CoreWeave and Meta Strike $21 Billion AI Cloud Deal

Six-Year Capacity Lock

CoreWeave announced on April 9, 2026 that Meta will pay approximately $21 billion for dedicated AI cloud capacity through December 2032. The arrangement extends an earlier $14.2 billion deal signed in 2025, bringing Meta's total known commitment to CoreWeave to roughly $35 billion.

The capacity will be deployed across multiple locations and include early deployments of NVIDIA's Vera Rubin platform. This is a take-or-pay agreement—Meta commits to paying whether it uses all the capacity or not, securing supply ahead of actual demand.

Inference Over Training

The expanded deal specifically targets AI inference workloads—serving billions of users across Facebook, Instagram, and WhatsApp—rather than one-time model training. Inference requires continuous, low-latency compute across distributed locations, a different profile than centralized training bursts.

CoreWeave's multi-location architecture and GPU-optimized infrastructure offer speed and efficiency that hyperscalers building their own data centers cannot match on a 2026-2027 timeline. Meta is effectively outsourcing part of the deployment risk while securing early access to next-generation chips.

Specialist Cloud Advantage

The deal reinforces the emergence of specialist AI cloud providers—sometimes called neoclouds—as strategic infrastructure backbones. CoreWeave, backed by a $2 billion NVIDIA investment, focuses exclusively on GPU-heavy workloads rather than general-purpose enterprise computing.

Meta could build equivalent capacity itself, but power grid constraints, permitting delays, and chip supply lead times make that impractical for immediate needs. CoreWeave now holds over $66 billion in total contract backlog and is guiding to $12-13 billion in 2026 revenue with $30-35 billion in planned capital expenditure.