Meta just told AWS it's not renting compute anymore — it's building a landlord empire.

The Summary

The Signal

Meta recruited Dave Brown, the AWS executive who ran EC2 — the bedrock compute service that powers half the internet. This isn't a lateral hire. Brown spent years building the infrastructure layer that OpenAI, Anthropic, and hundreds of AI labs depend on. Now he's building Meta's answer to it.

The new division, Meta Compute, signals a shift from customer to competitor. Meta already designs its own AI chips. It already runs one of the world's largest private data center networks. Now it's packaging that capability into a cloud platform that could rent compute to anyone building agents, training models, or running inference at scale.

"Meta is moving from renting compute to selling it — a vertical integration play disguised as infrastructure expansion."

The numbers matter here. Initial reports suggest $50B+ in investment, but other sources cite up to $145B earmarked for AI infrastructure. Either figure dwarfs what most cloud providers spend annually. For context:

  • AWS capex in 2023: ~$50B
  • Google Cloud capex in 2023: ~$32B
  • Meta's reported commitment: $50B-$145B over an unspecified period

This isn't just data center expansion. It's cloud ambition. Brown doesn't join a company to manage internal IT. He joins to build products. And the only product that makes sense at this scale is a public-facing compute platform that competes directly with his former employer.

Meta already open-sourced Llama, creating a moat of developer goodwill. It already has enterprise relationships through WhatsApp Business and Instagram. It already runs inference at a scale most startups will never touch. The missing piece was credibility as infrastructure provider. Brown brings that.

The Implication

If Meta Compute launches as a full cloud platform, the agent economy just got a major new infrastructure option. Developers building on AWS or Azure suddenly have leverage. Pricing pressure intensifies. And every AI lab currently renting GPUs from hyperscalers now has to recalculate their vendor risk.

Watch for Meta to bundle compute credits with Llama licenses, or offer inference-as-a-service at cost to undercut OpenAI's API. Brown didn't leave AWS to build a better internal IT department. He left to build the cloud that trains and runs the next million agents.

Sources

Crypto Briefing