The Machine That Eats Itself

There is a particular kind of irony in watching the technology that was supposed to cure inflation become its primary driver. Two data releases landed Thursday, separated by a few hours and a few floors of abstraction, and together they form a thesis that should unsettle anyone still treating the Fed's rate path as a tame variable in their models.

Micron's fiscal third quarter was not a good quarter. It was a generational restructuring of revenue. The company reported $41.46 billion in the period — up from $9.3 billion a year prior — with adjusted earnings of $25.11 per share against consensus of $20.78. The stock surged 17%. Those are not the metrics of a semiconductor company navigating a cycle. Those are the metrics of a company that has become infrastructure. HBM is not a product segment anymore. It is a toll road, and every hyperscaler on earth is paying the toll.

Then, two hours later, the PCE report arrived. Headline inflation in May ran at 4.1% year-over-year, a three-year high. Core came in at 3.4%. Both in-line with expectations, which is somehow worse — Wall Street fully anticipated this and bought Micron anyway. Personal income and spending both rose 0.7% month-on-month, above estimates. The consumer is not rolling over. The economy grew at 2.1% in Q1, revised up from 1.6%.

Put these two documents next to each other and read them as a system, not as separate data points.

Micron's revenues tripled because the AI industry is consuming memory in quantities that would have seemed fanciful three years ago. That demand has compressed supply so severely that Apple — Apple — announced price increases on MacBooks and iPads because it cannot source enough chips at old prices. AAPL dropped 6% Thursday. Not because the company is in trouble. Because it has become downstream of a supply chain it no longer controls. The iPhone maker is paying Micron's pricing power directly to the consumer, and the consumer is paying it to the CPI.

This is the structural feedback loop that nobody is naming clearly: the AI capex supercycle is contributing to the inflation that constrains the central bank that controls the rates that govern the cost of capital that funds the AI capex supercycle. Circle, meet its own tail.

---

Kevin Warsh has a theory. He has stated — carefully, in the way that lawyers speak when they know the transcript will matter — that supply-shock inflation ought to be looked through when formulating policy. The logic is defensible in academic terms: if prices are rising because of capacity constraints rather than demand excess, raising the cost of capital punishes investment without alleviating the underlying shortage. You do not cure a semiconductor supply bottleneck by making it more expensive to build fabs.

Fine. But the Fed's preferred inflation gauge just printed 4.1% headline, core PCE at 3.3% is now the floor rather than a spike, and markets are pricing roughly a 70% probability of a rate hike by September. The dot plot — minus Warsh's own dot, which he pointedly declined to submit — shows nine of eighteen officials anticipating at least one hike this year. The median funds rate projection moved to 3.8% by year-end, a 16-basis-point gap above the current floor. Warsh's theory has to survive contact with that institutional arithmetic, and the arithmetic is not cooperating.

What Thursday told us is that Warsh's thesis is in its most precarious position since he took the chair. The Hormuz reopening has pushed Brent back toward $73, down more than 35% from April's $114 peak — the energy-price shock that caused much of the headline surge is unwinding. If June PCE comes in soft, the hawks on the FOMC look premature. If it does not, the argument that this inflation was entirely supply-driven becomes progressively harder to sustain while core PCE is sitting above 3% with a resilient labor market behind it — 172,000 payrolls in May, unemployment steady at 4.3%.

The Fed is standing at a fork in the road with a blindfold on, and the two paths diverge further with every data release.

---

The market's reaction to all of this was illuminating in its incoherence. The Nasdaq fell for a fourth consecutive session — its longest losing streak since February — while the Dow closed at a fresh intraday all-time high. Caterpillar jumped 6%. UnitedHealth rose. Industrials, financials, and healthcare led, while the $41 billion memory company dragged the index lower because Apple's price hikes created inflation fears that cut into the multiple of every growth name.

Gold fell below $4,000 for the first time in seven months. In any other environment, a 4.1% PCE print, rising hike probability, and equity volatility spiking above 19 on the VIX would be a textbook bid for bullion. Instead gold is being liquidated into the dollar, which is up more than 2.5% in June and sitting above 101.50 on the DXY. Warsh's silence has somehow been more bullish for the dollar than any clear hawkish signal would have been. Markets filled the ambiguity with their most restrictive assumption.

SpaceX priced its $25 billion debt deal this week and reportedly met a skeptical audience in the bond market, absorbing the paper only at a premium. The equity, trading as SPCX near its IPO price of $150, has swung wildly. For something with no directly comparable credit, paying a spread premium is just the market doing its job — but the episode illustrates the financing environment: capital is not cheap, bond buyers are selective, and the moment you ask for $25 billion you find out exactly how many of your believers are long-only equity tourists who do not control a credit book.

---

The Micron quarter will be cited for months as vindication of the HBM thesis, the AI capex supercycle, the case for memory as the new oil. All of that is warranted. The numbers are extraordinary.

But extraordinary chip demand does not exist in a macroeconomic vacuum. It flows through Nvidia's compute orders, through TSMC's CoWoS packaging, through HBM supply from SK Hynix and Micron, and eventually into the chips Apple puts in a MacBook that now costs more because the hyperscalers bought up the margin first. The cost of building the AI economy is not borne solely by the companies building it. A portion of that cost is being distributed downstream through consumer prices, and the PCE is keeping score.

Warsh wants to look through supply-shock inflation. The supply shock is, in part, artificial intelligence itself. That is not a comfortable position for a central banker who also believes AI will eventually be disinflationary through productivity. The word "eventually" is doing a great deal of work in that sentence. And the Fed's mandate says nothing about patience.