The AI Equity Story Sorts Itself By Q1 2027
Michael Burry's largest disclosed position right now is roughly $912 million in Palantir puts. The reasoning isn't that AI is fake. The reasoning is that the market hasn't sorted what's real from what's narrative.
By the end of Q1 2027, the AI-narrative segment of the public equity market splits into two clearly separable groups: companies whose AI revenue can defend the multiple, and companies whose AI line is still aspirational. The second group re-rates downward by enough to matter. Several names in the second group don't survive the re-rating intact.
That's the prediction. What follows is the reasoning, the evidence, and what would have to happen to make me wrong.
What the bear position actually looks like
The shorthand version of Michael Burry's current position is "shorting AI." The actual filing is more precise. Scion Asset Management's most recent 13F discloses roughly $912 million in notional Palantir puts and around $186 million in Nvidia puts. Bearish put options also sit on Oracle, the iShares Semiconductor ETF, and the Invesco QQQ Trust, with expiries reaching into 2027. The Palantir position is structured around the view that the company's fair value is "low double digits at best," a fraction of where the stock has traded for most of 2026.
You can read the position three ways. It isn't a wholesale rejection of AI as a technology. It's a wager that a specific subset of AI-narrative equities is mispriced and will re-rate before the option contracts expire. And Burry's track record on bubble TIMING is mixed, early on housing for years before the call paid, so "Burry is short" doesn't translate to "Burry is right." What's worth taking from the position is the specificity of what he chose to short.
The CEO saying the quiet part
Anthropic's Dario Amodei spent the past two weeks telling interviewers some software companies will "completely go bust" if they don't adapt to AI. The line is being treated as bold. It's also self-interested. Amodei runs a frontier lab whose addressable market is the same set of incumbent software vendors he's predicting will fail. His forecast about which companies die is structurally indistinguishable from a sales narrative about who his lab's customers will be in 2028.
That doesn't make the forecast wrong. It means I won't repeat Amodei's framing without flagging the incentive. The underlying observation is more useful than the soundbite.
Where the bear case has teeth
The companies most exposed aren't the ones that ignored AI. They're the ones whose moat was the workflow friction that AI happens to dissolve. The shape is recognizable to anyone evaluating mid-market software this year. Imagine a $300M-revenue vertical workflow vendor whose product is essentially a structured form with reminders, sold into healthcare back-office or compliance ops. The product's value proposition is "this turns a multi-step manual process into a guided path." A single-agent approach that does the same work end-to-end with looser input constraints shrinks the moat to nothing.
That category trades at multiples that assume the AI strategy will defend the franchise. Their earnings transcripts say "AI-enabled." If you're sitting on the equity of one of these companies, the AI line in the next quarterly is going to be asked to show revenue, not initiative. The companies that can show revenue defend the multiple. The companies that can't will discover that the AI-narrative premium was always a loan, not a gift.
The bull case
The conventional bull-case response is that capex stories take three to four years to work through and that any re-rating is overstated for any single earnings cycle. There are two reasons that argument worked through 2024 and 2025 and doesn't work the same way now.
The comparison set has sorted. When every AI-narrative company sat in the same bucket, "we're early" was a category defense. The category has now produced enough quarters of attribution data that analysts can separate companies with material AI revenue from companies with aspirational AI revenue. The defense breaks when the comparison is no longer murky.
And the buy-side has been running this math for over a year. What's coming isn't a sudden shift in sentiment. It's the asking, on earnings calls, of questions the buy-side has been writing notes on. CFOs at exposed companies will be asked to defend the AI line with operational specifics. Some will. Others won't.
What would make me wrong
Three specific outcomes:
- A second-half 2026 acceleration in genuine enterprise AI revenue capture that lifts the exposed segment's AI line into the defensible range before Q1 earnings season.
- A capital-markets event large enough to suspend ROI scrutiny for another cycle: a major model breakthrough, an unexpected merger wave, or a rate cut sized to revalue the multiple risk.
- A Palantir-specific revenue print so strong that Burry's most prominent short target's stock holds, removing the most visible test case for the thesis.
If any of those land, the bifurcation pushes to mid-2027 or later. None are visible yet. The next two earnings cycles are the test.
The AI equity story has been about who's investing. It's about to be about who's earning.
Sources
Want to talk about this?
Get in touchMore on AI
The Ratepayer Subsidy Powering AI
Maryland just asked FERC to stop a $2B grid bill from landing on residential ratepayers to fund out-of-state AI data centers. It's the AI infrastructure story enterprise buyers should be tracking and aren't.
The AI Dividend Picks a Direction
Cloudflare just put AI on the receipt for 1,100 layoffs at record revenue. That makes the productivity story real and the 'AI creates more jobs' argument harder to sustain. The dividend has a direction.
The Speed Mismatch on AI
CNN says Wall Street is finally pressing on AI returns. The NYT op-ed page says federal policy isn't moving fast enough on AI risk. Two pieces, two sections of the paper, opposite framings, same underlying story. The systems around AI weren't built to operate at AI's clock speed.