The AI Boom Just Made Your RAM Cost 60% More. You're Paying For It, Not OpenAI.
I priced out a RAM upgrade for an old desktop in June. 32GB of DDR5 was $149. I went back three weeks later, same module, same retailer. $239. Nobody stole anything, no tariff hit, no supply-chain fire at a fab. The chip didn't get harder to make. It just stopped being made for me.
That's the part of the AI story nobody's telling you at the register. Every headline about the AI boom is about who profits — Nvidia's market cap, OpenAI's burn rate, which startup got which billion. Almost none of it is about who pays for the physical inputs those companies are hoarding, and the answer is: you are, every time you buy a laptop, a console, or a stick of memory, whether or not you've ever touched an AI product in your life. Samsung, SK Hynix, and Micron control north of 95% of global DRAM production, and in 2026 they made a coordinated bet that AI datacenter margins beat consumer margins, then reallocated their factories accordingly. The shortage on your end is the deliberate byproduct of a decision made entirely on someone else's.
Why AI Datacenters Are Eating the World's Memory Supply
High-bandwidth memory — the stacked DRAM that sits next to AI accelerator chips and feeds them data fast enough to keep a $30,000 GPU from idling — pays manufacturers dramatically more per wafer than the DDR5 that goes into your next PC. SK Hynix said its HBM, DRAM, and NAND capacity is essentially sold out for 2026. Micron went further and exited significant parts of the consumer memory market outright to serve enterprise and AI customers instead. That's not a company responding to a shortage. That's a company creating one, on purpose, because the AI-side buyer will pay more per unit than you will and doesn't complain about a 20% quarterly price hike the way a laptop shopper does.
The numbers aren't subtle once you see them lined up. Samsung and SK Hynix raised HBM3E contract prices roughly 20% heading into 2026. Conventional DRAM rose an estimated 80-90% in a single quarter in early 2026. Overall memory pricing climbed about 50% in Q4 2025 alone, with another 40-50% forecast on top of that by the end of Q1 2026. By July, Bloomberg was reporting DRAM spot prices up close to 700% year over year. Bank of America is calling it a supercycle on the scale of the 1990s boom, projecting global DRAM revenue up 51% year over year. A supercycle for the people selling the chips. A cost spiral for everyone buying a device built from them.
The Wealth Transfer Nobody's Auditing
Here's the reframe that matters: this is the first time the AI boom's cost has landed somewhere other than an AI company's balance sheet. Every prior AI-cost conversation — compute spend, model training bills, GPU scarcity — stayed contained inside the industry building the thing. This is different. A gamer buying a graphics card, a small business replacing office desktops, a student buying a laptop before the fall semester — none of them asked to subsidize a datacenter buildout, and none of them are shown that connection anywhere in the price tag. The receipt just says "memory," priced at whatever the market will bear once AI demand has first pick of the supply.
This matters because it's a preview of how AI infrastructure costs are actually going to get distributed going forward, and it isn't going to be "AI companies eat their own costs." It's going to be diffuse, indirect, and largely invisible — priced into components, passed through supply chains, showing up as a vague "market conditions" line in a hardware press release instead of a line item anyone can trace back to a specific datacenter contract. I wrote earlier this year about the gap between AI's promised enterprise ROI and its measured reality — the theatre of AI transformation dressed up as substance. The RAM shortage is the same instinct running in reverse: instead of overstating AI's internal value, it quietly offloads AI's internal cost onto a market that has no seat at the table where that decision got made.
What "Sold Out" Actually Means When You're Not the Customer
The phrase "sold out for 2026" sounds like scarcity. It's really an allocation decision dressed up as scarcity. There is no physical law preventing Samsung from running a DDR5 line at the same volume it ran in 2024. What changed is which customer gets first call on that line's output, and the answer, contractually, is whoever is building AI infrastructure. Consumer retail is what's left after that allocation clears — meaning the "shortage" you're experiencing at Best Buy is really a queuing system where you were moved to the back of the line without being told a line existed.
This is worth sitting with because it changes how you should read every future hardware price story tied to AI. When a manufacturer says supply is tight, the honest follow-up question is: tight for whom, and who got priority. A 60% jump in DDR5 pricing isn't a shortage story. It's a prioritization story, and the prioritization decision was made in a boardroom you'll never see minutes from, affecting a bill you'll pay without ever being told why.
The Ordinary Buyer Has No Leverage Here — and That's the Point
There's no consumer-side fix for this. You can't out-negotiate Micron's capital allocation strategy by shopping around, and "wait for prices to come down" assumes the AI datacenter buildout has a natural end date instead of a demand curve that every major AI lab is actively trying to extend for as long as possible. The closest thing to leverage an ordinary buyer has is simply understanding the mechanism well enough to stop treating it as a random inconvenience. It isn't random. It's a specific, traceable reallocation of a finite physical resource, made by three companies, in response to one industry's spending, at the direct expense of every other industry's customers.
The AI boom was always going to cost someone something real — chips, water, power, land. What's new is watching the bill for all of that get quietly forwarded to a stranger's shopping cart, itemized as nothing more specific than "memory," while the industry that caused the shortage keeps calling it a supercycle.