and Broadcom → · Bloomberg — Small Amount of Nvidia AI Chips Sold to China Via US License →