Michael Burry is shorting Nebius Group (NASDAQ: NBIS).
If you’re like most investors, that sentence is enough to make you sell.
Burry is the investor chronicled in The Big Short for betting against the entire U.S. housing market before the financial crisis of 2008. So when he takes a bearish position against what has been one of the hottest AI infrastructure stocks in the market, investors tend to react. (For how his public calls have fared since 2017, see Burry’s broader prediction track record.)
On Aug. 20th, 2026, Burry revealed that he was short Nebius at approximately $212 per share. This was part of a wider short position against capital-intensive AI infrastructure stocks. Burry’s thesis was centered around extreme infrastructure expenditures, levels of debt, execution risk, and — most importantly — how GPUs are depreciated for accounting purposes.
Since then, Nebius has traded down to about $187.97. In other words, Burry’s short trade is already profitable.
However, this is where we have to ask two very different questions.
Can Michael Burry make money shorting Nebius?
Absolutely.
Does that mean long-term investors should follow Michael Burry and short NBIS?
I think that could turn out to be a very expensive mistake.
It’s not crazy to worry about these things. Nebius is capital intensive, it trades at valuations that assume incredible growth, and it needs to spend billions laying down infrastructure before much of the revenue from that infrastructure comes in.
But I think the bear case misses one thing far more important: Nebius is quickly evolving from a speculative AI-infrastructure play into a company with tens of billions of contracted demand from Microsoft and Meta, backing from NVIDIA, improving operating economics, and a more nuanced financing approach.
That changes things.
What Exactly Is Michael Burry’s Nebius Short Thesis?
The specific case is about depreciation, but Burry’s critique of the AI infrastructure boom is rooted in depreciation as well.
You’ve got AI companies plowing billions into GPUs, servers, networking equipment etc. The question is on the accounting side is how many years will those assets be economically useful for the company.
Let’s say a company purchases a server for $5 million. If it depreciates that asset over 5 years rather than 3, its annual depreciation expense will look a lot smaller.
The company hasn’t spent any less cash, but earnings can look better.
Burry has made the case more generally that companies across technology are extending useful lives, which he believes understates depreciation expense for assets like AI hardware that are rapidly improving.
And with Nebius we have something investors can point to.
Beginning in Q1 2026, Nebius lengthened its estimated useful life for server and networking equipment from 4 years to 5 years. This change lowered depreciation expense by about $43.1 million in Q1. Net income was about $41.6 million higher as a result.
So please don’t @ me and tell me you know why Burry’s argument is stupid.
I don’t think you should dismiss it out of hand. It’s reasonable and deserves consideration.
I do think investors may be drawing the wrong conclusion from it though.
A Five-Year Depreciation Schedule Does Not Mean a GPU Becomes Worthless After Three Years
Treating technological obsolescence as economic obsolescence ignores the fact that they are two separate things.
If NVIDIA releases a GPU that is orders of magnitude faster tomorrow, it doesn’t automatically render previous generations worthless from a commercial perspective.
There’s a reason NVIDIA’s product roadmaps span multiple generations — older GPUs will likely be outperformed for frontier-model training, but they can still be economically useful for inference, fine-tuning, development, smaller models, scientific computing, and other workloads where customers are willing to trade raw performance for a lower price.
This is especially true now that the AI industry has moved past merely training larger and larger models to running those models at scale.
Nebius itself is positioning its offerings around the entire AI lifecycle. Training. Deployment. Inference. Nebius Cloud bundles GPU compute with storage, networking, and software. You’re not simply renting individual GPUs.
That matters because the commercial value of Nebius’s tech isn’t solely tied to whether NVIDIA’s latest GPU is the fastest chip available.
The more relevant question is:
Can Nebius generate enough revenue and cash flow from the hardware during its productive life to earn an attractive return on the capital invested?
And the early numbers are becoming increasingly difficult to ignore.
Nebius Revenue Is Growing at an Extraordinary Rate
Neebius provided $399 million of first quarter 2026 revenue. A year ago it provided $50.9 million.
Revenue grew by 684% year-over-year.
Crucially, nearly all of this growth was driven by its core AI infrastructure business. Nebius AI Cloud revenue grew from $41.4 million to $389.7 million, up approximately 841%.
This is quickly becoming a business generating hundreds of millions of dollars of enterprise AI infrastructure revenue. It’s not a business generating tens of millions of dollars of experimental revenue.
The infrastructure Nebius has been aggressively acquiring and building over the past few years is starting to come online and produce material revenue.
There are even hints of operating leverage.
The cost of revenue was 49% of total revenue in Q1 2025. One year later, the cost of revenue ratio dropped to approximately 26% as utilization and scale began to kick in.
Nebius AI Cloud also produced $174 million of adjusted EBITDA in the first quarter of 2026 compared to a negative $27.4 million adjusted EBITDA a year ago.
Depreciation is important.
But utilization is important as well.
If Nebius can rapidly deploy and utilize this infrastructure, there are still attractive economics to be generated from filling this capacity and producing healthy revenue per GPU, even with somewhat conservative depreciation rates.
The Biggest Problem With the Short Thesis: More Than $40 Billion of Contracted Revenue
This is where I think the Nebius bear case gets much harder.
Unlike a traditional data center company, Nebius is not just building billions of dollars of AI infrastructure and hoping customers eventually show up.
Big tech companies are already committing billions of dollars to use that infrastructure.
In September 2025, Nebius and Microsoft signed a multi-year agreement for AI infrastructure. The deal is worth approximately $17.4 billion through 2031, but could total approximately $19.4 billion if Microsoft exercises options to acquire additional services and capacity.
Then Meta came along.
In March 2026, Nebius announced it signed another five-year AI infrastructure agreement with Meta.
The details of the deal are particularly interesting.
Meta is agreeing to purchase $12 billion of dedicated capacity that would begin delivering compute power in early 2027. Meta is also agreeing to purchase available compute capacity across certain future Nebius clusters. This element of the agreement could be worth an additional $15 billion over five years.
Here is the key sentence from the SEC filing: To the extent that relevant capacity covered by the foregoing arrangement is not sold to other customers, Meta has committed and shall be required to purchase such unsold capacity for the applicable period under the terms of the order.
By July 2026, Nebius disclosed it had over $40 billion of extra contracted revenue commitments from investment-grade customers such as Microsoft and Meta.
Again, there is still tremendous execution risk. Nebius still needs to build the infrastructure on time. Nebius still needs to secure GPUs, power, data centers and financing. Nebius still needs to earn a reasonable return on that infrastructure.
The difference is that customers have already committed to using a significant amount of that infrastructure through contractual agreements.
Blindly building speculative data centers and hoping customers eventually come is much riskier.
Microsoft and Meta Are Also Helping Finance the Expansion
One of the strongest bearish arguments against Nebius is simple:
Look at how much capital this company needs.
And bears are correct.
Nebius invested roughly $2.47 billion into property, equipment and intangible assets just in Q1 2026, mostly toward GPUs, associated hardware and expanding its data center footprint.
$2.47 billion is massive compared to existing top line.
But looking at capex alone without understanding how the capex is being funded is only part of the story. In Q1 2026, Nebius generated roughly $3.2 billion of customer advances. On March 31st, Nebius had roughly $9.3 billion of cash and cash equivalents.
In July, Nebius closed roughly $775 million of senior secured financing against deployed GPU infrastructure and contracted future customer cash flows.
The loan was priced at SOFR + 2.50%. Nebius stated that the financing combined with cash flows from the related customer contract accounted for >100% of capex associated with the GPU deployment.
Could be one of the most significant developments in the whole Nebius thesis.
Nebius is gradually moving from:
Raise corporate capital → buy GPUs → find customers
toward:
Sign long-term customer → deploy infrastructure → use customer cash flows and asset-level financing to fund deployment
Those are very different business models.
If Nebius can repeatedly finance new clusters against contracted revenue rather than funding everything directly from its corporate balance sheet, its growth model becomes significantly more capital efficient.
NVIDIA Putting $2 Billion Into Nebius Should Not Be Ignored
There is another part of the story that I think makes shorting Nebius particularly risky.
NVIDIA itself is a strategic investor.
NVIDIA and Nebius announced a comprehensive partnership on AI factory design, infrastructure deployment, inference technology, fleet management and early access to future NVIDIA architectures in March 2026. As part of the deal, NVIDIA will invest $2 billion in Nebius.
The two companies said the partnership would enable Nebius to deploy over 5 GW of NVIDIA infrastructure by the end of 20×30.
NVIDIA investing in Nebius doesn’t automatically make the stock a sure winner.
NVIDIA can invest for smart strategic reasons even if public market investors overpay for Nebius stock.
However, the partnership is notable for one other reason.
One of the biggest risks any AI cloud company faces is access to leading-edge GPUs at scale. Nebius relationship with NVIDIA gives it access to something few companies can replicate: Intimate integration with the vendor that controls the AI computing ecosystem.
The partnership includes early access to NVIDIA’s future infrastructure architectures Rubin and Vera.
That’s not something any public cloud AI vendor can say about NVIDIA and it could become a key competitive advantage.
Burry Could Be Right About the Stock and Wrong About the Company
There is one more subtlety to keep in mind.
An investment can be a great long-term business that makes a wonderful short-term short.
Nebius has been massively volatile.
Since Burry reportedly initiated his short somewhere around $212, NBIS has dipped down to approximately $188.
So he doesn’t necessarily need Nebius to go bankrupt to make a profit.
If he is merely being tactical, a short-term consolidation of valuation may suffice.
That’s exactly why piggybacking off of famous investors once their positions are revealed can be hazardous to your portfolio.
You aren’t aware of their precise risk parameters.
You aren’t aware of their target holding period.
You aren’t aware of how that position may be hedged by other positions.
And you certainly don’t know when they plan on exiting the trade.
Burry may close his Nebius short after a 20%, 30%, or 40% drop while the person blindly following him stays short hoping for failure.
They are two very different trades.
The Real Nebius Bear Case
None of which is to say NBIS is without risk.
Not by any stretch.
For me, the biggest risks aren’t even whether Michael Burry is short the stock.
They center around whether Nebius can hit its infrastructure build-out timelines, whether AI compute costs remain favorable, whether customer concentration reaches unhealthy levels, whether too much debt/financing costs are taken on too quickly, and whether each successive generation of GPU infrastructure can generate acceptable returns on invested capital.
The depreciation issue also warrants continued discussion. Nebius bumped the useful life of their servers/networking equipment from 4 to 5 years back in 2026, which materially lowered depreciation expense. Something investors should keep an eye on rather than act like it doesn’t matter.
Management also continues to spend billions on capex before any of that future contracted revenue is recognized.
All of those risks are legitimate.
They just don’t have anything to do with whether Burry is shorting the stock.
What Would Actually Make Me Change My Bullish View?
This thesis would weaken substantially if several events occurred together.
If Microsoft or Meta materially scale back their commitments, if Nebius misses deployment deadlines repeatedly, if util suffers, if AI Cloud margins come down despite wildly expanding capacity, or if the company has to raise progressively dilutive equity just to maintain growth, I would rethink the story.
Likewise, if legacy generations of GPUs turn out to have absurdly worse economics than management expects, Burry’s depreciation thesis would warrant far more consideration.
Those are quantifiable events.
They matter way more than who the short is.
Why Following Michael Burry Into the Nebius Short Could Be Costly
I think Michael Burry deserves credit.
He has figured out how to short stocks when markets become unhinged from economic reality.
And as I laid out last week, parts of his thesis on Nebius are valid.
The business is insanely capital intensive. It needs NVIDIA hardware to run. Its capital requirements are massive. And yes, extending depreciation schedules on equipment from four years to five really did boost reported earnings in Q1 20×6.
Still, I don’t think investors that focus solely on those risks are paying attention to what’s going on under the hood at Nebius.
Recurring AI Cloud revenue doubled year over year. Nebius’s core AI Cloud business posted positive adjusted EBITDA for the first time. Microsoft & Meta are investing tens of billions of dollars into Nebius infrastructure. NVIDIA just committed $2 billion & entered into a strategic technology partnership with the company. Nebius began financing deployed infrastructure with contracted customer cash flows instead of purely corporate capital.
Does that sound like a company whose business model is falling apart to you?
It sounds like a massive AI infrastructure platform trying to scale exponentially.
Yes, exponential scale creates huge risk.
But it also creates tremendous operating leverage if management can execute.
With Nebius set to report 2Q 20×6 financial results on August 12, 20×6 revenue growth, utilization, margins, capex, financing and deployment will be especially meaningful figures to track. Burry may prove profitable on his short.
He may even be right that the market temporarily valued NBIS too high.
But that does not mean he is right that Nebius is fundamentally broken.
I think if you’re a long-term investor blindly betting against Nebius because Michael Burry is, you run the risk of missing the forest for the trees:
Microsoft wants the compute. Meta wants the compute. NVIDIA wants to build the compute. And Nebius is finding innovative ways to finance building that compute against contracted future cash flows.
If Nebius has continued executing don’t you think the biggest loser could be those that miss one of the fastest-growing AI infrastructure companies as its huge contracted backlog turns into revenue?
Disclosure: This article represents opinion and analysis, not personalized financial advice. Nebius is a highly volatile growth stock and investors should perform their own due diligence before buying, selling or shorting NBIS.
Keep learning: courses on this topic
Self-paced courses to put this article into practice.
Udemy
From Welfare to Wealth: Earn Online & Invest Wisely
If you're living on a low income, receiving benefits, or stuck in survival mode — this course is for you.“Financial Independence on a Low…
View course
Udemy
How to Invest in the Stock Market
A self-paced video course on stock investing, with practical lessons you can apply right away.
View course
Udemy
How To Confidently Invest & Analyze Stocks Like an Expert
Are you tired of feeling overwhelmed by the stock market or unsure if a stock is actually a good investment? This course, "Investing in…
View course
Leave a Comment