TL;DR

Payment Terms: Anthropic may pay xAI $1.25 billion a month for compute through May 2029. Capacity Lockup: The agreement covers all of Colossus 1 and more than 300 megawatts backed by over 220,000 Nvidia GPUs. Claude Impact: Anthropic says the added supply supports higher Claude Code and API limits. Contract Exit: Either side could end the compute contract with 90 days’ notice.

Anthropic appears set to pay $1.25 billion per month for compute through May 2029, giving the AI market one of its clearest public price signals yet for frontier-scale capacity. The details emerged in SpaceX’s S-1 filing with the SEC. At that run rate, the contract works out to about $15 billion per year. Terms at that level could nearly double SpaceX’s annual revenue.

Instead of adding compute in smaller cloud increments, Anthropic is tying Claude’s next growth phase to a dedicated supply lane with a defined price, term, and exit clause. A practical consequence becomes clear already: more supply is being connected to higher usage ceilings for Claude products.

What the Contract Buys

Anthropic agreed to buy 300 megawatts of compute, which puts the arrangement closer to a utility-scale infrastructure commitment than a routine software expansion. Colossus 1 would supply all of the compute capacity at the site near Memphis. Reserving the output of a named facility gives the company a fixed pool of power and hardware rather than overflow access on ordinary cloud terms.

Anthropic’s package adds more than 300 megawatts of new capacity and over 220,000 Nvidia GPUs. That means enough hardware to support far larger training and inference workloads across several Claude products at once. Brown, an Anthropic executive in an embedded social post, tied the next stage of the arrangement to Colossus 2 capacity, which keeps the expansion lane open without confirming that a second phase is locked in on the same terms.

Either side can terminate the compute contract with 90 days’ notice. A discounted rate for the first two months also suggests the infrastructure buildout still needs a ramp period before the full monthly bill applies. Together, those terms make the arrangement look more like a large supply purchase with defined checkpoints than an open-ended strategic partnership.

Why Anthropic Wants the Capacity

Anthropic links the added supply to higher Claude Code and API limits. More headroom lets the company raise usage ceilings without pushing the same infrastructure bottlenecks across every customer tier at once. More available compute is the direct bridge between a giant procurement number and a visible product effect.

Developers are already seeing part of that effect in doubled five-hour Claude Code rate limits. More throughput should also help enterprise and API customers whose workloads depend on steady model availability instead of short bursts of access. Claude Pro and Claude Max users stand to benefit if the added supply reduces queueing pressure at the same time Anthropic raises limits.

SpaceX’s IPO filing adds the other half of the rationale: the arrangement would monetize unused compute capacity that might otherwise sit below peak utilization. Capacity sales of that kind can turn expensive infrastructure into a revenue line even when the owner is not using every GPU for its own models at full intensity. That revenue logic helps explain why an xAI-linked site would reserve such a large block for an outside customer.

Supplier Strategy and Market Context

Anthropic’s recent compute deals with Amazon, Google, Microsoft, and SpaceX point to a broader effort to lock in enough infrastructure for Claude’s continued growth. Spreading supply across several partners gives the company more protection if one vendor hits deployment delays, procurement bottlenecks, or local power limits. Multiple supplier relationships also reduce the risk that one constrained facility becomes a single choke point for model launches or enterprise demand.

Relative to xAI’s wider footprint, the Colossus transfer also looks large. The arrangement could consume half its GPU fleet, putting xAI’s total around 500,000 GPUs. Capacity on that scale supports the idea that this is a material reservation, not a peripheral slice of infrastructure.

Memphis still adds a constraint to watch. In January 2026, an EPA ruling on xAI’s Memphis data centers showed that large-scale expansion can draw operational and regulatory scrutiny alongside the commercial upside. A later Anthropic or SpaceX filing could clarify whether Colossus 2 capacity would follow the same pricing and duration or move onto a different contract path.