Right now, a number is doing laps around crypto Twitter: 55x.
Not leverage. Revenue multiple. A European GPU cloud operator named Nebius is being valued at 55 times trailing revenue, according to a Crypto Briefing analysis that crossed my desk at 6 AM Nairobi time. My first instinct, after nearly a decade inside this circus: that's not a valuation. That's a narrative wearing a spreadsheet costume.
I have watched this movie before. ICO season. DeFi summer. The NFT gold rush. Every cycle produces the same character โ a token or a company sitting on a tiny base, inflated by a massive multiple, and the market calls the sum conviction. But Nebius is not vaporware. It has real data centers. Real GPU contracts. Real engineering muscle inherited from the Yandex mothership. And that is exactly what makes this one scarier than the ghosts of 2017.
Because the harder the underlying asset is, the easier it becomes to suspend disbelief. And in AI infrastructure, disbelief is the only margin that matters.
The Setup: Europe's Independent GPU Cloud
Let me catch you up, because this name deserves more than a glance at the ticker. Nebius is the reincarnation of Yandex N.V. โ the Amsterdam-domiciled parent of Russia's largest internet company. After the 2024 corporate split, the shell was reborn as Nebius Group, listed on Nasdaq under NBIS, with the entire balance sheet pointed at one target: AI compute.
The business breaks into three layers. Nebius AI Cloud is the core โ GPU-as-a-service, the cloud-native version of my chips make money while I sleep. Nebius AI Studio is the developer-facing platform, a place where startups land raw models, deployment tooling, and inference workloads without having to negotiate hardware leases at 2 AM. And there is the AI factory model: custom, massive-scale deployments for select partners, with at least one European supercomputing center in the early customer list.
What matters most is not the product names. It's the inheritance. Nebius walked out of the Yandex divorce with a mature data-center operations team, deep experience running hyper-scale infrastructure, and a network of European facilities already humming. It is building GPU clusters in Finland and France, tapping Nordic clean energy to cut power costs and satisfy EU carbon targets โ a detail that turns compliance from a burden into a sales pitch.
For European enterprises with data-sovereignty anxiety โ the banks, the health-tech firms, the government-adjacent projects โ Nebius markets itself as the safe lap. The pitch sells a kind of quiet: you cannot send this workload to AWS. You send it to us. That positioning is the entire ballgame, because the moment this becomes a pure price war on GPU rental, Nebius loses to people with bigger balance sheets.
And at 55x revenue, the market has already paid for that thesis in full.
Deconstructing the 55x
Here is the part where I slow down, because I have made the mistake of staring at a headline multiple without asking what sits underneath it. I'm not doing that twice.
First, the basic math. A 55x revenue multiple means the company's market capitalization is 55 times the revenue generated over the trailing twelve months. For a hypergrowth infrastructure business, that is not automatically insane โ but it is sitting on the upper edge of the envelope that contains reasonable. CoreWeave, the most direct comparable, spent 2024 and 2025 trading in a 20-to-40x range. Nebius at 55x implies the market is paying a premium for something CoreWeave doesn't have. That something is European scarcity.
There are very few pure-play, publicly traded GPU clouds with one foot inside the EU's sovereign AI ambition. Scarcity, in bull markets, gets priced like a controlled substance.
But watch the track of that math over a twelve-month horizon. If you believe Nebius can grow revenue at 100% compound annually โ and that is the working assumption embedded in every allocation rumor and contracted cluster โ then 55x trailing revenue compresses to something closer to 27x forward. Suddenly the multiple stops screaming. It's in line with the neighbor's yard. That is not a rationalization; it's the standard way infrastructure investors value assets that deliberately front-load losses for market share.
The problem is that all of it is an act of faith about numbers nobody outside the company can verify. And I have spent enough late nights reading GPU-cloud financial structures to know exactly where the bodies get buried.
Revenue recognition, for one. GPU cloud contracts are often multi-year and prepaid. Under standard accounting rules, a $100 million deal signed in January does not all hit the income statement in January โ it drips in across the contract's life. But some providers recognize hardware-linked components differently, front-loading the billings and the revenue when equipment ships. If the analysts citing 55x are using trailing revenue as reported, and if a meaningful chunk of Nebius's recent growth came from upfront recognition on prepaid deals, then the multiple is flattering to the upside. The denominator gets inflated by accounting timing, not by recurring demand. This is the kind of footnote that does not make the headline but changes the whole picture.
Then there is the base itself. A 55x multiple on a small number can still be a small absolute valuation. I have seen the tweets screaming bubble, but the math splits two ways. If Nebius's market cap is in the neighborhood of $17 billion, 55x implies trailing revenue around $300 million. That is not a cloud giant. That's a well-funded challenger with one strong growth year under its belt. If the market cap is closer to $5 billion, the revenue base is under $100 million โ and then you are paying for a startup's roadmap, not a company's trajectory. Either way, the multiple is not the story. The trajectory is.
The deeper question โ the one the spreadsheets skip โ is what the 55x is actually buying. Here is my honest technical read after tracking NVIDIA supply chains and reading too many S-1 documents: the market is not paying for Nebius's software. It's paying for allocation. H100s, H200s, the Blackwell units coming down the pipe โ the scarce resource in this industry is not innovation. It's the right to buy NVIDIA silicon before everyone else. Nebius's valuation is a premium on GPU-slot reservations, delivery logistics, and power contracts that are already signed. That's the real asset. The Kubernetes wrapper and the Studio interface are table stakes.
And that is precisely why the valuation is fragile in ways the bulls do not like to admit. GPUs are not indefinite assets. The moment NVIDIA ships a generational jump โ and it is shipping them on a brutal cadence now โ last year's high-end stock becomes a discount line item. If Nebius loads up on debt to buy this generation, and the next generation makes that hardware obsolete in eighteen months, the depreciation cliff hits the income statement exactly when the razor-thin margin conversation starts. Capital expenditure comes first. Revenue arrives late. That timing mismatch has killed every infrastructure overbuild in history.
I lived through the post-Dencun rollup debates โ everyone promising blob space would stay cheap forever, while the capacity math said otherwise. Same disease, different organ. This era's version: everyone promises GPU scarcity is permanent. Capacity has a nasty habit of arriving exactly when the narrative needs it most.
The Utilization Question Nobody Can Answer
Here is the metric that matters more than the multiple, and the one no one is talking about: utilization.
I have been asking GPU cloud CFOs for utilization numbers since this cycle began, and the answer is always the same โ a smile and a deflection. Industry models put average utilization somewhere between 50% and 70% for the private GPU clouds. If Nebius is running at 80% or better โ meaning its expensive silicon is generating revenue most of the time โ then 55x has hard assets underneath it. If it is at 50%, the valuation is funding a lot of idle metal humming in Finnish server halls.
This is where my DeFi instincts kick in, because the pattern rhymes. During DeFi summer, I watched projects buy total value locked with liquidity mining rewards. Shiny numbers, subsidized by token emissions. The instant the incentives stopped, the users evaporated. The same logic applies here. The subsidy in this market is NVIDIA supply plus narrative capital. Strip out the allocation advantage and the European sovereign AI fairy dust, then ask a brutal question: would a European bank pay Nebius's price for raw compute if AWS dropped GPU prices by 30% tomorrow?
Because that is the real competitive threat, and it is not CoreWeave. It is not even Azure. It's NVIDIA. The company whose chips built this entire industry is quietly entering the service layer with DGX Cloud, and NVIDIA's broader investment ecosystem โ including a strategic stake in CoreWeave โ means the supply chain is increasingly owned by the same players who are supposed to be competing downstream. If NVIDIA decides the cloud is its own turf, every middleman's margin structurally tightens. Nebius becomes a distribution channel with a fixed ceiling.
Meanwhile, the hyperscalers are doing what they always do: building their own silicon. AWS Trainium. Google TPU. These are not idle science projects anymore. Every custom chip that avoids an NVIDIA purchase erodes the pricing power that Nebius's entire margin structure depends on. Nebius is a pure NVIDIA-ecosystem shop. When your supplier gains pricing power over you and your customers simultaneously, growth story turns into squeezed middle remarkably fast.
Look at the unit economics, because this is where the rubber meets the road. A top-tier GPU rents at an hourly rate that has already fallen from its 2024 peak. The costs underneath that sticker โ power, cooling, facility debt service, cluster networking, staffing โ are sticky and rising. Electricity prices in Europe are not going down. The AI factory model helps here because it front-loads committed revenue: one anchor customer, one giant contract, predictable utilization. But it concentrates risk. If your anchor customer's own model roadmap stalls, you are left holding the most expensive idle hardware in Europe. And if the anchor is tied to a sovereign program, you are exposed to political cycles, not just product cycles.
The Contrarian Read
So what is the angle nobody is writing?
Conventional wisdom frames the story as Nebius versus the hyperscalers. I think the more dangerous dynamic is the one where Nebius wins.
If the EU's AI Factory initiative actually matures โ if Brussels champions, sovereign wealth funds, and compliance-driven enterprises start signing real contracts โ Nebius could become one of the designated providers. That is the bull case, and it is legitimate. But notice what happens then. The moment European governments treat Nebius as strategic infrastructure, the scrutiny arrives. The Yandex history does not dissolve because the corporate entity moved to Amsterdam. Geopolitical review, sanctions-compliance auditing, board-level anxiety about Russian-rooted engineering talent โ these become recurring costs. Being Europe's AI champion in a bull market is a blessing. Being it during a geopolitical storm is a liability that does not show up in the revenue multiple.
And there is a second contrarian layer. That 55x is not just a number โ it's a magnet for capital already conditioned to pay for narratives. This column is read by crypto people, and I know my audience. We are collectively the most tolerant investors on the planet when it comes to valuation math that depends on future miracles. The AI-infrastructure trade has become the new crypto trade: high-beta, narrative-driven, story-first. The crowd that threw money at tokens with no revenue has found a vehicle with actual data centers and an excuse to say institutional. That is a dangerous combination. Not because Nebius is a scam โ it is not. But because the capital around it behaves like it is entering a casino, while the asset is actually a long-dated power plant with debt covenants.
High multiples don't crash. They deflate. The revenue keeps growing, the margins keep compressing, and one day the market realizes the story was priced two years ahead of reality. That is how the data-center darlings of 2000 went from 100x revenue to 20x. It took years, and the world did not end. It just repriced.
Technical Check: The metric that matters is not on any headline chart. It is utilization, revenue recognition dates, and the replacement value of current-gen GPU inventory when the next silicon generation lands. Verify the denominator before you vibe the multiple. Speed is nothing without verification โ I learned that the hard way in the NFT era, when a casual praise of a roadmap cost me credibility I had to earn back.
What I'm Watching Next
So here is my honest scorecard. The 55x revenue multiple is high, but not absurd โ on forward numbers, with real contract flow, it is top-of-range fair. The bull case collapses if GPU supply floods and rental prices fall. The bear case collapses if utilization stays near 80% and European sovereign deals start landing.
I am watching four signals. Quarterly revenue growth โ does it hold above 50% quarter over quarter? Gross margin โ does the platform business lift the blended number above the raw-rental floor? Book-to-bill โ are prepaid, multi-year contracts expanding? And the quiet one: depreciation policy and debt terms on the next GPU purchase round.
The silence after the pump tells the real story. Every time I have seen a 55x narrative break, it did not break with a scandal. It broke with a spreadsheet footnote. Nobody screamed. The valuation just quietly caught up to what was actually operating in the building.
Right now, Europe's most credible independent GPU cloud is running at a euphoric multiple, armed with a real balance sheet and a genuine chance at becoming a sovereign-infrastructure cornerstone. I genuinely want that story to work. But after the ICO glass ceilings, the DeFi summers, the NFT honeypots, and watching founders survive the 2022 wreckage, I know one thing for certain.
Strip out the subsidy. Strip out the scarcity. Strip out the narrative capital that crypto-adjacent money brings to every shiny new thing.
What is left is a European infrastructure company with excellent engineering and a brutal fight ahead. And at 55x, that company is priced as if the fight is already over.
The fight is never over. That is not a warning. That is the opportunity.