Nebius Group is no longer a fringe AI vendor—it’s become the go-to cloud supplier for hyperscalers, and the order book is already larger than its current capacity.
While hyperscalers race to build out artificial-intelligence capacity, one small European operator has quietly become the industry’s most coveted subcontractor. Nebius Group (NASDAQ: NBIS) reported triple-digit revenue growth last quarter, citing demand that outstrips supply—a phrase Wall Street rarely hears outside of GPU circles.
The catalyst: a pair of mega-contracts that dwarf the company’s current market cap. Microsoft signed a framework worth up to $19.4 billion for dedicated AI clusters, and Meta quickly followed with a $3 billion agreement limited only by Nebius’ ability to deliver racks fast enough.
From regional host to AI gatekeeper
Nebius began life as Yandex’s infrastructure arm, spinning out in 2023 with a handful of Scandinavian data halls and a niche cloud business. Management pivoted hard toward AI, betting that hyperscalers would rather rent purpose-built GPU farms than fight Nvidia’s allocation queue themselves.
The wager paid off. Nebius now offers:
- Priority access to Nvidia H100 and next-gen B200 GPUs through a direct supply agreement
- Managed Kubernetes clusters pre-loaded with CUDA, PyTorch and MLX libraries
- Liquid-cooled racks that push density past 100 kW per cabinet—double the industry norm
That technical edge let the company quote power-efficient AI at $1.85 per GPU-hour, undercutting AWS p4d instances by 34% while promising 99.9% availability.
Balance-sheet firepower for a build-out blitz
To bridge the supply gap, Nebius raised $4.1 billion in December through a convertible note plus a follow-on equity offering. Bond filings show proceeds earmarked for:
- 500 MW of new data-center shell across Finland, Kansas and Illinois
- 70,000 additional H100 GPUs before mid-2027
- A 400 Gbps private fiber ring linking European and U.S. campuses for low-latency training jobs
Capex guidance jumps to $2.3 billion in 2026, triple last year’s spend, yet net-debt-to-Ebitda is projected to stay below 1.3× on contracted cash flows.
Why Microsoft and Meta are writing multiyear checks
Cloud customers increasingly demand reserved AI capacity rather than best-effort spot. Microsoft’s framework guarantees 95% GPU availability for OpenAI and Copilot workloads through 2030, insulating the software giant from Nvidia’s chronic shortages. Meta uses the same clusters to train its next-generation Llama models, with the option to scale to 64,000 GPUs per job.
Both contracts include take-or-pay clauses, meaning Nebius books revenue even if the client underutilizes the hardware—an arrangement normally reserved for telecom tower REITs, not five-year-old cloud firms.
Valuation: priced like a services firm, growing like a chip maker
Nebius trades at 6.2× 2026E sales, a discount to pure-play AI silicon names (Nvidia at 14×, AMD at 8.7×) despite similar growth trajectories. Analysts model revenue compounding at 82% annually through 2029, driven by:
- AI cloud revenue hitting $7.4 billion by 2029 versus $800 million in 2024
- GPU-hour volumes surpassing 12 billion annually, equal to 1.4 million H100s running 24/7
- Ebitda margins expanding to 46% as utilization climbs past 85%
Bulls argue the stock could rerate to 10× sales if the company sustains 70%-plus growth, implying 120% upside from current levels.
Risks lurking in the rush
No story this hot comes without hazards:
- Supply choke: Nvidia allocation remains the single biggest bottleneck; any slip delays revenue recognition
- Power politics: Nordic regulators can cap energy use during winter peaks, capping rack density gains
- Customer concentration: Microsoft and Meta could represent >70% of 2026 revenue, exposing Nebius to renegotiation risk
Short interest has climbed to 18% of float as skeptics bet margin guidance proves too aggressive once competition from Amazon, Google and CoreWeave scales.
Bottom line for investors
Nebius has transformed from an obscure European spin-off into the critical middleman of the AI build-out. With contracted revenue covering the next decade and Nvidia on the cap table, the company enjoys a pipeline most infrastructure firms can only dream of. Shares could stay volatile around quarterly GPU delivery numbers, but the secular tailwind is intact: every hyperscaler on the planet needs more AI capacity than they can build themselves.
Position sizing matters—this is still a single-product story levered to Silicon Valley’s allocation whims—but the risk-reward skews sharply upward so long as demand continues to exceed supply.
Stay ahead of the next wave of infrastructure winners—bookmark onlytrustedinfo.com for the fastest, most authoritative finance analysis on the market.
