Meta Compute: Everyone Wants To Be A Neocloud Meta’s aggressive expansion into the neocloud market has sparked intense speculation, with analysts and investors debating the company’s strategic moves. Despite initial market reactions that suggested a slowdown in datacenter investments, the reality is that Meta’s procurement of compute resources is accelerating. The company has already contracted over 5GW of capacity across cloud and colocation services in the first half of 2024, with self-built datacenters adding to this figure. Analysts at SemiAnalysis argue that the notion of “half of US datacenters being delayed” is misleading, as just two of Meta’s datacenters represent more than half of the current construction pipeline. The core of Meta’s strategy lies in its ability to monetize vast compute resources through multiple high-value use cases. One key area is scaling its recommendation systems (RecSys) by more than 10x in complexity to drive revenue growth. This requires both inference and training compute, enabling more sophisticated generative advertising. Additionally, Meta is reportedly in final talks with Anthropic to access private instances of Claude, akin to Amazon’s Bedrock or Google’s Vertex. This partnership could power internal tools and external SaaS platforms leveraging Frontier AI agents, with plans to launch a token-as-a-service endpoint. Meta’s ambitions extend beyond internal use. The company is positioning itself to replicate Elon Musk’s “SpaceX-type” deals, where large-scale on-demand compute is sold at premium pricing. A potential $10B deal with Anthropic could kickstart a flywheel effect, leveraging Meta’s network to scale compute offerings.#meta #anthropic #coreweave #semi_analysis #neocloud
