Nebius: The Highest-Conviction Asymmetric Bet in AI Infrastructure

A year ago, you had to squint to see a business. Today, $46 billion in committed contracts, a $27 billion Meta deal, and a $2 billion NVIDIA investment later, the stock is down 31% from its all-time high, and the fund managers who own it are telling you exactly why that's the opportunity.

A year ago, you had to squint to see a business. Nebius Group was a Dutch holding company with a Russian founder, a handful of GPU contracts, and a lot of skepticism. Today, after $46 billion in committed customer contracts, a $27 billion deal with Meta, a $2 billion investment from NVIDIA, and 684% revenue growth, the stock sits at $205, down 31% from its June 22 all-time high of $300. The business has never been stronger. The stock has never been cheaper relative to what it's already signed.

The origin story

Nebius was born from the largest corporate exit from Russia in history. Arkady Volozh built Yandex, the "Google of Russia," into a $30 billion company over 25 years. After Russia invaded Ukraine, Volozh divested every Russian asset in a $5.4 billion transaction, reconstituted the company in Amsterdam, and kept the crown jewels: a 1,300-strong engineering team with decades of experience building large-scale computing infrastructure, plus ownership stakes in ClickHouse, Avride, Toloka, and TripleTen.

He had $2.3 billion in cash, a world-class team, and a blank slate. He decided to build something that didn't exist yet: a purpose-built AI cloud, vertically integrated from the hardware up, optimized for the workloads that the hyperscalers were too slow to serve.

What they actually built

Nebius is an AI infrastructure company in three layers.

Layer 1: Bare Metal GPU. Nebius rents GPU-packed data centers to the world's largest technology companies. The contracts are massive: $19 billion with Microsoft, $27 billion with Meta. These are long-term, take-or-pay deals that generate 20%+ operating margins. The customer prepayments cover roughly 60% of Nebius's capital expenditure, and the contracts are structured so that Meta and Microsoft effectively underwrite the buildout. As Crossroads Capital put it in their Q1 2026 letter: "Meta's deal provides access to investment-grade borrowing costs with no equity dilution, while acting as a backstop customer if enterprise demand for the AI cloud doesn't materialize on schedule."

Layer 2: AI Cloud Platform. This is the real thesis. Nebius is building a full-stack cloud specifically for AI workloads: multi-tenant compute, unified storage, inference-as-a-service, managed databases, security certifications, and a growing suite of AI-native services on top. CEO Arkady Volozh describes it as "not just multi-tenant cloud, which means basic cloud services, multi-tenancy billing, security, storage, managed databases, it's all there. But on top of it, there is a lot of services specific for AI builders, like our Token Factory, inferencing platform, or agentic search services." Early customers include Shopify, Cloudflare, and Revolut, validation that the platform works for real enterprise workloads.

Layer 3: Hidden Assets. Nebius owns approximately 30% of ClickHouse, the fast-growing real-time analytics database company, whose valuation has been reported in the range of $6 billion to $15 billion. That stake alone could be worth $2 billion to $4.5 billion. It also holds equity in Avride (autonomous vehicles, ranked alongside Tesla and Waymo), Toloka (AI data labeling, backed by Jeff Bezos), and TripleTen (tech education). Baron Global Opportunity Fund, a top-10 holder, estimates these stakes are worth "high-single-digit billions" in total, roughly 15% of Nebius's current market cap at the low end.

The numbers that changed everything

When Crossroads Capital first bought NBIS in late 2025, the bear case was straightforward: no anchor customer, no enterprise counterparties, a small GPU fleet, burning cash, and a Russian-adjacent corporate structure that might spook capital markets. One year later, every single question has been answered in sequence.

In January, Nebius won a contract to build a supercomputer for the Israeli government. In March, NVIDIA invested $2 billion directly, not a purchase order, not a partnership press release, but Jensen Huang putting his own company's capital behind Nebius and designating it an "Exemplar Cloud" for GB300, the next-generation AI GPU. The same month, Meta signed a five-year, $27 billion compute deal, the largest external compute contract in history. Combined with Microsoft, committed contract value exceeds $46 billion.

Then came the Q1 2026 results in May. Revenue of $399 million, up 684% year-over-year and 75% quarter-over-quarter. Growth accelerated. Adjusted EBITDA swung to a $130 million profit. The AI cloud segment hit 45% margins. Annualized recurring revenue reached $1.9 billion. The company raised guidance for the full year to $3.0 to $3.4 billion in revenue, $7 to $9 billion in ARR, and approximately 40% adjusted EBITDA margins.

The bond market took notice. In March, Nebius issued $4 billion in convertible debt at 1.25% to 2.625%, rates typically reserved for investment-grade companies. The conversion premiums were 55% to 57.5%, meaning dilution is delayed far into the future. Crossroads Capital noted: "Nebius walked into the Meta negotiation with silicon supply that its competitors can't match. Two days after the Meta deal closed, Nebius issued debt in size, and the terms were better than anything we'd seen for a company at this stage."

The moat: why this isn't just another neocloud

The bear case for every neocloud is the same: they'll run out of money before they reach scale. Nebius has structurally solved this. Customer prepayments from Microsoft and Meta cover roughly 60% of 2026 capital expenditure. The remaining 40% is financed through convertible debt at sub-2.6% rates. Compare that to competitors diluting shareholders by multiples of the outstanding count. Nebius is financing growth with customer cash and cheap debt while delaying dilution.

NVIDIA's investment is not passive capital. Crossroads Capital explains: "Nvidia is enabling capacity at a preferred customer by paying the reservation fee. Nebius gets priority GPU allocation at zero cash cost and modest dilution. Jensen gets a free option on execution he can self-catalyze. Both sides win." In a market where GPU supply is still the binding constraint for every AI infrastructure company on Earth, priority access is a structural advantage that no competitor can replicate by simply raising more money.

Then there is power. GPUs are increasingly a commodity: anyone with capital can buy them. Power and land are not. Nebius controls more than 3.5 gigawatts of contracted power, with over 75% of it owned rather than leased. The 1.2 gigawatt Pennsylvania campus is a generational asset: land, power, and water secured for decades. When the AI buildout runs into the power grid bottleneck, and every data center operator in the world expects it will, Nebius's owned capacity becomes the moat.

Finally, the team. Volozh and his engineers built Yandex Cloud, one of the few non-US cloud platforms that competed effectively against AWS, Azure, and Google Cloud. They understand what it takes to build a cloud from scratch, and they are doing it faster the second time.

The valuation: three ways to see it

At $205 per share, Nebius trades at a market capitalization of approximately $53 billion. Consensus analyst estimates put the stock at $244 on average, with a high target of $380, implying 16% upside to the mean and 85% upside to the high target. The PEG ratio, price-to-earnings divided by growth rate, sits at 0.63, a level that typically signals a stock is undervalued relative to its growth trajectory.

Baron Global Opportunity Fund values Nebius from three angles. First, the bare metal contracts alone are worth "mid-single-digit billions in equity value." Second, the hidden assets, ClickHouse, Avride, Toloka, and TripleTen, are worth "high-single-digit billions." Third, the AI cloud business itself. If Nebius captures even 2% to 3% of a trillion-dollar AI cloud market, the upside from the current price is substantial. Arkady Volozh's stated ambition is "approximately 10%."

The fund's closing words on the position: "At today's price, we are essentially buying a contracted bare metal business and a portfolio of valuable stakes at a fair price, and getting what could become a very big idea, one of the world's next great AI cloud platforms, with a world-class Founder/CEO and a team of engineers who have succeeded before, at an attractive entry point."

The selloff: what the market is missing

NBIS hit $299.86 on June 22. It is now at $205, down 31% in three weeks. The selloff has three drivers, none of which touch the fundamentals.

The first was a July 1 Bloomberg report suggesting Meta might expand its own cloud and monetize excess data center capacity. The stock dropped 17% in a single day. The market panicked that Meta would compete with Nebius rather than buy from it. But the $27 billion contract is structured with committed capacity, not optional. Meta does not sign $27 billion deals and then walk away. As one analyst noted on X, "Meta doesn't randomly buy $48 billion worth of neocloud contracts if they overbuilt capacity and can cut capex."

The second driver is a broader rotation out of AI infrastructure names. On July 13 alone, the sector saw ALAB down 14%, CRDO down 9%, and MU down 6%. This is sector-wide profit-taking, not a Nebius-specific problem.

The third is technical. The stock is below its 20-day and 50-day moving averages. The Relative Strength Index sits at 31.2, oversold territory. The algorithmic flow shows consistent distribution on the tape. But this is exactly what a flush looks like: institutional hands accumulate on the way down, and the narrative catches up later.

The RSI of 31.2 is worth dwelling on. An RSI below 30 is considered oversold, and 31 is right at the doorstep. The last time NBIS was this washed out on a technical basis, it was in late April at $135. It rallied to $300 in eight weeks.

The LEAPS market agrees

The January 2028 options chain tells a story of conviction. The largest single long-dated call position is the $350 strike, with 17,664 contracts of open interest. At roughly $205 spot, that is a bet on a 71% recovery by early 2028. The $250 strike holds 16,439 contracts. The $410 strike, deep out of the money, added 308 contracts of open interest on July 13 alone. These are not retail YOLO trades at this scale, they are deep-pocketed, long-dated conviction positions.

The real risks

This is not a risk-free story. The company is spending $2.5 billion per quarter on capital expenditure. Even with 60% customer prepayment, the outspend runs to roughly $8 billion to $10 billion annually. The $9.3 billion cash balance covers 2026, but continued expansion at this pace requires continued access to capital markets. If the convertible bond market closes, Nebius would need to dilute shareholders.

Meta and Microsoft are the two anchor customers. Combined, they represent approximately $46 billion of the contract book. If either relationship deteriorates, the impact is enormous. The contracts are legally binding, but no contract can prevent a relationship from souring.

There is a governance question. This is a Dutch company with a Russian founder, listed on NASDAQ. Arkady Volozh controls the company through super-voting shares. If something happens to Volozh, or if geopolitical tensions escalate between the West and Russia, the stock could reprice regardless of the business fundamentals.

Jim Chanos, the noted short-seller, has argued that NBIS trades at roughly 27 times estimated 2030 earnings per share while NVIDIA trades at 8 times the same metric. "Predicting the future is hard," Chanos wrote, "so why pay a premium?" It is a fair question. The bull case relies on Nebius capturing a meaningful share of a market that does not yet exist at scale.

What to watch

Nebius reports Q2 2026 earnings on August 6 before the market opens. The consensus estimate is a loss of $0.64 per share, but the market will be watching revenue growth, ARR trajectory, and the capacity buildout timeline more closely than the bottom line. Any update on the Meta and Microsoft contracts, the Pennsylvania campus, or the ClickHouse stake will move the stock.

The AI cloud thesis is a 5-to-10-year story. The volatility is the price of admission. But the sequence of events over the past 12 months, anchor customer, NVIDIA investment, investment-grade debt, accelerating revenue, expanding margins, has de-risked the hardest questions. The stock is pricing in a failure that the business has not shown any sign of delivering.

More research at bargo.ai/research.