Good morning and happy Sunday.

For two years the AI trade has been a story about the top of the stack: the chips that train the models, the hyperscalers that buy them, the power plants being dragged out of retirement to feed them. The numbers were always somebody else's problem, settled between Nvidia (NVDA), the cloud giants, and a handful of suppliers in Asia. Then, this week, Apple (AAPL) chief executive Tim Cook said the part out loud, conceding that higher device prices are now "unavoidable," and the whole abstraction finally landed somewhere you can touch. The water that's been filling the data center has started running down the pipes into your gadget drawer, and that plumbing is the subject of today's deep dive.

Start with the commodity nobody outside the supply chain used to think about: memory. Every phone, laptop, game console, and car needs DRAM (the working memory) and NAND flash (the storage). For most of the last decade it was cheap, abundant, and boring, the rice of electronics, priced like a sack of it.

Artificial intelligence broke that calm. Training and running large models is enormously memory-hungry, and the most prized variety, high-bandwidth memory (HBM), is made by stacking DRAM dies into dense towers that sit beside the AI accelerators. The catch is that HBM is brutally capacity-intensive to produce. Because the dies are larger and the stacking yields are lower, a single HBM wafer swallows the production capacity of two or more conventional DRAM wafers, as Tom's Hardware has detailed. Every tower built for a server is bits not built for a handset.

The result is a classic crowding-out, and the prices tell the story. TrendForce's latest survey expects conventional DRAM contract prices to climb 58% to 63% quarter-over-quarter in the second quarter of 2026, with NAND flash contract prices rising 70% to 75%, the research firm said, as suppliers keep reallocating wafers toward server-grade parts. These aren't the gentle single-digit moves the memory market is built around. They are the cost of an input doubling while everyone downstream is still trying to ship the same phones.

If consumer demand is starving, why don't the memory makers just point more capacity at it? Because the math has stopped making sense for them.

Samsung, SK Hynix, and Micron (MU) together control north of 90% of the world's DRAM, and all three have spent the past year converting lines to HBM and server memory for one unsentimental reason: those parts can earn several times the revenue per wafer that a commodity smartphone chip does. When the most profitable customer in your history, the hyperscaler building out AI, is willing to sign multi-year, take-or-pay agreements, the rational move is to sell them the reservoir and let the garden hose run dry. Samsung and SK Hynix have warned the resulting shortages could persist through 2027 and beyond, with buyers already reserving supply years in advance, per Tom's Hardware.

That is the heart of why this isn't a normal cycle. Memory has always been brutally cyclical — gluts crater prices, makers cut output, scarcity returns, repeat. What's different now is that the glut-and-bust rhythm has been overridden by a structural buyer who values the same wafers far more than your phone does. The pipes haven't sprung a leak; they've been deliberately re-plumbed, and the pressure is all flowing one way.

Here is where the upstream squeeze surfaces downstream, in numbers a shopper can feel. Gartner estimates a 130% surge in combined DRAM and SSD prices by the end of 2026, enough to push PC prices up 17% and smartphone prices up 13% versus 2025, the firm projected in February. Memory is expected to balloon to 23% of a PC's bill of materials, up from 16% in 2025, which is why Gartner thinks the sub-$500 entry-level laptop disappears entirely by 2028 and why basic-phone buyers are exiting the market roughly five times faster than premium ones.

Apple is the cleanest read-through, because it sells the priciest hardware and reports its margins for all to see. Cook told The Wall Street Journal the memory-cost situation has become "unsustainable" and that Apple is even "willing to use our balance sheet" to lock down supply, according to a report carried by Yahoo Finance. The scale of the jump is stark: TechInsights pegs the memory and storage bill inside an iPhone 17 Pro at roughly $50 last year, but closer to $200 for the coming iPhone 18 Pro, MacRumors reported. A company that guards its margins like a vault is telling you, in advance, that it can't fully absorb this one.

The shipment forecasts follow the prices down. Gartner sees global PC shipments falling 10.4% and smartphone shipments dropping 8.4% this year, calling it the steepest contraction in device shipments in over a decade and warning that buyers will simply hold their gear longer. That's the second-order kicker: an AI infrastructure boom is quietly lengthening the upgrade cycle for the consumer electronics industry, draining unit volume from the very companies that helped popularize AI features in the first place.

If you want one gauge of whether the squeeze tightens or eases, it prints this week. Micron reports fiscal third-quarter results on June 24, and the company has already said its entire 2026 HBM output is sold out, Investing.com noted. The line to watch isn't revenue, it's gross margin, which the Street expects to push toward 80%, a level that would confirm the maker is pricing from a position of pure scarcity rather than the wafer-thin margins memory is famous for.

That single percentage is a tell for the whole tape. A fat memory margin is the supplier's reward and the device-maker's tax, and it tells you the reservoir is still being filled faster than it can be refilled. For everyone downstream — the phone in your pocket, the laptop on your desk, the car in your driveway — it signals the bill stays high a while longer.

The honest uncertainty is timing, not direction. Memory has humbled every forecaster who called the top too early, and a genuine glut would arrive the moment AI capacity finally outruns demand or the makers over-build, sending those same contract prices into reverse. But for now the water is running the wrong way for consumers, and the people who set memory prices are in no hurry to redirect it. The AI boom spent two years as a story about what the future would cost someone else. This week it quietly itemized the charge and slid it across the counter to everyone.