
Meta’s reportedly in early talks to lease up to $10 billion of computing power to Anthropic over two years. The negotiations are fluid. Either side can walk away.
I’ve learned to pay less attention to the deal itself and more to what it reveals. This one reveals a structural shift in who controls compute distribution.
The Signal Was Already Public
At Meta’s shareholder meeting in May, Mark Zuckerberg said entering cloud computing was definitely on the table. Companies were approaching Meta nearly every week, asking to buy spare capacity at a premium.
That detail matters. Buyers offered to pay above Meta’s own cost. When your excess inventory commands a premium, you’ve got a product.
Meta raised its 2026 capital expenditure guidance to $125 to $145 billion, an 87% jump from the $72.2 billion spent in 2025. Investors flinched. Shares dropped roughly 10%, erasing about $150 billion in market value.
The market read that spending as risk. I read it as inventory for a second business.

Anthropic’s Side Tells You More
Anthropic is closing in on a $1 trillion valuation and still can’t build capacity fast enough. The company places usage limits on its most advanced models because chip access is the bottleneck.
Its revenue growth makes the squeeze visible. Annual recurring revenue hit $47 billion by mid-2026, up from $5 billion in early 2025, with Claude Code driving roughly a third of it.
And Anthropic’s paying $1.25 billion per month for SpaceX data center capacity. That’s $15 billion a year on compute alone.
Compute access now sets competitive velocity. Model architecture comes second.
What This Means for Your Next Twelve Months
The five largest U.S. infrastructure providers have committed $660 to $690 billion for 2026, nearly double 2025 levels. Every one of them reports supply constraints, and demand keeps outrunning capacity.
💡 This is the part that’s commonly overlooked. Meta buys capacity from Nebius under a $27 billion agreement while preparing to sell capacity to Anthropic. One company, positioned on both sides of the same marketplace. That dual position creates pricing power that pure providers can’t replicate.
If you’re building on AI infrastructure, your planning question shifts. Reliable capacity at scale now determines what you’re able to ship. A slightly smarter model that times out under load helps you less than a capable model that answers ten thousand queries an hour.
The Recap
Three things worth holding onto. Meta is turning capital expenditure into a revenue line. Anthropic’s growth proves compute cost is the binding constraint on AI margins. Infrastructure position now shapes competitive outcomes across the entire sector.
Watch where the capacity contracts flow. The repricing follows.