For three years the AI build-out has been an abstraction for most people: trillion-dollar valuations, data centers in distant counties, debates about jobs that have not quite arrived. This week it became concrete in the most ordinary way imaginable. Apple raised prices on nearly its entire hardware lineup, from the Mac and iPad to the HomePod, Apple TV and Vision Pro, with increases averaging roughly $247 per product and climbing as high as $1,300 on the top Mac Studio. The reason is not tariffs or a new design. It is that the chips Apple needs to build a laptop are now being bought, in staggering quantities, to build the machines that run AI.
The pinch point is memory. Every AI server stuffed with expensive Nvidia processors needs even more expensive high-bandwidth memory to feed them, and that demand has drained the same pool of DRAM and flash storage that goes into consumer devices. Tim Cook, who warned a week earlier that price rises had become "unavoidable," described the situation to The Wall Street Journal as a "hundred-year flood," the worst he had seen in four decades. The numbers behind his alarm are stark: research firm TechInsights projects that memory and flash costs could roughly quadruple by autumn, pushing the bill of materials on a flagship iPhone up about 25 percent. Apple's stock fell 6 percent on the day it confirmed the hikes, its worst session in more than a year.
What is a crisis for device makers is a windfall for the three companies that dominate memory. Micron, the only major American player, has watched its shares climb more than 85 percent in a year as Wall Street recasts it from a boring commodity supplier into "the next Nvidia." High-bandwidth memory modules now sell for upwards of $1,000 each, at margins estimated near 60 percent against 30 percent for ordinary chips, and Micron's revenue from them is forecast to leap from almost nothing in 2023 to more than $8 billion this year. South Korea's SK Hynix, an early and once-doubted bet on the technology, has become the most valuable company in the country and is preparing a US listing to raise as much as $29 billion. Samsung booked over $1 billion in sales of its newest HBM4 chips within four months of starting production.
The more interesting shift is structural. Memory has always been a brutally cyclical business, prone to gluts that crater prices and stocks by 70 percent or more. To escape that trap, makers like Micron are now signing long-term "take-or-pay" deals, locking customers such as Nvidia into guaranteed payments whether or not they use the supply. It is an attempt to convert memory from a commodity that booms and busts into a strategic input with predictable cash flow, much as oil producers learned to do. Whether it holds through the next downturn is the open question.
That question has teeth, because not everyone believes the demand is permanent. Jefferies strategist Chris Wood argues the AI boom will end not from a chip glut but from the moment investors conclude that hyperscalers cannot earn an adequate return on hundreds of billions in spending. For now, the spending continues, and the bill is being passed along. The most telling sign that AI has arrived is not a benchmark or a demo. It is a shopper at a store discovering that this year's laptop costs more, for reasons that have nothing to do with the laptop.