The app for independent voices
My good friend Daniel J. Arbess nails the core thesis here. Apple is not competing in the AI capex arms race. It is taxing it.
The hyperscalers are spending $630 to $690 billion this year building the infrastructure. Apple is spending $14 billion to sit at the end of the pipe and collect tolls from 2.5 billion verified endpoints. That is not a cheap bet. That is the most capital-efficient positioning in technology history.
But here is the angle I think even Daniel’s excellent WSJ piece understates.
The same AI infrastructure supercycle that makes Apple the distribution chokepoint is simultaneously destroying Apple’s hardware competitors through a mechanism nobody is modeling correctly.
AI demand has locked approximately 20% of global DRAM wafer capacity via forward contracts. TrendForce projects LPDDR5X contract prices surged roughly 90% quarter-over-quarter in Q1 2026. The steepest increase on record. Memory costs have effectively tripled over two quarters.
Apple absorbs this at 48.2% gross margins with Services at 76.5%. Flesh wound.
Mid-range Android OEMs absorb this on single-digit margins. Extinction event.
Asus has halted smartphone development. Meizu has reportedly ceased operations. Transsion’s profit collapsed 53%. OPPO, Vivo, and Honor enacted 15 to 25% coordinated price hikes across China.
And what did Apple do on the exact same week? Launched iPhone 17e at $599 with doubled storage and upgraded specs. Held the line while every competitor raised prices.
This is the part Wall Street misses entirely. The AI capex boom that Daniel correctly identifies as making Apple the “price-maker” is simultaneously starving the memory supply chain of consumer-grade silicon, which is destroying the competitive set that Apple needs to disappear for its installed base to keep expanding.
The margin headwind and the competitive destruction are not separate effects. They are the same physics. And the net terminal value impact is positive.
IDC projects smartphone shipments down 12.9% this year. The sub-$100 segment declared permanently uneconomical. Market share projected to concentrate among the largest OEMs.
Apple does not need to win the AI training race. Apple needs to own the endpoints where AI gets delivered. It already does. And every quarter that the memory crisis persists, it owns more of them.
Read Daniel’s full WSJ piece. Then read my deep-dive synthesis on Substack for the mechanism underneath.
