Amazon raised its 2026 AI spending plan to $220 billion because of memory prices. Apple's margins are taking the same hit from the other side.

Two of the biggest companies in AI are being squeezed by the same input cost, from opposite directions. Amazon told investors this week it now expects to spend roughly $220 billion in cash capital expenditure in 2026, up from the roughly $200 billion it had guided to previously — a $20 billion increase CEO Andy Jassy attributed directly to the rising cost of the memory chips its AI data centers run on. Days earlier, Apple's finance chief told analysts that memory pricing alone explained more than the entirety of the gross-margin decline the company is projecting for its next quarter. Both numbers trace back to the same shortage: AI accelerators have pulled so much DRAM and NAND manufacturing capacity toward high-bandwidth memory that everyone else buying conventional memory chips is now bidding against the AI buildout itself.

The $20 billion Amazon didn't plan on

Amazon's underlying quarter was strong on its own terms: $200.6 billion in Q2 2026 revenue, up 20% year over year, with AWS posting $42.2 billion in revenue — up 37%, its fastest growth in 18 quarters — and AWS operating income nearly doubling to $16.6 billion. AWS's contracted backlog reached $496 billion. None of that is in question. What moved was the spending plan sitting on top of it: Amazon now expects to spend about $220 billion in cash capex this year rather than $200 billion, and executives were direct that memory, not new construction or new orders, drove the revision. "Even at that amount, we will still not have enough capacity to meet all the demand we have in 2026," Jassy told analysts on the earnings call — a capacity constraint the company expects to persist into 2027.

What's actually driving the extra $20 billion

The mechanism is straightforward and, per TrendForce's tracking, accelerating faster than the analyst firm itself has been able to predict. High-bandwidth memory for AI accelerators consumes wafer capacity that used to go toward conventional DRAM and NAND, so as HBM demand grew, everyone still buying ordinary memory — for servers, laptops, phones, everything — started competing over a shrinking supply. Contract prices for conventional DRAM rose an estimated 93-98% quarter over quarter in the first three months of 2026, then a further 58-63% in the second quarter, both according to TrendForce; the firm's most recent forecast puts the increase moderating to 13-18% in the third quarter, as consumer buyers reach what TrendForce calls their "affordability limit" even as server demand stays strong.

That deceleration is a forecast, not a confirmed result — worth flagging because TrendForce's own actual 1Q26 print (93-98%) came in well above what the firm had projected earlier in the year. If 3Q26 runs hotter than the 13-18% currently forecast, both Amazon's $220 billion figure and Apple's margin guidance are numbers that could move again before the year is out.

The same shock, from the paying side

Apple's own quarter looked strong by the same headline measure Amazon's did: $109.4 billion in revenue, up 16% year over year, with gross margin at a stronger-than-expected 50.1%. But that margin is a lagging number, and Apple's finance chief was explicit that it's about to turn. "We paid more in March for memory than [the] December quarter, and we paid more in the June quarter than March," CFO Kevan Parekh told analysts. "We expect this to continue to get worse in September." CEO Tim Cook put it more bluntly: Apple is "in a 100-year flood on the memory pricing with exponential increases," and the company has "reluctantly raised prices" on iPad and Mac as a result. Apple is guiding fiscal Q4 gross margin down to 47-48%, and Parekh said memory alone accounts for more than the entire projected sequential decline.

“We're in a 100-year flood on the memory pricing with exponential increases.” — Tim Cook, Apple CEO, July 30, 2026 earnings call

That last line is the reconciling fact this story turns on: the same price spike is not a uniform shock. [Samsung's memory division posted a record 89.2 trillion won profit the same quarter its own phone business posted a loss](#/article/samsung-q2-2026-record-chip-profit-mobile-loss) from the identical input cost — a company positioned on both sides of the same trade at once. Amazon and Apple have no such offset. Both buy memory; neither makes it.

That asymmetry is also why the two companies' responses look so different despite facing the same input cost. Amazon can route the hit through its own capex line and keep building, because AWS's $496 billion backlog means the spending is pre-sold; a higher bill for the same server fleet mostly shows up as a bigger number on a slide, not a product Amazon has to reprice. Apple doesn't have that option — it sells a finished device at a fixed price point set months in advance, so a mid-cycle input-cost spike either compresses the margin it already booked or gets passed to a consumer who didn't sign up for a memory shortage. Both companies are describing the same shortage in their own vocabulary: Amazon calls it a capacity constraint, Apple calls it a flood. They're the same event, arriving on two different balance sheets.