Last week’s Bloomberg scoop sent a clear signal: Meta is finally turning its AI infrastructure into a revenue stream. The stock jumped over 10% in a single session, adding nearly $100 billion in market cap. Meanwhile, pure‑play GPU rental shops like CoreWeave and Nebius dipped. The market got the message.
The new unit, reportedly called “Meta Compute,” will let external developers rent spare GPU capacity and access Meta’s in‑house models. Think AWS Bedrock, but with a social‑media giant’s twist.
Why now? The spending numbers are staggering. Meta’s 2025 capex hit $72.2 billion. For 2026, guidance is $125‑145 billion. And as of Q1 this year, the company had committed over $182 billion to AI infrastructure over the next few years. That’s a lot of silicon sitting in data centres.
Unlike Google or Microsoft, Meta has no legacy cloud business to absorb those costs. Until now, there was no clear line item for AI revenue – investors kept asking “where’s the payback?” Renting out unused compute is the most direct answer.
How will it work? Two tracks:
- Model‑as‑a‑service – access to Meta’s Muse Spark and other models, pay‑per‑use.
- Raw compute – bare‑metal GPU hours, competing directly with CoreWeave and Lambda Labs.
Zuckerberg actually telegraphed this back in May. At a shareholder meeting, he said: “Almost every week, external companies reach out wanting to buy our compute – some even offer a premium.” Now the infrastructure is ready.
A neat comparison: Elon did it first. xAI started renting excess capacity from its Memphis data centre to Anthropic and Google – at monthly rates of $1.25 billion and $920 million respectively. That works out to roughly $26 billion annualised. A single under‑utilised cluster turned into the world’s largest compute‑rental business overnight. Meta would be silly not to copy the playbook.
What this means for you – if you’re an enterprise, you’ll soon have one more hyperscaler to choose from, potentially with competitive pricing. If you’re an investor, it gives Meta a credible monetisation story for its AI spend. The caveat? AWS, Azure and GCP spent decades building their ecosystems. Meta is starting from zero in enterprise sales – that won’t be easy.
Tags:
Technology