Mark Zuckerberg just pulled the rug out from under the neocloud party. CoreWeave and Nebius shares cratered Wednesday after reports surfaced that Meta is considering selling access to its massive AI compute infrastructure. The market’s message was clear: the party might be over before the champagne got warm.
CoreWeave plunged 18% in afternoon trading. Nebius, the Amsterdam-based cloud operator, fared worse, shedding 22% of its value. The selloff wasn’t just about two companies—it was a referendum on an entire sector that had been riding high on AI hype.
Neoclouds—specialized cloud providers built for GPU-heavy AI workloads—have been the darlings of the infrastructure boom. Unlike AWS, Azure, or Google Cloud, they promised speed and flexibility for AI startups hungry for Nvidia GPUs. For a while, it worked. CoreWeave went from a crypto-mining shop to a $19 billion valuation. Nebius, reborn from Yandex’s ashes, became Europe’s answer to the AI compute shortage.
Meta’s Moat Becomes a Weapon
The threat is existential. Meta isn’t just any potential competitor—it’s the largest single buyer of Nvidia H100 GPUs in the world. The company has stockpiled hundreds of thousands of these chips to power its own AI ambitions. If Zuckerberg decides to rent out that spare capacity, he can undercut every neocloud on price. Scale works in his favor. Meta’s infrastructure was built for internal use; selling excess capacity is just gravy. For CoreWeave, every GPU they buy has to generate a profit. Meta’s chips are already paid for.
"This is a classic platform play," says Ravi Patel, a cloud infrastructure analyst at Forrester. "Meta doesn’t need to make money on cloud services initially. They just need to make it unprofitable for everyone else."
"Meta doesn’t need to make money on cloud services initially. They just need to make it unprofitable for everyone else."
The numbers back that up. Meta’s capital expenditure hit $35 billion last year, most of it dedicated to AI compute. Even a fraction of that capacity, offered at marginal cost, would wreak havoc on neocloud margins. CoreWeave’s gross margins are already under pressure—they fell from 65% to 52% over the past two quarters as competition for GPU supply intensified.
The Zuckerberg Playbook
This isn’t Meta’s first rodeo. Zuckerberg has a history of entering markets late, copying successful features, and using his user base to smother competitors. He did it with Instagram Stories (killing Snapchat’s growth), Reels (stalling TikTok’s momentum), and WhatsApp Payments (squeezing fintech startups across Asia). The pattern is brutal: let others prove the market, then leverage existing infrastructure to dominate.
Cloud compute is no different. Meta already has data centers spanning the globe, a private fiber network, and deep relationships with Nvidia. Announcing a cloud service would be trivial—flip a switch, set up a billing system, and start charging. The Wall Street Journal report that sparked Wednesday’s selloff suggests Meta has been quietly testing the waters with select enterprise customers.
"The neocloud thesis was always fragile," says Maria Gonzalez, a venture capitalist who backed several AI startups. "If Meta or Amazon ever decided to compete aggressively, the neoclouds had no moat. Their only advantage was speed, and that evaporates when a trillion-dollar company shows up."
The timing is brutal. CoreWeave was set to file for an IPO next month, hoping to raise $3 billion. That prospect now looks shaky. Nebius, which went public via a SPAC last year, has seen its stock cut in half since January. The sector’s total market capitalization has shrunk by over $10 billion in a single day.
What the Neoclouds Are Saying
CoreWeave’s CEO, Michael Intrator, tried to calm investors in an afternoon memo. He argued that Meta’s infrastructure is optimized for its own workloads—social media recommendation engines and large language models for internal tools—not for the diverse needs of third-party developers. "Our infrastructure is purpose-built for AI training and inference at scale," the memo read. "Meta’s data centers are designed for Meta."
It’s a weak defense. Meta runs PyTorch and hosts some of the largest AI models in existence. Their hardware stack—Nvidia DGX systems, custom networking, and software—is identical to what neoclouds offer. And Meta’s engineering team can adapt to any workload. The only real barrier is organizational inertia. Once the decision is made to go to market, execution would be swift.
Nebius CEO Arkady Volozh struck a more defiant tone in an interview with Bloomberg: "We’ve seen this before. Hyperscalers talk about opening up, but they never do it well. Their focus is always internal. We live and breathe cloud-native AI. They don’t."
Investors weren’t buying it. The stocks kept falling.
The Bigger Picture
Wednesday’s selloff is a warning for any sector built on scarcity. The AI boom has been fueled by a GPU shortage, allowing neoclouds to charge premium prices. But that shortage is ending. Nvidia is ramping production, new competitors like AMD and Intel are entering the market, and hyperscalers are building their own custom chips (AWS Trainium, Google TPU, Meta MTIA).
The neocloud business model—buy GPUs, rent them out at a markup—works when GPUs are hard to get. It fails when supply catches up. And if the biggest GPU buyer in the world starts selling access too, the pricing power completely evaporates.
"The neoclouds were a bet on inefficiency in the market," says Patel. "That bet is now expiring."
Some analysts see a potential silver lining: Meta could become an acquirer. "If CoreWeave’s stock falls far enough, Zuckerberg might buy them," says Gonzalez. "It would give Meta instant talent and customer relationships." But acquisition—at a fire-sale price—is cold comfort for current shareholders.
For now, the market is voting with its feet. CoreWeave and Nebius are down, and the neocloud sector is in crisis mode. The question isn’t whether Meta will enter the cloud business—it’s when, and how brutally.
Zuckerberg is playing the long game. He always does. And the neoclouds are learning a painful lesson: when a giant decides to sit at your table, there’s not much food left for anyone else.
