Meta has signed contracts with infrastructure providers CoreWeave and Nebius worth approximately $62 billion in total—and now plans to resell excess AI computing capacity as a cloud service. Reuters, citing Bloomberg News, reports the move, though Meta has not officially confirmed a commercial cloud product. The market reacted sharply: CoreWeave stock fell 10.8 percent, Nebius dropped 12.4 percent.
The essentials
- CoreWeave supplies Meta capacity worth $35.2 billion ($21 billion in April 2026 + $14.2 billion from September)
- Meta plans to rent excess AI compute capacity as a cloud service—potentially including access to proprietary models like Muse Spark
- 200 megawatts of compute could generate roughly $10 billion per year for Meta—about 7–8 percent of its 2026 capex guidance
- Meta's capex pace must double: from $19.84 billion in Q1 to $35–42 billion per quarter for the remaining nine months of 2026
The capex acceleration challenge
Meta has guided for $125 to $145 billion in capex for 2026. After spending $19.84 billion in Q1, the company must deploy $105 to $125 billion over the remaining nine months—or $35 to $42 billion per quarter. That's a 77 to 110 percent acceleration from Q1 pace.
The math reveals Meta's strategy: it's not building solely for internal use. A planned cloud unit could convert excess capacity from pure expense into inventory, easing investor concerns—if Meta can actually rent it at premium rates.
Internal AI roadmap under pressure
There's a catch. CEO Mark Zuckerberg told employees that AI agent development is moving slower than expected and some restructuring bets "haven't come to fruition yet," Reuters reports. He expects material AI benefits only in the next three to six months.
This creates timing pressure. If internal AI projects don't deliver returns quickly, cloud rental becomes a hedge against slower internal ROI. Conversely, if Meta can't sell the capacity, the massive capex increase remains pure infrastructure risk.
Winners and losers in the neocloud space
The market reaction was brutal for neocloud specialists—AI infrastructure providers. D.A. Davidson managing director Gil Luria told Reuters that firms like CoreWeave and Nebius rely heavily on Meta, while Meta "may not need them anymore."
This is the core risk for these suppliers: their business models assumed hyperscalers would buy external capacity. If Meta becomes a vendor instead, their addressable market shrinks.
What this means for European enterprises
For German infrastructure providers and cloud specialists, this is a wake-up call. As US hyperscalers vertically integrate and begin reselling capacity, pressure mounts on all intermediaries. Yet opportunity exists: suppliers of specialized components, cooling, or networking technology for this massive buildout could benefit—provided contracts aren't already locked in. The next three to six months will reveal whether Meta's AI bet pays off or becomes a cost driver.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
