I’ve Designed AI Data Centers for 15 Years. Burry’s Nebius Short Exposes a Dirty Secret

Posted on by

(SeaPRwire) –   By: Ethan Gallagher
Michael Burry didn’t just short a stock last month. He called out the AI infrastructure sector’s biggest delusion. Nebius Group’s 13.29% single-day drop after his disclosure says more about market fragility than Burry’s track record. I sat in on a dinner with three data center procurement leads two weeks before the news broke. Every single one was complaining about GPU lead times stretching past 12 months. None of them could tell me when their capital spending would start generating real, recurring returns. That’s the same vibe hanging over Nebius right now. I’ve seen this cycle before, back during the 2021 cloud boom. Companies were raising billions to build data centers, only to get crushed when demand cooled 18 months later. The stock fell 31.1% in July total, yet still trades up 125% for 2026. That disconnect is the problem. Investors have been pricing AI cloud players for infinite growth, ignoring the physical and capital limits of scaling compute. Burry’s bet is a wake-up call no one in Silicon Valley wants to answer right now. Too many people are confusing a gold rush with owning the mine.

The official Nebius narrative checks all the hypergrowth boxes investors crave right now. Q1 revenue jumped from roughly $50 million a year prior to $400 million. That’s 700% year-over-year growth, a number most enterprise software firms can only dream of hitting at scale. The company attributes that growth entirely to surging demand for AI compute capacity, which aligns with every major tech firm’s AI spending plans. Contracted power capacity has grown from a minimum of 1 GW last August to over 4 GW today. For context, 1 GW of data center capacity can support hundreds of thousands of high-end GPUs running training workloads. Scaling that fast in 12 months is a massive logistical feat, even for seasoned infrastructure teams. The company secured 1.2 GW of power and land for a new AI factory in Pennsylvania, adding to its growing pipeline of buildouts. It raised $6.3 billion in Q1 alone, including a $2 billion equity investment from Nvidia. That Nvidia backing is seen by many bulls as a major vote of confidence from the most important company in the AI chip space. Analysts at Simply Wall St put fair value at $245.43, implying 22.6% undervaluation at recent prices. That estimate leans heavily on the assumption that Nebius can sustain hyper-growth in AI compute infrastructure for years to come. The stock closed recently at $187.97, with a 52-week range of $62.01 to $299.86 and a market cap around $47 billion. On paper, it looks like the next big AI infrastructure winner.

Dig past the press releases, and the math gets far less flattering. Nebius spent $2.5 billion in Q1 alone, most of it on GPUs and related hardware. That’s six times its quarterly revenue, with no sign of spending slowing down. I’ve talked to infrastructure CFOs who say 2x revenue capital spend is the red line for sustainable growth. Nebius is three times past that line, and it’s still ramping up buildouts. Its contracted power capacity has hit 4 GW, but contracted power doesn’t equal revenue-generating capacity. It takes months to install GPUs, configure networks, and onboard paying customers. A lot of that capacity will sit idle for quarters before it starts pulling in revenue. Mid-July, the company raised another $775 million in debt to fund expansion. That’s on top of the $6.3 billion it raised in Q1, which included $4.3 billion in convertible securities. Convertible debt might look cheap when interest rates are high, but it dilutes shareholders heavily if the stock stays elevated when the notes mature. Its P/E ratio sits at 66.1x, more than double the U.S. software industry average of 31.4x and above peer averages of 39.4x. That means investors are paying a massive premium for growth, even compared to other high-flying tech stocks. The stock trades at roughly 14 times forward sales. For comparison, most mature cloud infrastructure players trade at 3 to 5 times forward sales. Management has flagged rising co-location costs, operating lease expenses, and recruitment as persistent cost drivers. These aren’t one-time startup costs. They’re recurring expenses that will keep climbing as the company adds more data centers and hires more staff. Any slowdown in customer cloud capacity commitments would put immediate pressure on the stock. We already saw a preview of that with the 13.29% drop on Burry’s short news. The market is skittish, and any miss on growth metrics will trigger a bigger selloff. Q2 earnings, due Wednesday, Aug. 12, will be the first major test of that risk. Investors are watching for updates on compute capacity growth, customer pipeline, and whether Q1 cost pressures have eased. Even a small miss on cost guidance could send the stock tumbling further from its 52-week high of $299.86.

The AI compute supply chain still funnels most of its value straight to chip makers. Nebius and its peers burn billions on GPUs just to stay competitive, while Nvidia locks in both equity stakes and guaranteed long-term sales. Smaller infrastructure players will run out of capital first when AI demand cools, and the market will consolidate around a handful of players with the deepest pockets.

Author bio: Ethan Gallagher, a Silicon Valley hardware architect with 15 years of experience in AI data center infrastructure strategy.