SemiAnalysis: Meta will accelerate computing power procurement rather than slow down, and it is negotiating with Anthropic to build its own AI model service platform.

Mars Finance News, July 3 — In its latest report, SemiAnalysis said that after news emerged suggesting Meta could become a new Neocloud, the market’s first reaction was to sell off compute-cloud companies such as CoreWeave and Nebius, and to worry again about “AI compute power oversupply.” But the institution’s assessment is exactly the opposite: this concern may be wrong. Meta’s data centers and its procurement of computing power will not slow down; they will instead continue to accelerate.

The article notes that in just the first half of this year, Meta has already signed deals for more than 5GW of capacity in cloud services and managed data centers—and this figure does not even include the self-built projects it is accelerating. SemiAnalysis says Meta is in final negotiations with Anthropic, hoping to gain access to private instances of Claude. If this comes to pass, the significance is not only that “Meta is buying more compute,” but that Meta may be building its own AI model services platform.

This approach is somewhat similar to AWS’s Bedrock, Microsoft’s Foundry, and Google’s Vertex. Meta could use Claude internally first, and later package the model capabilities into a token-as-a-service offering to provide services externally. In the short term, it may use its own models externally and Anthropic’s models internally; in the long term, Meta could even incorporate Anthropic and OpenAI’s models into its own external service ecosystem.

SemiAnalysis said that the underlying logic is that Meta has computing power, advertising customers, social-network distribution capabilities, and entry points at the consumer end. If it can combine frontier models, agents, and sales-and-marketing SaaS, it would not be just a company that buys GPUs—it would be moving up the stack toward AI applications and model distribution.