RAM Prices Are Up as Much as 89% This Year Because AI Datacenters Are Eating the Global DRAM Supply. Your Next Dev Machine Isn't Exempt.

I priced out a 64GB dev machine in January and put off the purchase because I figured RAM would come back down after the holiday season, the way it usually does. It didn't. Depending on which SKU you're tracking, consumer DRAM has climbed as much as 89% in a single quarter this year, HP now says memory eats 35% of what it costs to build a laptop, and Gartner is telling manufacturers to expect PC prices up roughly 17% by year-end. None of that is a temporary component squeeze. It's AI datacenters buying the same wafers your laptop needs, and they have more money than you do.

The receipts

Start with the number that started this post: HP's own CFO told investors on the company's Q1 2026 earnings call in February that memory now makes up 35% of the bill-of-materials cost on a PC, up from 15-18% just one quarter prior. That's not a rounding error, that's memory going from "one line item among several" to "over a third of what the machine costs to build" in about ninety days. Gartner backed that up from the demand side a day later, forecasting combined DRAM and SSD prices up roughly 130% year-over-year by the end of 2026, translating into PC prices up about 17% and smartphone prices up about 13% compared to 2025, with global PC shipments expected to drop over 10% and smartphone shipments down over 8% as buyers hold onto what they have.

On the raw component side, tracking sites and trade press have logged LPDDR5X prices up as much as 89% in a single quarter this year, and 32GB DDR5 desktop kits that were selling for under $100 in early 2025 now regularly list north of $375. I'll flag one figure I could only partly verify: a Bloomberg report cited in July coverage claimed some smaller Android phone makers, buying DRAM on the spot market without long-term supply contracts, are paying prices reportedly up around 700% year-over-year. That's a real but narrow slice of the market (spot buyers without contracts always eat the worst of a shortage), and I wouldn't generalize it to "RAM is up 700%" the way some aggregator headlines have. Treat it as evidence of how ugly things get for buyers with no leverage, not as the number that applies to you.

Three companies, one bottleneck

The reason none of this is spread evenly is that DRAM manufacturing is genuinely a three-company market. Samsung, SK Hynix, and Micron control more than 95% of global DRAM production between them, and all three have been reallocating wafer capacity toward high-bandwidth memory (the HBM stacks that go into Nvidia and AMD AI accelerators) because HBM carries far better margins than the DDR5 that ends up in your laptop or a cloud provider's server fleet. Multiple industry estimates now put AI datacenters on pace to consume roughly 70% of global memory output in 2026, up from something like 20-30% back in 2022. When the same fab lines can either fill your DDR5 order or fill Nvidia's HBM order, and one of those orders pays a much better margin, the DDR5 order loses. That's not collusion, it's just what capacity allocation looks like when demand outstrips supply and one buyer is willing to pay more.

This isn't a blip: ask the people who run the fabs

The instinct with any component shortage is to wait it out, because historically DRAM cycles boom and then crash into oversupply within a year or two. This one's being described differently by the people who'd actually know. SK Hynix CEO Kwak Noh-jung told Reuters in July, on the day his company began trading on the Nasdaq, "we forecast that next year will be the worst year in the industry's history from the supply perspective," and added that he expects customer demand to keep exceeding SK Hynix's production capacity "even beyond 2030." Intel CEO Lip-Bu Tan said in February, after talking to two of the major memory makers, that he'd been told there's "no relief until 2028." Those aren't dueling analyst guesses: that's the CEO of a company that makes memory and the CEO of a company that desperately needs memory, both pointing past 2026 and past 2027 before this normalizes.

Where it actually lands on your budget

For a solo operator, this hits two places. The first and more obvious one is your own next machine. Any 32GB-or-up laptop or desktop refresh you're planning in the next six to twelve months is going to cost more than the same spec would have cost a year ago, and the RAM line item specifically is the part driving that, not the CPU or the display. If you were planning to go from 16GB to 32GB, or 32GB to 64GB, for local model inference or just to stop your IDE from swapping, that upgrade is materially pricier than it was in 2025 and getting pricier through at least Q3 2026 per the latest TrendForce tracking.

The second, less obvious place is your cloud bill. Hyperscalers buy DRAM from the exact same three suppliers your laptop's OEM does, at volumes that dwarf consumer demand. Memory-heavy instance types (anything with a high RAM-to-vCPU ratio, which covers a lot of database, cache, and inference workloads) sit on hardware that gets refreshed and repriced against the same input costs. Cloud providers haven't been as loud about passing this through as HP has, but there's no reason to assume they're immune to an input cost that's up double digits to triple digits depending on the segment. I'd watch your VPS and managed-database renewal pricing over the next couple of quarters rather than assume it's locked in.

What I'd actually do

If a dev-machine refresh was already on your near-term roadmap and your cash flow allows it, I'd move it up rather than wait for a price that multiple chip-company CEOs are telling you not to expect this cycle. Buying now at an inflated-but-known price is a more defensible bet than buying later at a price nobody currently predicting improvement before 2027 or 2028. If the refresh wasn't already planned, I wouldn't manufacture urgency: don't panic-buy RAM you don't currently need on the theory that it'll be worse in six months, because that kind of stockpiling behavior is part of what makes shortages worse than the underlying supply-demand gap justifies.

The honest counter-take here: this is a market, not a law of physics, and markets have surprised people in both directions before. Tom's Hardware reported in July that the price surge is already showing signs of cooling: TrendForce's own data has conventional DRAM contract prices projected up "only" 13-18% quarter-over-quarter for Q3 2026, down from roughly 60% jumps in Q2, as consumers and PC makers simply stop absorbing further increases. That's a demand-side ceiling forming, not a supply fix, but it shows this isn't a one-way ratchet. And there's a second wildcard: several outlets have flagged real investor scrutiny of hyperscaler AI capex, and if that sentiment turns and a few of the biggest buyers pull back their accelerator orders, DRAM demand could ease faster than the multi-year forecasts assume. I don't think that's the likely outcome given what SK Hynix and Intel's own CEOs are saying, but I'd be lying if I said it was impossible. Plan for expensive memory through 2027, and don't be shocked if you're pleasantly wrong.

Author

Lukas

@lukcombinator

Sources

• HP says memory costs doubled in one quarter, now account for 35% of PC build materials — Tom's Hardware
• Gartner Says Surging Memory Costs Will Reduce Global PC and Smartphone Shipments in 2026 — Gartner
• SK Hynix says 2027 will be the "worst year" for memory shortage, forecasts crunch to last until 2030 — Tom's Hardware
• Intel CEO Says There's "No Relief" on Memory Shortage Until 2028 — Bloomberg
• Memory price surge begins to cool as consumers hit affordability limit — AI demand still keeps DRAM and NAND prices climbing through Q3 2026 — Tom's Hardware
• AI-driven RAM shortages threaten more than just the PC market — datacenters expected to consume 70% of global memory in 2026 — Windows Central