AI Stocks Bounce Back After the Situational Awareness Fire…

The AI stocks at the center of last week’s forced unwind of Situational Awareness are rallying hard. Nebius jumped roughly 4% on Tuesday, SanDisk added 6%, and CoreWeave surged close to 20%, part of a broad snapback across the semiconductor and data center complex.

The instinct is to read the bounce as the all-clear. It is not that simple. A rebound that follows a margin-call liquidation can mean two very different things, and telling them apart is the difference between buying a bottom and buying the same fragile trade that just broke.

What Actually Happened

Situational Awareness, the roughly $20 billion AI fund run by former OpenAI researcher Leopold Aschenbrenner, sold most of its public equity book to Citadel last Wednesday after leveraged positions in AI-hardware names, including Nebius and SanDisk, fell sharply and triggered margin calls. The fund had reportedly returned more than 400% in the year through June, a run built on concentrated, borrowed-money bets on AI compute and infrastructure.

That leverage is the whole story. It amplified gains on the way up and forced selling on the way down, as lenders demanded collateral and the fund had to dump large blocks into a falling market. Citadel, along with Millennium and Jane Street, bid for the book; Ken Griffin’s firm took the majority. A forced seller of that size hitting the same handful of stocks is exactly what drove July’s declines in these names, and its removal is exactly what set up the bounce.

The AI Stocks Bounce Is Real, but Look at the Month

Here is the fact that complicates the bullish read. Tuesday’s gains are large, but they are recovering a fraction of the damage. Nebius jumped nearly 30% at one point during the rebound, its strongest daily move since September 2025, yet remains down about 5% over the past month.

SanDisk rose 6% but is still down roughly 37% on the month, its single-day pops repeatedly failing to dent the larger slide. CoreWeave gained nearly 20% in a session and is still essentially flat over thirty days.

The pattern is a violent snapback inside a larger drawdown, which is precisely what a post-liquidation bounce looks like whether or not it marks a bottom. Nvidia, the sector’s bellwether, tells the sober version: it reclaimed the $200 level and rose about 3% on the day, but is up less than 5% over the month and stalls at $210 resistance, with its real test, August 26 earnings, still ahead.

Wall Street’s consensus target sits near $300, but as FinanceFeeds’ NVDA scenario analysis notes, every published target implies upside, which means the risk the market is least prepared for is a demand disappointment rather than a beat.

The Bull Case: A Clean Flush

The optimistic read is that this was a healthy capitulation. In this view, the July selloff was a leverage problem, not a fundamentals problem: overleveraged holders like Situational Awareness were forced out, the weak hands are gone, and real buyers, Citadel foremost among them, stepped in at better prices.

There is evidence for it. The largest cloud providers are spending more aggressively than ever, and their latest results were read as proof the demand is real. Amazon raised its 2026 capital-expenditure plan to about $220 billion, up from $200 billion, on the back of AWS growth of 37%, its fastest in 18 quarters, with CEO Andy Jassy saying even that spend would not meet all the AI demand he sees. Alphabet lifted its own 2026 guidance to as much as $205 billion, and Microsoft and Meta both guided higher, taking the four largest hyperscalers’ combined 2026 infrastructure spending past $700 billion.

Some analysts read the episode as a necessary reset in which forced selling by overleveraged players cleared the overhang and set up a rebound. When the most sophisticated distressed buyer on Wall Street shows up to take the other side of a fire sale, and the companies funding the buildout keep raising their bets, that is not a trivial signal.

Investor Takeaway

The Bear Case: The Same Trade, Re-Leveraging

The cautious read is that nothing structural has changed. The AI trade is still crowded, still concentrated, and still valuation-stretched, and a fast bounce may simply be the same leveraged positioning rebuilding into the same fragility that just blew up.

That fragility has a name the market started worrying about in July: circular financing. Nvidia holds a roughly $2 billion equity stake in CoreWeave, whose multi-billion-dollar capex program is, as FinanceFeeds’ NVDA scenario analysis laid out, largely a forward purchase of Nvidia chips. It is the kind of vendor-financing loop that looks efficient on the way up and concentrates risk on the way down, where a chipmaker’s revenue quality is tied to the funding health of the customers buying its chips.

The spending itself has become the fault line. In the same week, the market split sharply on who was spending well: Amazon’s stock jumped about 10% on its raised capex because AWS growth justified it, while Meta fell 8% on its own increase because investors could not see the return. Alphabet had set the tone a week earlier, raising guidance and posting its first-ever negative free cash flow quarter, then watching its stock drop 7% the same day. Record AI spending is no longer read as automatic confidence; it is now graded on whether revenue is visibly arriving to justify it.

The clearest warning sits in the one stock that did not join the party. Meta rose on Tuesday but has been the sector’s notable loser, falling sharply on questions over whether its AI spending will generate a return. That divergence matters as infrastructure names bouncing while an application-layer giant gets punished for capex says the market has not resolved the central question of whether the AI buildout pays off. The infrastructure layer may have found a floor, but the layer that has to monetize it has not.

Investor Takeaway

How to Tell Which One This Is

The honest answer is that a single green session cannot distinguish a bottom from a bull trap, and the signals to watch are specific. Whether the infrastructure names hold their gains or give them back over the coming weeks is the first. Nvidia’s August 26 earnings is the second, and the most important, since it is the clearest read on whether data-center demand justifies the spending. And whether the monetization doubt weighing on Meta spreads to the infrastructure names, or resolves, is the third.

For now, the infrastructure layer looks like it has found a floor and the application layer still has to prove the model. The rebound is real. Whether it is a bottom or a pause depends on questions that will take more than one trading day, or one fire sale, to answer.