Nebius Group disclosed $46 billion in binding AI cloud infrastructure commitments on Tuesday, including a named partnership with Meta and expanded Nvidia collaboration. The stock closed up 21.3% in New York trading. The company spun out of Yandex eighteen months ago with data center assets in Finland and ambitions to compete with AWS and Google Cloud on GPU-dense workloads.
The $46 billion figure represents multi-year infrastructure commitments, not booked revenue. Nebius confirmed Meta as an anchor tenant for training clusters and disclosed it is co-designing liquid-cooled rack architecture with Nvidia for H200 and Blackwell deployments starting Q2 2025. The company operates 40,000 GPUs today across three European facilities and has purchase orders for an additional 80,000 units through 2026. Management stated average contract duration is four years with take-or-pay clauses at 85% of committed capacity.
This matters because hyperscaler GPU wait times have stretched to nine months for H100 clusters above 16,000 nodes, and no tier-two provider has previously disclosed a pipeline of this scale with named enterprise anchors. Meta's participation signals the company is diversifying beyond its Llama infrastructure investments with AWS and is willing to lock in capacity with a relatively young vendor. Nvidia's involvement in rack co-design grants Nebius early access to Blackwell at volume, a privilege typically reserved for the hyperscale four. The timing is precise: OpenAI, Anthropic, and xAI are all competing for the same Q2-Q3 2025 Blackwell allocation window. Nebius is effectively pre-empting that bottleneck by securing commitments now and building to meet them.
The market is repricing Nebius from a regional play to a credible alternative for frontier model training, particularly for customers who want European data residency or need to avoid hyperscaler conflicts of interest. The company's revenue run rate remains modest—roughly $400 million annualized as of Q4 2024—but the disclosed pipeline implies $11.5 billion in average annual contract value if commitments deploy evenly. The stock now trades at 3.2x forward revenue on those pipeline assumptions, below Coreweave's last private round multiple of 4.1x. The risk is execution: Nebius must build out 120,000 GPU equivalents in 24 months without the balance sheet or vendor relationships of an Azure or GCP.
Allocators should watch for Nebius's Q1 2025 earnings in late April, when management will disclose how much of the $46 billion converts to recognized backlog under ASC 606 and whether Meta's commitment is capacity reservation or dedicated infrastructure. Nvidia's next GTC keynote in March will clarify Blackwell production timelines; any slippage there would compress Nebius's build schedule. Additionally, watch for debt or equity raises in the next 90 days to fund the capex ramp—the company will need roughly $8 billion in project financing to meet 2026 delivery targets.
The company reports Q1 results on April 28. Meta's Q1 capex guidance, due February 5, will show whether the Nebius deal is incremental or cannibalizes existing hyperscaler spend.