The economics of frontier AI inference start with one hard constraint: access to data centre capacity at a scale that no single operator can provision quickly. Meta, which has committed $145 billion to infrastructure spending, is in talks with Anthropic for a data centre arrangement worth up to $10 billion. That conversation is running alongside Meta's reported consideration of entering the cloud business itself.
Where this sits in the compute stack
The mechanism behind that is straightforward. Training and serving large language models requires power, cooling, and networking density that took hyperscalers years to build. Anthropic needs that capacity in volume, and Meta is building considerably more of it.
A deal of this size would move Anthropic from buying capacity piecemeal to a larger, more stable supply arrangement. The specific unit that drives the economics here is contracted capacity at known power and cooling specifications. That matters because it determines how a lab plans training runs and inference throughput months in advance, not quarter to quarter.
Meta's cloud ambitions
The constraint here is that Meta has historically built infrastructure to serve its own products and internal AI operations. Selling capacity to third parties is a different business with different margin structures, compliance obligations, and customer relationships. The organizational muscle is not the same.
Entering cloud services while simultaneously committing $145 billion to infrastructure is an aggressive posture. The talks with Anthropic, if they close at or near the reported ceiling, would represent one of the largest early customer commitments a nascent cloud operation could bring to market.
What the deal structure means for Anthropic
For Anthropic, the relevance is capacity and cash runway. Large data centre contracts typically involve committed spend over multi-year terms. The terms of the reported talks are not publicly detailed beyond the $10 billion figure, so the structure, duration, and any equity components remain unknown.
The size alone is material. Anthropic competes for both talent and training capacity across a field of well-capitalized rivals. A long-term infrastructure arrangement with a provider building toward a cloud business creates a different supply-chain posture than relying on existing hyperscaler allocations.
Neither company has confirmed the talks or their terms publicly. The $145 billion infrastructure commitment from Meta and the up to $10 billion figure attached to the Anthropic discussions are the specific numbers in play.