Michael Burry's short position in Nebius Group is underwater after the AI cloud company reported a 454% year-over-year revenue increase to $582.3 million and the stock surged 34% to $259, roughly 22% above his disclosed short price of $212.
While the quarter demonstrated strong demand and solid funding, Burry's fundamental concern—that AI infrastructure companies depreciate hardware too slowly and overstate profits—remains unrefuted: Nebius' depreciation and amortization ($259.7 million, or 45% of revenue) exceeded its adjusted EBITDA, and the company extended the useful life of servers from four to five years.
What happened
Michael Burry disclosed a short position in AI cloud company Nebius Group (NBIS) on August 6 at around $212 per share. Six days later, Nebius reported second-quarter revenue rose 454% year over year to $582.3 million, and the stock jumped about 34% to around $259—roughly 22% above Burry's disclosed short price.
Why it matters
Nebius' AI cloud business grew revenue 514% year over year to $575 million, with an annualized run rate reaching $3.0 billion (up from $1.9 billion at end of March). The company closed four contracts averaging more than $1 billion each, with about 70% of deals including customer prepayments covering 50% to 60% of capital spending. Operating cash flow was $2.2 billion and cash on hand was $8 billion at end of June, addressing concerns about funding for rapid data-center buildouts. However, Burry's core argument—that AI infrastructure companies depreciate chips too slowly, overstating profits—remains difficult to dismiss: depreciation and amortization were $259.7 million (45% of revenue), exceeding adjusted EBITDA of $236.2 million, and Nebius extended useful lives of servers from four to five years.
What to watch
Nebius trades at about 20 times guided full-year revenue of $3.0 billion to $3.4 billion. The company spent $5.7 billion on capital expenditures in the second quarter alone; if depreciated over five years, that would imply more than $1.1 billion of annual depreciation once in service—exceeding current annualized adjusted EBITDA. Shareholders are also funding the buildout through share dilution: the company sold 12.7 million shares through June 30 at an average price of $223.60, raising about $2.8 billion, with 12.3 million shares still available under the program.
On August 6, prominent investor Michael Burry disclosed a short position in Nebius Group at around $212 per share—a bet he characterized as larger than his concurrent short against Oracle. Six days later, on August 12, Nebius reported its second-quarter earnings before market open. The results vindicated the company's execution and confounded Burry's near-term positioning: total revenue rose 454% year over year to $582.3 million, management reaffirmed its full-year guidance, and the stock jumped approximately 34% to around $259, leaving Burry's position underwater by roughly 22%.
The depth of Nebius' operational performance left little room for the traditional demand-side objections to the short case. The company's AI cloud business, which represented the bulk of growth, expanded revenue 514% year over year to $575 million. More striking than the top-line growth was the acceleration in the business's annualized run rate—the trailing month's AI cloud revenue multiplied by twelve. At the end of December, that annualized rate was $1.25 billion. By the end of the first quarter (March), it had grown to $1.9 billion. By the end of the second quarter (June), it had reached $3.0 billion. The deals themselves bore signs of durability: Nebius closed four contracts in the quarter, each averaging more than $1 billion in total value, and approximately 70% of all deals closed in the quarter included customer prepayments that covered 50% to 60% of the related capital spending. In a shareholder letter, founder and CEO Arkady Volozh underscored the strength of the contract book: "We could sell our entire 2027 capacity on these terms today." The company's balance sheet also addressed a traditional worry for any infrastructure company undertaking rapid expansion. Operating cash flow came in at $2.2 billion in the quarter, and Nebius held $8 billion of cash as of June 30.
Yet Burry's critique does not rest primarily on demand weakness—it rests on accounting. In a widely cited post published in November, Burry argued that AI infrastructure stocks systematically depreciate hardware too slowly, extending the useful life of chips that turn over on a two-to-three-year product cycle and thereby overstating accounting profit. He estimated Oracle's earnings could be overstated by approximately 27% by 2028 under such assumptions. His August 6 short disclosure added a second concern: obligations, including lease commitments not yet commenced and purchase commitments, that remain off the balance sheet and unquantified by investors.
Nebius' own second-quarter report validated the difficulty of dismissing this argument. Depreciation and amortization expense totaled $259.7 million—45% of the company's $582.3 million in revenue and more than the company's entire adjusted EBITDA of $236.2 million. On a GAAP basis (under full accounting rules, not adjusted metrics), Nebius recorded a net loss of $190.4 million in the quarter. The company depreciates its servers and network equipment over five years, a useful life it extended from four years at the start of 2026. Nebius cited "usage patterns and utilization commitments" as the reason, a move that aligns precisely with Burry's thesis that companies in the space extend asset lives to suppress depreciation and support reported profits. The forward implication is stark: in the second quarter alone, Nebius spent $5.7 billion on capital expenditures. If that entire amount were depreciated over five years (the company's current policy), it would imply more than $1.1 billion in annual depreciation expense once the equipment is in service. That figure alone exceeds Nebius' current annualized adjusted EBITDA, suggesting that the company's adjusted profitability today depends materially on the extended five-year depreciation schedule.
From a valuation perspective, Nebius does not offer an obvious reprieve either. At $259 per share, the stock trades at approximately 20 times the company's guided 2024 full-year revenue of $3.0 billion to $3.4 billion—a steep multiple even assuming the company hits its targets. Shareholders are also funding the buildout not only in cash but in equity dilution. Through June 30, Nebius sold 12.7 million shares at an average price of $223.60, raising approximately $2.8 billion, with 12.3 million shares still available under its repurchase authorization. This means future capital will arrive in both dollar commitments and shareholder dilution. The quarter thus presents a paradox for Burry's thesis: it refuted the straightforward demand-side short case while leaving the harder—and potentially more damaging—accounting-based concern untouched.
Burry's short thesis rests on two layers: the easier claim, about demand weakness, and the harder claim, about accounting. The quarter substantially undermined the easier argument. Nebius' AI cloud revenue accelerated across three consecutive quarters—$1.25 billion annualized run rate in December, $1.9 billion by end of March, and $3.0 billion by end of June—suggesting not just growth but accelerating momentum. The contracts backing that growth are also fortified: four deals in the quarter averaged over $1 billion each, and roughly 70% included customer prepayments covering half to two-thirds of capital spending. CEO Arkady Volozh's statement—"We could sell our entire 2027 capacity on these terms today"—implies forward revenue is not only contracted but prepaid, removing demand risk for the foreseeable future. The company's cash position ($8 billion) and operating cash flow ($2.2 billion in the quarter) similarly addressed the funding concern that typically haunts hyperscalers in rapid buildout phases.
Burry's harder argument, however, remains standing and harder to dismiss. His core claim, advanced publicly in November, is that AI infrastructure companies stretch the useful lives of chips—treating them as if they'll last five-to-seven years when the product cycle is truly two-to-three years—thereby deferring depreciation expense and inflating accounting profit. The Nebius quarter provides fresh ammunition for that view. Depreciation and amortization hit $259.7 million in Q2, representing 45% of the company's revenue and exceeding its adjusted EBITDA ($236.2 million). On a GAAP basis, Nebius posted a net loss of $190.4 million. The timing of the useful-life extension—from four to five years at the start of 2026, just as growth accelerated—also fits Burry's narrative that companies lengthen lives to smooth reported earnings. Looking forward, Nebius' $5.7 billion in second-quarter capital expenditures, if depreciated over five years, would imply $1.1 billion in annual depreciation once deployed—likely exceeding the company's current adjusted EBITDA run rate once the full asset base is in service. This creates an uncomfortable future state: if Nebius' reported operating profit today relies on stretched depreciation, a normalization of asset lives could reveal a much smaller or negative operating margin underneath.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Google launched the Pixel 11, Pixel 11 Pro, Pixel 11 Pro XL, Pixel 11 Pro Fold, and Pixel Watch 5
Gallium arsenide (GaAs), a semiconductor material, is gaining adoption in short-reach optical links for AI dat…
Antrique Holdings, through its Brunei subsidiaries, unveiled an AI Innovation Platform on August 13, 2026, mar…
Meta announced a $1B "Future Is For Everyone Fund" targeting communities affected by its AI infrastructure pro…
Meta is simultaneously launching open-weight AI models like Muse Spark 1.2 and Muse Glimmer, expanding data-ce…
Blackstone, BlackRock, Apollo, Brookfield, Goldman Sachs, and KKR signed MOUs with Nvidia to deploy at least $…
The AI news that matters, in one minute each morning.
Sign up free