Particle.news

Meta Developing 'Meta Compute' to Sell AI Compute and Hosted Models

The plan would convert excess data‑center capacity into a new revenue stream that risks squeezing specialized cloud rivals.

Overview

  • Bloomberg reported Wednesday that Meta is quietly building a cloud unit reportedly called Meta Compute to offer hosted AI models and raw GPU compute to third parties, and Meta has not confirmed product details, pricing, or a launch date.
  • News of the plan drove a sharp market move with Meta shares rising roughly 6% in premarket trading and opening higher in some reports while neocloud names such as CoreWeave and Nebius fell roughly 9 to 13 percent.
  • Meta’s decision follows a steep increase in AI capital spending for 2026, with company guidance now at $125 billion to $145 billion, and comes as Meta shifts budget away from Reality Labs and trims headcount to free funds for infrastructure.
  • If rolled out, Meta Compute would pit the company directly against AWS, Microsoft Azure, and Google Cloud but would likely start by selling specialized AI compute services rather than the full cloud software stack, and enterprise trust and privacy concerns could limit adoption.
  • The move builds on Meta’s existing cloud ties — including a 2025 multi‑year Google Cloud deal and expanded CoreWeave commitments — and would turn part of its massive infrastructure build into an operational revenue option rather than reliance on new equity financing.