Anthropic signed a five-year agreement to purchase computing capacity from xAI at $1.25 billion per month, committing $15 billion through 2029 in what marks the largest disclosed AI infrastructure contract on record. The deal gives Anthropic guaranteed access to xAI's compute clusters while Elon Musk's venture monetizes hardware investments made over the past eighteen months.

The structure is monthly payments against reserved capacity, not usage-based cloud pricing. Anthropic secures predictable compute availability for Claude model training and inference workloads. xAI converts its Memphis data center build-out—reportedly 100,000 H100 GPUs online as of Q4 2024—into contracted revenue before those chips depreciate. Neither party disclosed uptime guarantees or penalty clauses for capacity shortfalls, though contracts of this scale typically include quarterly reconciliation windows.

This matters because foundation model economics now hinge on secured compute, not algorithmic efficiency alone. Training runs for frontier models require coordinated access to tens of thousands of accelerators for weeks at a time. Spot capacity and interruptible cloud allocations no longer suffice at GPT-4 scale and beyond. Anthropic's move follows similar patterns: OpenAI's $10 billion Microsoft Azure commitment, Google's multi-year TPU agreements with DeepMind. The difference here is cross-company contracting—Anthropic buying from a direct competitor rather than a cloud hyperscaler. That signals two shifts. First, xAI is positioning as infrastructure landlord, not just model developer. Second, the compute constraint is severe enough that Anthropic accepts vendor risk with Musk-controlled hardware to guarantee training windows for Claude 4 and successors.

The deal also clarifies xAI's business model pivot. Musk's venture raised $6 billion in May 2024 at a $24 billion valuation, primarily to build out compute infrastructure in Memphis. Converting that capital into a contracted revenue stream—$15 billion over sixty months—derisks the hardware bet and provides cash flow to fund Grok model development in parallel. For Anthropic, the cost is steep but the alternative is worse: training delays that let OpenAI, Google, and Meta extend leads in multimodal capabilities and enterprise deployments. Compute access is the choke point. Anthropic is paying to remove it.

Allocators should track three follow-ons. First, whether Anthropic's next funding round—expected in Q2 2025 after its $7.3 billion Series D in March 2024—prices in this $15 billion obligation. The company is now carrying a material off-balance-sheet compute liability that affects cash flow projections through 2029. Second, xAI's capacity utilization. If Anthropic's reserved allocation sits at 70-80% utilization, xAI has room to sign secondary contracts with other buyers, layering revenue on the same hardware. Third, regulatory attention. A deal this large between two frontier AI labs may draw antitrust review in the EU and UK, particularly if it forecloses compute access for smaller model developers.

The contract starts in Q2 2025, with first payments due in May. Anthropic's Memphis workloads go live sixty days after that.