Nebius at 55x Revenue: The Market Is Buying a Story, Not a Balance Sheet

Over the past seven days, one data point has dominated every conversation I have had about AI infrastructure: 55. That is the trailing revenue multiple assigned to Nebius, the Amsterdam-headquartered GPU cloud company born from Yandex N.V.'s 2024 split. A 55x multiple triggers the same instinct I developed while auditing token distribution back in 2017: do not yell bubble, do not nod along. Decompose the number. In a sideways market, clarity is more valuable than conviction, and valuation multiples are where clarity goes to die. Nebius is not a financial abstraction. It operates a GPU cloud called Nebius AI Cloud, an AI-native platform called Nebius AI Studio, and a custom "AI factory" model that has paired it with European supercomputing centers. The split from Yandex left it with something few new entrants can claim: real data center operations, cold-start engineering talent, and a balance sheet large enough to place huge GPU orders. It also inherited a complicated geopolitical shadow, even though the Russian assets were removed. In Europe, that shadow cuts both ways. Some enterprise buyers will avoid it; some sovereign-buyers will welcome an EU-domiciled alternative to American hyperscalers. The market narrative is straightforward: this is the European CoreWeave. And CoreWeave spent 2024 and 2025 trading in a 20 to 40 times revenue band. So Nebius at 55 times is not off by an order of magnitude; it is at the upper edge of a frothy peer group. But that comparison can mislead as easily as it clarifies. The 55x is almost certainly based on trailing twelve-month revenue, not forward revenue. If you apply the market's typical forward discount, Nebius may be trading at 20 to 30 times next year's expected revenue. That is a completely different conversation. Every projection now hinges on three numbers: utilization, gross margin, and booked contracts. During DeFi Summer in 2020, I watched protocols with 100x revenue multiples implode when utilization fell below 50%. GPU clouds face the same arithmetic. Industry utilization for specialized GPU providers is often opaque, but the available evidence suggests a 50 to 70 percent range for many operators. If Nebius can keep its clusters above 80 percent while locked into long-term contract revenue, the 55x multiple is a reasonable growth premium. If utilization sits near 50 percent, that multiple is simply a down payment on a future that hasn't happened yet. The middle term is gross margin. Nebius is, at its core, an NVIDIA hardware reseller wrapped in software and power contracts. NVIDIA's pricing power determines the ceiling. A pure GPU reseller in a tight supply environment can generate healthy near-term margins, but as supply catches up, the floor drops. The good news is visibility into contract momentum is starting to improve. Book-to-bill, quarterly CapEx guidance, and disclosures around total contract value will tell more than any price chart. A safe harbor for forecasters is to watch the two quarters. If revenue is still growing 50 percent or more quarter-over-quarter, and if the company announces contracts measured in billions, the narrative is being validated. If revenue growth stalls or gross margins compress, the multiple will reset toward 10 to 20 times revenue—not because the company is failing, but because that is what infrastructure assets trade at when the scarcity premium disappears. One factor rarely enters the valuation debate: electricity. The bottleneck has shifted from GPU allocation to grid interconnection and power prices. Nebius chose Finland and France partly because Nordic clean energy lowers operating costs and aligns with EU climate rules. But grid permits take years. Every megawatt of delay pushes break-even utilization higher. This is not software; it is physical infrastructure. I treat any AI infrastructure estimate that ignores power as incomplete. Now for the contrarian angle. The obvious threat list includes AWS, Azure, and Google Cloud. They are all dropping prices on GPU instances, and each has custom silicon—Trainium, TPU—that reduces their dependence on NVIDIA. But the sharper threat is NVIDIA itself. Nebius exists at NVIDIA's pleasure. NVIDIA can prioritize its own DGX Cloud, or funnel scarce supply to partners it has invested in, or raise the price of H100 and Blackwell systems to squeeze the margin out of lease-only competitors. Nebius is not a platform in the way AWS is a platform. It is a dependent variable in someone else's supply chain. The next risk is asset depreciation. A data center full of H100s is an asset that becomes obsolete with every new silicon generation. If Blackwell or Rubin deployments accelerate, the resale value of existing GPU clusters can collapse. Nebius has to keep borrowing and ordering new hardware while those old assets sit on its balance sheet. That is a leveraged bet on NVIDIA's roadmap, not just on AI demand. There is also a political element. The European Union's AI Act and the AI Factory initiative create demand for privacy-conscious, EU-controlled compute. Nebius is well positioned to benefit from that. But the company's Yandex origin will attract continuous scrutiny. Every procurement decision will involve a geopolitical review. That is a cost of doing business, and it is also a moat. American hyperscalers face their own sovereignty challenges inside Europe. What would make me change my mind? Two signals. One, if Nebius wins a formal role in the EU AI Factory project or a comparable sovereign compute contract, that would validate the scarcity premium and make the 55x look cheap in hindsight. Two, if quarterly gross margin stabilizes above what pure GPU resellers usually earn, that would prove the software platform is adding value beyond the hardware. I have not seen enough data to know whether either will happen. But this is what 55x means right now when you push past the math: someone is paying for a future where European AI sovereignty is real, where compute is the new currency, and where Nebius becomes the trusted intermediary. Code is law, but people are purpose. Community is the new central bank, and in Europe the community is a coalition of regulators, researchers, and enterprises looking for dignified compute. The 55x revenue multiple is not a fraud. It is a test. It will pass if utilization stays high, contracts keep landing, and gross margins show pricing power beyond NVIDIA markups. It will fail if depreciation accelerates, access to supply dries up, or the next generation of chips leaves those clusters stranded. Don't trust, verify. But also, connect. Resilience beats hype every time. The same math that killed many DeFi protocols is now running quietly through European data centers. This time, let's watch the data before we celebrate or mourn.