AINEWS
Apple's margin slipped on memory costs, but the real bottleneck is at the foundry line—the chip fabrication tier (3nm and below) that builds Apple's own silicon—where Amazon is positioned to clear it and Apple is not.
Apple's stock fell roughly 8% the week of its fiscal Q3. Andy Jassy's Amazon climbed between 11% and 15% on the same supply chain the next day. Both CEOs named the same upstream cost line on their July 30 and 31 calls. The divergence is not a company story. It is a supply-chain story about where, exactly, the constraint now lives.
Apple's June quarter, the period ended June 27, delivered $109.4 billion in revenue, up 16% year over year, with diluted EPS of $2.02, up 29% (Apple press release). iPhone set a June-quarter record at $54.25 billion, and Mac posted its best June quarter ever at $10.35 billion. On the surface, a clean beat.
Beneath it, the Apple Q3 10-Q shows reported gross margin at 50.1%, with roughly two points of favorable tariff-refund benefit. Strip that out and the walk from March's 49.3% to June's 48.1% is what management actually has to explain. On the Q3 earnings call, CFO Parekh said memory cost explained the sequential gross-margin decline.
Memory is the cost line most earnings coverage has been carrying. The binding constraint sits one layer up, at the foundry line that builds Apple's own silicon.
Tim Cook, on the same call, said availability of advanced process nodes is constrained. He framed it as a demand-forecast problem rather than a classical supply problem. iPhone and Mac are selling well above plan; the chain cannot flex fast enough to absorb a forecast error. Apple's September-quarter guide of 9% to 11% growth, against roughly 12% consensus, embeds the same constraint: about 2.5 points of FX, plus a sequential increase in supply tightness.
Memory is a six-month fab ramp. Advanced-node foundry capacity is a multi-year build, with reservation lists locked in by 2023 and 2024. That is the layer the "memory shortage" narrative misses.
Amazon's quarter sits on the same supply chain and tells the other side of the trade. AWS revenue hit $42.2 billion, up 37% year over year, the fastest growth in more than four years (memeburn recap). Adjusted EPS of $1.97 beat the $1.82 consensus. On the call, Jassy said Amazon's AI business and its chip business are each pulling in more than $25 billion a year.
Amazon raised 2026 cash capex from about $200 billion to about $220 billion. Jassy explicitly tied the increase to higher memory cost.
This is where the two stories fork. Amazon is buying the bottleneck. Most of that $220 billion lands on data-center shells, accelerators, and the high-bandwidth memory those accelerators need. AWS is positioned upstream of the constraint, with capital commitments that pre-empt the same memory and advanced-node lines Apple competes for as a buyer. When memory tightens further, the unit cost of a token on AWS rises, but the customer's effective bill still falls because the constraint is being cleared with internal capex rather than passed through in full.
Apple is the buyer on the other side. The Fortune recap frames Cook's call as his last before John Ternus takes the CEO seat on September 1. That timing is reported across the recaps; Apple has not published a formal transition memo in the materials captured here, and the date should be checked against the company's own announcement before being treated as final. The supply story is the one Ternus inherits: the same chain, the same cost line, and a position further from the cure.
The portable check for the next four to six quarters is short. Memory spot prices will rise, fall, and rise again. The question is whether advanced-node wafer capacity has actually expanded underneath. If it has not, Apple stays margin-constrained and Amazon's capex moat widens. If it has, the setup flips, and the next iPhone cycle starts to look like a margin-recovery story rather than a memory-cost story.
The wire copy got the cost line right. It got the layer wrong.