The math that makes this irresistible: roughly $50B of annual revenue per gigawatt at SpaceX-style pricing, so renting out just 200MW throws off more than $10B a year at very high margin.
SpaceX's Anthropic and Google compute contracts earn three to four times the revenue per megawatt that peers charge. Cost structure is roughly the same, so the profit gap per megawatt is even wider than that.
The Google deal is nominally three years but either side can cancel on 90 days' notice — effectively a rolling three-month rental. Neoclouds cannot play there because financing a large cluster requires locking in a multi-year offtaker.
Cancellable contracts are exactly why this suits Meta: if Superintelligence Labs needs the GPUs back, it takes them back in 90 days. Its fast-to-erect 'tent' datacenters fit the same logic — lower quality, online early, monetized immediately.
After roughly 10GW of deals signed since early 2024, most of Meta's incremental capacity now comes through third parties rather than self-build. That makes Meta a source of contracted backlog growth for CoreWeave and Nebius, not a competitor crushing them.
Why it matters
Gives engineers building on hyperscaler compute or model APIs concrete signal on where GPU supply, pricing, and access (e.g., potential Anthropic resale via Meta) may shift in the near term.
Key quotes
“Everything is computer and everything is a neocloud.”
articlexAI’s Colossus 2 – First Gigawatt Datacenter In The World, Unique RL Methodology, Capital Raise
articleAmazon’s AI Resurgence: AWS & Anthropic’s Multi-Gigawatt Trainium Expansion
articleSpaceX 10GW in 2027 – Why It’s Real, Will Drive $300B ARR for SpaceX, and Why Microsoft Will Be the Largest Offtaker
