AI Infrastructure Selloff Looks Like Rotation, Not Collapse

The action in the megacap and AI complex backs that up. Nvidia has slid to $206.84 from a recent high above $235, even after trading well above its 200-day moving average and holding above its 50-day line for much of the summer. Microsoft has fallen to $381.70 from more than $460 in early June and now sits below both its 50-day and 200-day averages. AMD, meanwhile, has dropped to $521.95 from a June peak near $580 after a parabolic run. These are not charts telling you the AI story is over. They are charts telling you positioning got too crowded, too fast.

The market’s real mistake is to treat every rise in AI capital spending as a margin threat instead of the cost of owning the next utility layer of the economy. Alphabet can beat cloud expectations and still get sold if investors balk at the spending curve. Salesforce can pivot into AI and still go nowhere if the payoff is not immediate. That is short-termism, not analysis. The money is flowing to the picks-and-shovels layer: chips, memory, networking, power, cooling and data-center buildout.

We believe that is where the asymmetry sits now. Super Micro Computer’s 20% surge on record AI-related orders is the cleaner signal than the hand-wringing around hyperscaler capex. The same is true of AMD’s 10-Q, which showed data-center revenue jumping 57% to $5.8 billion in the quarter. Those numbers say the buildout is still accelerating, even if the stock market is temporarily punishing the suppliers and beneficiaries alike.

The pullback also creates a more interesting opportunity set than the leaders alone. If investors are de-risking the obvious names, the next leg of the trade can come from second-order beneficiaries: Arista Networks, which sits inside the AI networking bottleneck; Micron, which benefits from memory tightness and accelerated server demand; and the broader semiconductor and infrastructure complex through ETFs such as SOXX and SMH. Microsoft remains a core long-term AI platform, but the better entry points often come when even the highest-quality compounders are forced lower by emotion.

This is why the selloff matters economically. AI is not just another software cycle; it is a capital-intensive infrastructure wave that pushes through chips, electricity, cooling, optical gear and cloud contracts. Those capex dollars do not disappear when sentiment turns. They get repriced, redistributed and, eventually, justified by growth. The near-term losers are momentum chasers and late longs. The longer-term winners are investors who use fear to accumulate the toll roads of the AI economy.

The next catalyst is simple: as the panic fades, the market will go back to asking which companies are actually monetizing the buildout and which ones are just talking about it. That is when the best returns usually begin. I believe investors should be buying the AI infrastructure dip, not the narrative on the way down.