Meta Platforms, which has spent aggressively on AI infrastructure to power its own models and recommendation systems, is reportedly weighing whether to sell compute access to third parties — effectively entering the commercial cloud and AI-infrastructure-as-a-service market. The move would follow a playbook Amazon pioneered when AWS was spun out of internal need, and it carries significant weight given Meta's reported plans to spend $60-65B on capex in 2025 alone.
The immediate casualties are neocloud names CoreWeave and Nebius, whose entire investment thesis rests on the idea that GPU compute supply remains tight and that independent cloud operators can command premium pricing from AI developers who can't get capacity from hyperscalers. Meta's entry — even as a threat rather than a confirmed product — undermines that scarcity argument.
The bear case for neoclouds is straightforward: a vertically integrated operator with Meta's scale, financing cost, and existing infrastructure can undercut on price while still turning a profit, the same dynamic that crushed independent data-center operators when AWS scaled. CoreWeave in particular IPO'd at a premium valuation that demands a prolonged tight-supply environment.
The bull case for the neoclouds is that Meta's infrastructure is optimized for its own workloads (inference at massive scale, recommendation systems), not general-purpose GPU rentals — retooling for multi-tenant cloud is operationally and contractually complex, and Meta may never actually launch a commercial offering. Additionally, demand for GPU compute continues to expand faster than any single supplier can satisfy.
What to watch: any official Meta announcement confirming or denying a commercial cloud product, CoreWeave's next earnings print and customer concentration data, and whether Nebius or CoreWeave see booking cancellations or pricing pressure in upcoming guidance.