How they use Nebius AI Cloud

Meta buys dedicated AI infrastructure from Nebius. Nebius reported that both tranches under its earlier agreement had been delivered by February 2026. Nebius expected to recognize 12 months of revenue for Meta’s first tranche and roughly 11 months for the second in 2026. Nebius secured the memory and storage components for that contract in the second half of 2025. At Inflection 2026, Nebius placed Meta among users of its bare-metal infrastructure. Nebius says these large customers run their own software stacks and buy basic bare-metal services. Nebius described supplying Meta and Microsoft with custom racks and servers, GPUs connected through InfiniBand, and multi-tier storage to help build their internal infrastructure.

A separate March 2026 agreement covers additional dedicated capacity built around NVIDIA Vera Rubin systems, with delivery scheduled to start in 2027. Those systems were not scheduled to be delivered until 2027. Meta also committed to purchase remaining available capacity across certain upcoming Nebius clusters over five years. Nebius intends to sell that capacity to third-party AI cloud customers first. The companies have not named the internal models served. The Motley Fool reports that Meta's first contract requires payment whether or not Meta uses the capacity, with individual tranches cancellable for missed delivery dates or repeated availability failures. It also reports discounted monthly fees for late delivery under the March agreement. TradingNEWS reports that Meta’s commitments to Nebius have been partly funded through advance payments.

The London Institute for Mathematical Sciences describes Nebius as helping Meta train and deploy frontier models at scale. A MongoDB event description also names Meta among organizations running production workloads on Nebius.