Burry Re-Enters Oracle and Shorts Nebius-Are AI's Earnings or Moods About to Crack?

Burry is re-exposing himself to two AI-linked names with different downside paths

Oracle: a re-entry after a closed short

Burry is not signaling a broad AI unwind here. He re-entered Oracle after closing a short position earlier this week and said he may re-enter Oracle puts if volatility cools. The setup is straightforward: Oracle is an AI-adjacent name still tied to cloud-spending optimism, and Burry is positioning for another leg down if sentiment cracks again.

Bears will argue the easy damage is over. Bulls will argue a sharp pullback is exactly where mean reversion begins. The bearish case lives as long as investors worry that cloud demand, earnings quality, or both may not hold up cleanly.

Nebius: a direct short into a hot momentum trade

Nebius is the other side of the same coin. Burry established a larger short exposure in Nebius at $211.77 and chose direct equity shorts rather than puts because implied volatility exceeding 100% made options expensive. That suggests the trade is less about calm valuation math and more about a sharply rerated stock with crowded expectations.

Bulls see a rising AI compute winner. Bears see expensive insurance and stretched sentiment. Burry's choice of a straight short, rather than pricey puts, implies he thinks downside risk outweighs rerating risk if expectations wobble.

The shared bearish thread is GPU depreciation and earnings quality

How slower GPU depreciation can distort profits

Burry's core argument is not that AI demand is fake. It is that reported earnings can look cleaner than the hardware economics underneath them. If companies aren't depreciating their GPUs fast enough, quarterly expense stays too low and profit stays too high. That does not require a demand collapse; it only requires the market to question the useful-life assumptions behind compute hardware.

Think of it like a trucking company that assumes a heavy-use vehicle lasts 10 years when it is really wearing out in three. Near-term profits look better than the underlying economics. In AI, revenue can still grow while the market underestimates how quickly compute hardware is consumed.

Oracle is the cleaner accounting test case

Oracle is the cleaner test case because the mechanism can become a material earnings number. Burry estimates Oracle's earnings could be overstated by approximately 26-27% by 2028 if GPU depreciation is too slow. That is large enough to turn headline beats into less impressive results.

The backdrop matters too. Burry points to $176 billion in AI-related capital expenditures across hyperscalers as context for the scale of the problem. When that much spending is tied to hardware that may depreciate faster than companies acknowledge, even a modest change in assumed useful life can move billions through depreciation expense.

Nebius adds financing strain and future-commitment risk

Nebius shifts the debate from accounting timing to contract duration and balance-sheet resilience. The caution here is not just about depreciation; it is about whether the business can carry its commitments cleanly. That lines up with the view that there is too much debt in the setup, paired with revenue forecasts that look aggressive.

The fragility, then, is not only whether future contracts exist, but whether operating performance and financing can support them without masking margin pressure.

What would confirm or break the bearish call

This moves the discussion from theory to positioning. Oracle looks more vulnerable to a fresh earnings disappointment, while Nebius looks more vulnerable to a squeeze if expectations stay hot.

Ownership alone does not guarantee a squeeze

Nebius has 660 institutional owners, including 615 long-only holders and 36 long/short holders. That does not prove a tight short float, but it does show a broad holder base with differing mandates. If sentiment stays enthusiastic, even hesitant shorts can be forced to cover quickly after positive news.

Three signals worth watching

• Price response to good news: If Oracle keeps struggling after decent headlines, or Nebius fades after strong commentary, the market may be signaling weakening confidence before any obvious demand break.
• Depreciation scrutiny: A more candid discussion of GPU useful lives or hardware turnover would support the idea that earnings quality is weaker than reported.
• Sentiment and positioning: Rising optimism, expensive options, and tight coverable float would raise squeeze risk, especially in Nebius.

The thesis can still be right on the mechanism and wrong on timing. If the market absorbs the bad news slowly rather than punishing the stocks immediately, the bearish view may be more of a gradual rerating risk than an imminent trade.