Meta just agreed to spend $27 billion renting someone else's servers while simultaneously spending $125 billion building its own.
The Signal
Nebius, a Dutch data center operator you've never heard of, saw its stock jump 15% Monday after Meta signed a multi-year deal worth up to $27 billion for cloud capacity. This isn't some strategic partnership. This is Meta admitting it can't build fast enough to keep up with its own AI ambitions.
Meta's spending $125 billion this year on capex. That's more than the GDP of Ukraine. They're building data centers at a pace that would make any CFO sweat. And they're still running out of room. The gap between "how much compute we need" and "how much compute we can build" has gotten so wide that throwing tens of billions at third-party providers is the only play.
This is what demand destruction looks like in reverse. The AI compute bottleneck isn't easing. It's tightening. And it's creating a market for anyone who can stand up GPUs fast enough. Nebius isn't special. They're just there, with capacity, at a moment when Meta needs it more than it needs efficiency.
Jensen Huang spent Monday at GTC talking about Nvidia suppliers enjoying "booming business." He wasn't exaggerating. The whole supply chain is printing money because companies like Meta have moved past "let's optimize costs" into "get us capacity or we fall behind."
The Implication
Watch the second-tier infrastructure plays. If Meta is this desperate, everyone else is too. The companies winning right now aren't the ones with the best technology. They're the ones with available racks. That gap won't last forever, but while it does, expect more deals like this and more stock pops for operators you've never heard of.
Source: The Information