The AI capex story has been built on one assumption: OpenAI grows fast enough to pay for the chips it ordered. Today the Wall Street Journal reported that growth isn't there. The CFO told her own board she's worried about covering the bills. By 11 a.m., every name with an OpenAI contract was down at least 5%. By 4 p.m., half of them had bought it back. The supplier reprices first — and then the market argues with itself for the rest of the day.

Different tape than Monday. The S&P 500 closed down about half a percent. The Nasdaq fell almost 1%. The Dow held flat. Small caps held in — the Russell 2000 added 0.4%, and the index is now up 11% on the year, on pace for its best month since 2024.

The Wall Street Journal report at 7:30 this morning lit the fuse. OpenAI missed its goal of hitting a billion weekly active users by end of last year. Revenue came in light. The CFO told leadership she was worried about whether the company can pay for the computing contracts it has already signed. ChatGPT's share of AI web traffic has dropped from 87% a year ago to 65%. Google's Gemini took most of that share.

The market repriced the supply chain first. Oracle led the way down — off more than 7% premarket on the news that its $300 billion, five-year compute deal with OpenAI might be in trouble. The company just raised $50 billion in debt and equity to build the data centers. By the close, Oracle had cut the loss in half, finishing down 3.7%. Same round trip we saw in Microsoft yesterday. Nvidia closed off 3%, AMD off 5%, Arm off 7%, Broadcom off 4%. All four have multibillion-dollar capacity deals with OpenAI signed in October. Now those deals are sitting in front of a customer whose own CFO is worried about cash flow.

Oil ran. Brent crossed $112 a barrel intraday and closed near $111 — about $6 above where it ended Monday. WTI moved above $100 for the first time since the early days of the Hormuz crisis. The trigger was the United Arab Emirates announcing it will leave OPEC effective May 1. The cartel just lost its third-largest producer in the same week the market is pricing higher oil for longer.

The odd one: Bed Bath & Beyond up 28% on a revenue beat — the kind of small-cap pop that doesn't happen unless short positions are piled into a name. GM up 4% after raising 2026 guidance. Illinois Tool Works down 9% on geopolitical pressure ahead of Thursday's earnings. Corning down 10% even after a Q1 beat — investors looked through the print to softer Q2 guidance. Up and down the tape, capital is rotating from AI infrastructure into industrials and small caps that don't depend on data center capex.

Stocks don't price today. They price the next six to twelve months. And what got priced today wasn't OpenAI's miss — it was the math underneath the AI capex cycle.

Here's the chain. For the past eighteen months, every chipmaker, every data-center builder, every power equipment company has been running on the same assumption: the AI workload demand from OpenAI and a handful of other model makers will grow fast enough to fill the new capacity. The forecasts to back that up were enormous. Mag Seven companies are expected to confirm more than $700 billion in capex commitments for 2026 alone when they report this week.

That number works only if the customer can pay. Today the Journal reported that OpenAI's own CFO is asking the question internally. That doesn't mean the AI cycle is over. It means the math underneath the cycle just got tested for the first time in a year.

The supplier reprices first. That's how this kind of repricing always starts. The customer keeps spending — they have to, the chips are already ordered. But the supplier's stock starts pricing in cancellations, write-offs, and a slower growth ramp. By the time the customer formally walks back the order, the supplier is already 30% lower. On Wall Street, this is what they call the order-book reset — when the gap between what was promised and what's getting paid for finally shows up in the share price.

I've seen this before. November 2018. Apple cut its iPhone production forecast on a Friday afternoon. By the following Monday, every chipmaker that supplied iPhones — Skyworks, Lumentum, Qorvo, Cirrus Logic — was down 15% to 25%. Apple itself fell 10% over the next two weeks. But the suppliers fell 35% to 45% over three months. They had built capacity assuming Apple's order book held. When the order book moved, they had no other place to put the chips.

Now layer the oil story on top. Brent at $111 is a different problem than Brent at $90. The Fed's preferred inflation gauge has to fight a $35-a-barrel oil shock at the same moment the AI capex story is getting questioned. Two of the three pillars under this rally — cheap money soon and limitless data center demand — both have something to prove this week.

The forward-looking read: this week's earnings are now the most important the market has seen in three years. Microsoft, Meta, Alphabet, Amazon all report Wednesday. Apple Thursday. The questions analysts will ask have changed in twenty-four hours. It's no longer "are you spending enough on AI?" It's "who's actually paying you for the AI you're building?" If even one of those four guides capex lower or talks about pushed-out revenue, the chip selloff that started today extends. If all four come in clean and confident, the OpenAI story becomes a one-day event.

Three things I'm watching:

Four prints, all between 4:00 p.m. and 4:30 p.m. Eastern, all guiding on 2026 capex. Watch for two specific lines: total capex for 2026 (the Street is at $700 billion combined for the Mag Seven) and any commentary on cloud revenue growth tied to OpenAI. Microsoft is the most exposed because of its Azure-OpenAI revenue line. If Azure decelerates, the chip names that fell today extend lower Thursday morning. If even one company raises capex confidently, the selloff stops there.

The Fed isn't expected to cut. Futures are pricing 100% odds of a hold at this meeting. The presser at 2:30 is the action. Today's move proves something about the market — Conference Board consumer confidence came in at 92.8, the highest reading of 2026, but the tape sold tech anyway. That tells you the rate-cut bid that powered last week is still there underneath, looking for a reason to come back. With Brent now above $111, Powell has more cover to hold than he had on Friday. The question is whether he closes the door on summer cuts entirely.

The Fed's preferred gauge. Consensus is around 2.7% year-over-year for the core reading, which strips out food and energy. With Brent above $111 and the UAE walking out of OPEC, energy is going one direction. The number to watch is the three-month annualized rate under the headline. Below 2.5% gives the Fed cover to cut into the second half. Above 3% locks rates in until at least September.

The supplier always reprices before the customer does. By Thursday morning, we'll know whether today was a repricing or just a sale.

That's it for today. See you tomorrow after the close.