The Market Was Running on Vibes. The Vibes Just Called In Sick.

Wednesday night, Micron drops earnings. Revenue crushes. Guidance crushes. HBM demand is not a fabrication, not a Wall Street ghost story — it's $9.7 billion a quarter of real silicon shipped into real data centers eating real power. The stock rips 16% in after-hours like it has somewhere to be. Good. Fine. The AI capex supercycle has a heartbeat, and you can measure it in DRAM shipments.

Then Thursday opens and the Nasdaq just... sits there. Doesn't celebrate. Industrials run — up 2.19% on the session. Healthcare runs. The equal-weight S&P quietly posts gains while the cap-weight version bleeds from the megacap anchors. MU's 16% surge barely registers as a footnote. You could feel the rotation happening in real time, like watching a crowd quietly shuffle toward the exits before anyone's officially yelled fire.

Then Friday morning, the New York Times publishes a single article, and the whole thing comes apart.

OpenAI is considering delaying its IPO to 2027. The reason cited — almost comically on the nose — is SpaceX. Specifically, what happened after SpaceX's IPO: soared 60% above its offer price, then gave back nearly all of it. Which was supposed to be the proof of concept. The validation that the AI trade could convert narrative into durable public market equity. If SpaceX, the most hyped non-AI IPO in a decade, can't hold its gains, Sam Altman is apparently not eager to find out what happens when OpenAI tries.

Micron, which had just told the world that HBM supply was tight and getting tighter, gave back 5% by lunch. AMD down 2%. Intel down 2%. SoftBank — which owns a massive OpenAI position and had been pricing that position in at roughly the curvature of the moon — dropped 13% in Tokyo and dragged the Nikkei down 4.2%. The Japanese benchmark lost 1,889 points in a single session. For context, that's the entire year's gain of a normal market. Gone. Because a newspaper reported that a private company might not go public on schedule.

That should tell you everything about the architecture of this trade.

The AI bull case was never really about earnings, not yet. It was about capital market perpetual motion. Hyperscalers spend hundreds of billions. That spending flows into Nvidia, into Micron, into the whole HBM stack, into the picks-and-shovels names. Meanwhile, the AI-native companies — OpenAI, Anthropic, the rest — absorb venture capital at astronomical valuations, promising to eventually IPO and justify the whole chain of upstream spending with downstream revenue. The IPO is the exit valve. It's the moment when "this company will eventually make money" becomes "this company is now priced by a market that needs to believe it will make money, forever, at increasing rates."

Remove that exit valve, and you don't just lose the IPO premium. You start asking whether the upstream spending is itself rational. JPMorgan said it plainly on Friday: the delay raises questions about "sustainability of their infrastructure spending given the delay in funding from the capital markets." That's the anxiety now. Not whether AI is real — Micron's numbers settled that. But whether the equity market's pricing of every company adjacent to AI is contingent on a narrative loop that requires fresh money to keep validating itself. An ouroboros denominated in B100s.

And then, twenty-four hours earlier, the PCE report.

Headline PCE at 4.1% year-over-year — the highest since April 2023, the fourth consecutive month of acceleration. Core PCE at 3.4%, a tick above the 3.3% consensus. The energy shock from the Iran war is the primary driver, and yes, oil has since pulled back — Brent fell another 3% Friday to around $73 as Hormuz ceasefire talks normalized shipping — so there's a plausible argument that May was the inflation peak. Apple announced price increases on Macs and iPads last week due to the memory shortage, which means AI infrastructure bottlenecks are now showing up in consumer electronics prices. The AI buildout isn't just bidding up HBM futures. It's making your laptop more expensive.

The Warsh Fed held rates at 3.50%–3.75% at its June meeting, but nine members signaled at least one hike before year-end. Deutsche Bank wants two. The Fed's own projections now see PCE ending the year at 3.6% — versus 2.7% in their prior forecast. The Dallas Fed's trimmed mean PCE tells a quieter story: 2.4% over the twelve months through May, which is actually close to target, suggesting the headline spike is concentrated in a handful of energy-contaminated categories. Warsh has argued that supply-shock inflation should be looked through. The problem is he can't say that loudly. Any public dovishness right now and the bond market reads it as a green light, yields fall, financial conditions ease, and inflation becomes self-sustaining through a different channel. So he holds. He signals. He leaves optionality on the table. September rate hike bets stay live.

Meanwhile consumer sentiment sits at 48.9, near historic lows. The S&P 500's cap-weight and equal-weight correlation is at its lowest since 2003. The market is not one thing right now. It's two markets sharing an index: a narrow strip of megacap AI-adjacent names under genuine fundamental and valuation pressure, and a much wider universe of industrials, healthcare, and rate-sensitive names quietly benefiting from yield softness and rotation flows. The Dow hit an intraday record Thursday while the Nasdaq posted its fifth straight losing session. That divergence is not noise.

Here's what I keep coming back to. Micron's earnings were genuinely good. HBM demand is not going away. The hyperscalers are not going to suddenly stop building. But the market has been pricing AI equities not on the profits those companies make today, but on the profits the entire AI stack will make once the capital market flywheel reaches escape velocity — once OpenAI goes public at $300 billion and validates everything above it in the chain. OpenAI at an ~$852 billion pre-money valuation and 38x sales is not a company. It's a thesis. And a thesis delayed is a thesis that has to survive on its own merits in the meantime.

The market is now finding out if it can.

The rotation will continue until someone answers the AI spending question with a balance sheet, not a press release.