Nebius: Priced for 2028, Financed Until Then

The Setup

Two years ago, Nebius was an obscure collection of assets spun out of the former Yandex. Today it is one of the most talked-about AI infrastructure names on the market, up roughly 175% this year, valued near $50 billion at its recent price around $218, and briefly worth closer to $65 billion at its June peak. The company builds and rents out large clusters of Nvidia GPUs, the picks and shovels of the AI buildout, and demand for that capacity is genuinely fierce.

This is not an argument that the demand is fake. It is not. Nebius management describes a business where power is sold out and four or more customers compete for every GPU cluster it brings online, and the contracts it has signed are real and large. The argument here is narrower: the price, and even the reported profitability, describe a company much further along and much more self-sustaining than the one in the filings. The gap between the story and the statements is the subject of this piece.

Nebius belongs to the same family as the AI infrastructure names we have examined before, including the IREN avoid thesis, where a real technology story ran well ahead of the realized business. The specifics differ. Nebius has better contracts and a stronger strategic position than most. But the pattern of pricing a distant, capital-intensive outcome as though it were already secured is the same, and the accounting this quarter makes it especially easy to miss.

The Contracts That Made It a Large Cap

The bull case is strong and belongs up front. In its first quarter of 2026, Nebius reported revenue of $399 million, up 684% year over year, with the core AI cloud business growing 841% to nearly $390 million and making up 98% of the total. The metric Wall Street fixates on, annualized recurring revenue, reached $1.92 billion exiting the quarter, up 54% in three months. This is one of the fastest revenue ramps in the public market.

The contracts behind it are the reason the stock re-rated from curiosity to large cap. In March, Nvidia invested $2 billion directly in Nebius and granted it Exemplar Cloud status with priority access to next-generation GPUs, a genuine strategic advantage in a supply-constrained market. In the same month, Nebius signed a five-year agreement with Meta valued at up to $27 billion, including $12 billion of dedicated next-generation compute and $15 billion of committed capacity. It added multi-billion-dollar arrangements with Microsoft and raised its contracted power guidance to 4 gigawatts, anchored by a new 1.2 gigawatt facility in Philadelphia.

Take it at face value, because it is real. Nebius has Nvidia as an investor and supplier, some of the largest technology companies in the world as customers, and a demand pipeline it describes as capacity constrained rather than demand constrained. That is the story the rally is buying, and it is a legitimately impressive one. The rest of this article is about what that story costs and what the most recent results actually show underneath it.

The Profit That Came From a Markup

Here is where a careful reader has to slow down. Nebius reported net income of $621 million for the quarter, a headline that looks like a profitable, scaled business. It was not operating profit. The bulk of that figure came from a one-time valuation adjustment tied to a funding round at ClickHouse, a company in which Nebius holds a stake. A markup on the paper value of an investment is a real accounting event, but it is not money earned from renting GPUs, and it will not recur each quarter.

The same care applies to cash. Nebius reported operating cash flow of $2.3 billion, a startling swing from an outflow a year earlier. That number was driven largely by upfront payments from customers, cash collected in advance for capacity to be delivered later. That is a genuine vote of confidence from those customers, and it is useful funding. But it is deferred revenue, a timing benefit, not a measure of sustainable cash generation from the ongoing business. Count it once and you flatter the run rate.

None of this means the underlying business is unprofitable. The AI cloud segment reported an adjusted EBITDA margin around 45%, which is real and healthy. The point is narrower and important: the headline $621 million profit and $2.3 billion of operating cash flow, the two numbers that make Nebius look like it can fund its own ambitions, are inflated by items that do not repeat. Strip them out and you are left with a fast-growing, high-margin, but still deeply cash-consuming buildout. Which brings us to the capital plan.

The Capex No Cash Flow Can Cover

Nebius raised its 2026 capital expenditure guidance to $20 to $25 billion, up from a prior range of $16 to $20 billion. Set that against full-year revenue guidance of roughly $3 to $3.4 billion. The company plans to spend on the order of seven times its annual revenue building data centers this year alone. One analyst model puts cumulative capital expenditure at more than $76 billion through 2028. This is not a criticism by itself. Building ahead of contracted demand is the entire neocloud strategy. It is, however, the central risk, and it is what the reported profit disguises.

That spending cannot come from operations, because operations do not generate anything close to it. It comes from the capital markets. In the first quarter alone, Nebius raised more than $6 billion, including $4.3 billion of convertible senior notes and the $2 billion Nvidia equity investment, and it ended the quarter with $9.3 billion of cash. Convertibles are debt that becomes dilution, equity raises are dilution directly, and a plan of this scale will require repeated trips back to investors. The model works only as long as the capital markets stay open and generous on the way to a revenue base that does not yet exist.

This is the structural bet inside the stock. Nebius is spending tens of billions of dollars it must raise, to build capacity for revenue the Street models at roughly $11 billion in 2027 and $21 billion by 2028. If that revenue arrives on schedule and the GPUs stay full at good margins, the buildout is brilliant. If the ramp slips, if funding tightens, or if AI compute pricing softens as supply catches up, the company is left servicing an enormous, mostly debt-and-dilution-funded asset base against revenue that came in lighter than planned. That asymmetry is not in the reported profit. It is the whole question.

Concentration and Commoditization

Two more risks sit under the contracts. The first is concentration. The deals that made Nebius a large cap are with a small number of enormous counterparties, chiefly Meta and Microsoft. That is a strength today and a vulnerability tomorrow, because those same hyperscalers are building vast amounts of their own capacity. A customer that can eventually make its own compute is a customer whose commitment is only as durable as the current shortage. When capacity catches up with demand, the party with the leverage is the buyer, not the landlord.

The second is commoditization. Renting Nvidia GPUs is a business with real scale advantages but a thin moat, which the platform's Moat score of 5 out of 15 reflects. Nebius competes with CoreWeave, with the hyperscalers' own clouds, and with a growing field of neoclouds, all buying the same chips from the same supplier and renting them to overlapping customers. GPUs depreciate quickly and the next generation resets the field. In a commodity buildout, the winners are decided by cost of capital and utilization, not by brand, and a valuation that assumes durable premium economics is making an assumption the industry structure does not obviously support.

What the Wealth Engine Scores Say

Before the valuation verdict, here is what the Wealth Engine Pro platform's systematic scoring shows for Nebius right now.

The platform scores Nebius Moderate, not Weak, and that is the honest read: this is a real, fast-growing business with a defensible cash position, not a broken one. The Financial Health score of 57 is the highest of any name this column has flagged in weeks, reflecting the large cash balance. But Fair Value is flagged Expensive at $59.68, roughly 74% below the current price, and the Outlook is Bearish across every recent reading. The systematic message is not that the company is bad. It is that the price has moved far beyond what the reported fundamentals support.

These scores are systematic. They evaluate companies based on reported financials, balance sheet quality, moat characteristics, and valuation models. They measure what a company is today, not what it might become, and they do not give credit for a 2028 revenue ramp until it appears in the results.

In this case, the editorial thesis and the platform scores point in the same direction. The article argues that the price and the headline profit both overstate the realized business, and the systematic data, reading Moderate strength against an Expensive, Bearish valuation, lands in the same place from a different angle. When both the quantitative model and the qualitative analysis flag the same gap between price and substance, that convergence is worth paying attention to. Research Nebius yourself on the platform and decide which signal matters more for your situation.

The Valuation Verdict

At roughly $50 billion, Nebius trades near 75 times trailing earnings and around 15 times this year's revenue guidance, on a business whose reported profit came from a markup and whose cash flow was flattered by prepayments. The platform's blended fair value of $59.68 sits about 74% below the current price. That fair value is strict and backward-looking, and Nebius is growing into it faster than almost any company its size. But the gap is enormous, and closing it requires the future to arrive nearly on schedule.

It is telling that even constructive analysts struggle to find upside. One detailed sum-of-the-parts model, built on a favorable 2028 peer multiple, arrives at a fair value near $234, barely above the recent price, and downgrades the stock to Hold on that basis. When the bull case's own arithmetic lands roughly where the stock already trades, the margin of safety is gone. It is also worth noting that insiders have been net sellers, disposing of more stock than they bought over the past year, with one selling near $276.

The buyer near today's price is not paying for a fast-growing AI cloud with an option on the future. The buyer is paying for the 2028 outcome, the revenue at $21 billion, the margins intact, the buildout financed and full, arriving on time. Each assumption is plausible. Stacked together and paid for in advance, they leave very little room for reward if everything works and a long way to fall if any link breaks.

What Could Go Wrong

This is a demanding steelman, because Nebius is a genuinely strong company and the bull case is not hypothetical.

The demand may be as durable as management says. If AI compute stays supply-constrained for years, Nebius fills every gigawatt it builds at healthy margins, and the capex that looks reckless today looks visionary in hindsight. Being early and aggressive in a genuine shortage is how infrastructure fortunes are made.

Nvidia's backing is a real edge. Priority access to the newest GPUs in a market where supply is the binding constraint is worth a great deal, and Nvidia's equity stake aligns the most important supplier in the industry with Nebius's success. That is not a relationship competitors can easily replicate.

The contracts de-risk the ramp. A five-year commitment of up to $27 billion from Meta, plus Microsoft, means a large share of the 2027 and 2028 revenue is contracted rather than speculative. If those hold, the revenue chart the bulls point to is closer to a backlog than a forecast.

Cash and capital access reduce the funding risk. With more than $9 billion in cash, Nvidia as an aligned investor, and open convertible markets, the financing that worries this thesis has so far been readily available on good terms.

The thesis breaks if the buildout stays funded, the hyperscaler contracts hold, and the 2028 revenue ramp lands close to plan. In that world the profit becomes real, the capex proves prescient, and the stock grows into its price. What the avoid thesis argues is not that this is impossible. It is that the current price already assumes it, and pays in advance for an outcome that is years away and financed with borrowed money. It is the same pattern examined in The Quantum Computing Bubble and across the infrastructure names in Who Is Winning the AI Race: a real opportunity, priced as though the hard part is already finished.

The Bottom Line

Nebius is a good company in the right place at the right time. Its growth is extraordinary, its contracts are marquee, its most important supplier is also an investor, and it has real cash on the balance sheet. If you own it and believe the AI compute shortage runs for years, nothing in this thesis says the business is failing, because it is not. The caution is about price and about what the reported numbers quietly conceal.

The headline this quarter was a $621 million profit and $2.3 billion of operating cash flow. The reality was a one-time markup and a batch of customer prepayments, sitting on top of a business that intends to spend $20 to $25 billion it must raise, against $3 billion of revenue, to build for a 2028 that the stock already treats as delivered. The platform's systematic scoring reads Moderate, Expensive, and Bearish, with a fair value roughly 74% below the price, and even the bull case's own model lands near where the stock trades today.

That is the entire philosophy here. Wealth Engine Pro evaluates companies on what they are and what they have actually earned, not on the best version of the story during the most exciting part of the cycle. The narrative says Nebius is a profitable AI infrastructure champion funding its own rise. The data says it is a fast-growing, cash-consuming buildout whose reported profit came from an accounting entry and whose future is financed on the way to a revenue base that does not yet exist. When the gap between those two pictures is this wide, the data is the part worth trusting until the results prove otherwise.