SemiAnalysis: Meta to Accelerate Compute Power Procurement Instead of Slowing Down, Actively Engaged in Negotiations with Anthropic to Build Proprietary AI Model Service Platform

BlockBeats News, July 3rd, SemiAnalysis stated in its latest report that after the news broke that Meta may become a new Neocloud, the market's initial reaction was to sell off CoreWeave, Nebius, and other such cloud computing companies, and once again worry about "AI hash rate oversaturation."

However, the firm's assessment is quite the opposite: this worry may be unfounded. Meta's data center and computing power procurement will not slow down, but will instead continue to accelerate. The article mentioned that in just the first half of this year, Meta has already secured over 5GW of capacity in the cloud services and colocation data center space, and this doesn't even account for its self-built projects that are rapidly progressing.

SemiAnalysis stated that Meta is in final negotiations with Anthropic, hoping to gain access to Claude's privatized instance. If this materializes, the significance would not just be that "Meta has acquired more computing power," but that Meta may be in the process of constructing its own AI model services platform.

This model is somewhat similar to AWS's Bedrock, Microsoft's Foundry, and Google's Vertex. Meta could initially use Claude internally and in the future package the modeling capability as a token-as-a-service to provide external services. In the short term, it may be internal use of proprietary models and internal use of Anthropic models; in the long term, Meta may even incorporate Anthropic and OpenAI models into its own external service ecosystem.

SemiAnalysis stated that the rationale behind this is that Meta has computing power, advertising clients, social network distribution capabilities, and consumer entry points. If it can integrate cutting-edge models, intelligent agents, and sales and marketing SaaS, it will no longer just be a GPU buyer but will be moving toward AI applications and model distribution.