Nebius Climbs 21% on $1B+ Meta AI Deal, Nvidia GPU Partnership

Nebius Group shares rose 21.3% following announcement of a multi-year AI infrastructure agreement with Meta and a strategic partnership with Nvidia that grants the Amsterdam-based company priority access to H100 and H200 GPU clusters. The move positions Nebius as one of three non-U.S. firms with direct Nvidia allocation outside China, alongside European sovereign compute initiatives.

The Meta deal structures as a capacity-as-a-service arrangement where Nebius provides dedicated AI training infrastructure across its Finnish and North American data centers, with initial deployment targeting 10,000+ GPUs by Q3 2025. Financial terms remain undisclosed, though industry pricing for comparable H100 clusters suggests annual contract value between $800M-$1.2B based on current market rates for dedicated AI compute. Nebius will maintain ownership of the physical infrastructure while Meta secures multi-year reserved capacity with performance guarantees tied to model training throughput rather than raw compute hours.

The Nvidia partnership delivers two mechanisms: preferential allocation from Nvidia's constrained supply queue and co-engineering on liquid-cooling deployment at scale. Nebius already operates 40MW of GPU-dense capacity in Finland's Mäntsälä facility, where ambient temperatures and hydroelectric power create 15-20% lower operational costs than comparable U.S. sites. The Nvidia relationship effectively converts Nebius from a regional cloud provider into a strategic infrastructure partner for frontier model training, particularly for European customers navigating data residency requirements under the AI Act.

What matters for allocators: this is the first major hyperscaler validation of a European AI infrastructure pure-play since the Yandex spinoff completed in November 2024. Nebius inherited Yandex's cloud engineering talent but shed the Russia exposure that kept institutional capital sidelined. The Meta contract suggests Meta is pursuing geographic diversity in AI compute after concentration risk became visible during recent U.S. power grid constraints in Virginia. For Nvidia, Nebius serves as a European beachhead that doesn't trigger the same geopolitical scrutiny as direct sales to sovereign compute projects.

The stock traded at $23.40 before the announcement, implying a pre-deal market cap around $2.1B. Post-surge valuation of approximately $2.5B still prices the company at a significant discount to pure-play cloud infrastructure multiples, which typically command 8-12x forward revenue for high-growth assets. If the Meta contract alone represents $1B in annual recurring revenue at maturity, current valuation assumes either significant execution risk or investor unfamiliarity with the separated entity.

Operators and allocators should track three developments: GPU deployment pace through Q3 2025 earnings calls, any follow-on hyperscaler announcements (Microsoft and Anthropic both face similar European compute gaps), and whether Nebius pursues debt financing to accelerate data center buildout. The Nvidia partnership also positions the company for potential H200 early access in H2 2025, which would create a 6-9 month technological lead over competitors still running H100 clusters.

Nebius management has signaled plans to expand Finnish capacity by an additional 60MW by late 2025, which would support approximately 15,000 more GPUs at current density ratios. That timeline aligns with the second phase of European AI Act compliance, when data residency requirements become binding for frontier model deployment. Meta's willingness to lock in long-term European capacity suggests the hyperscalers are pricing regulatory risk into infrastructure strategy faster than the market expected.