AI Infrastructure Stocks Face a Reality Check as Capex Spending Comes Under Investor Scrutiny

CoreWeave and Supermicro step up to report earnings next, and Intel has upsized a stock sale into a sell-off flagged by The Straits Times. The market is about to demand receipts, and your equity sleeve is collateral damage either way.

The Capex Stress Test

We have been here before: a thematic trade gets so crowded that every incremental data point either confirms or invalidates the thesis. This time the thesis is brutally simple. Hyperscalers keep spending tens of billions on data centers, and a tightly linked supply chain — CoreWeave, Supermicro, and Intel — has to convert that spending into real margins. The Straits Times framed the recent sell-off as a test of "AI hype" rather than the end of the infrastructure story. Read that carefully. The question is no longer whether AI infrastructure gets built. The question is who gets paid for it, and at what multiple.

The Microsoft numbers in the Sahm screen are the cleanest case study. About $171 billion of revenue runs through the United States and $161 billion through other countries, with Intelligent Cloud at $138 billion carrying the AI workload. That revenue base only matters to you if margins survive the buildout, and right now they do. The risk is asymmetric: if capex keeps compounding, the infrastructure layer compounds with it. If the cycle cracks, the same revenue base does not save a multiple priced for perfection.

Cerebras sits at the other end of the spectrum. The company pulls roughly $681 million entirely from semiconductors, is still loss-making, and depends on a handful of large customers plus external borrowing. That is not a balance sheet issue — that is a thesis risk. When Intel upsizes a stock sale into this backdrop, the signal is obvious. The adults are raising liquidity before the window closes.

What You Actually Control

You do not decide whether CoreWeave prints a beat. You do not decide whether Supermicro guides up. But you do decide three things this week.

First, your exposure to the AI infrastructure theme. If it is more than 15% of your equity sleeve, you are running a concentrated bet, not a diversified portfolio. Trim back to a level where a 40% drawdown in the basket does not force your retirement plan to sell at the bottom.

Second, your split between the picks-and-shovels layer and the model layer. The infrastructure layer has contracted revenue and visible order books. The model layer has narrative. We prefer contracted revenue in a scrutiny year.

Third, your cash buffer. Intel raising equity into a sell-off is a signal. So is CoreWeave stepping up to earnings into the same tape. Cash is not a yield drag when the next 90 days carry this much event risk.

The binary choice is straightforward. You can own the AI infrastructure theme at full size and pray the capex curve keeps bending. Or you can size it so the thesis has room to be wrong. One of those is a strategy. The other is a bet.