Sam Altman's RAM Warning 2026: Buy Now or Wait?

This week, the memory shortage got two new chapters — one from the man whose company helped create it, and one from the company that makes the memory. They agree on the facts and disagree on the conclusion, and together they give buyers the clearest picture yet.

This is part three of our memory series. Part one explained why AI data centres are eating the world's memory, and part two updated the picture with Intel's PC outlook and SanDisk's extraordinary margin. What changed this week is not the direction — it is the deadline. The shortage now has a date on it: 2030.

What Sam Altman actually said

On 29 October 2026, OpenAI CEO Sam Altman appeared on episode 39 of the Cheeky Pint podcast, hosted by Stripe co-founder John Collison. The conversation turned to local inference — the idea that open-weight AI models running on people's own computers could reduce the need for data centres. Altman pushed back, in unusually blunt language:

"From a global allocation perspective, actually, it's quite terrible to have everybody have a bunch of the world's limited supply of DRAM sitting in their own computer and being used 5% of the time. You want maximum utilization in the cloud if you're trying to deal with this very supply-constrained environment."

Three caveats matter before anyone quotes that line. First, the "5%" is Altman's illustrative argument about utilisation, not a measured average of how PC owners use their RAM — your computer's memory does far more than run AI models, from games and video editing to everyday multitasking. Second, Altman announced no policy and no restriction on consumer RAM; he was making an argument about where scarce memory is best used, not telling anyone what they may buy. Third, he did not say local AI should disappear — his claim is that, while memory is this tight, spreading it across millions of home machines is a poor use of a constrained resource compared with keeping it busy in shared data centres.

The online reaction was hostile, and that backlash is itself part of the story. Social posts about the interview drew thousands of reactions, with commenters arguing that they paid for their hardware and should decide how it is used — and that a company buying memory at data-centre scale lecturing consumers about waste was, to put it mildly, unwelcome. Some commenters quoted dramatic price rises for RAM kits over the past two years; treat those as commenter claims, not verified prices, but the sentiment behind them is real: people who just paid more for memory do not enjoy being told their purchase was a misallocation.

What SK Hynix just told investors

If Altman supplied the demand-side argument, SK Hynix supplied the supply-side deadline. On its third-quarter earnings call on 28 October 2026, the world's second-largest memory maker reported a record quarter — revenue of 104.62 trillion won and operating profit of 83.33 trillion won — and then told investors something with direct consequences for every PC buyer.

Its DRAM and high-bandwidth memory (HBM) volumes are effectively sold out through 2029, taking into account long-term agreements already signed and contracts still under discussion. Long-term agreements with 14 key customers already cover more than 50% of its planned production capacity for 2027 and 2028. HBM volumes customers are requesting amount to roughly twice what the company can actually supply. And chief financial officer Kim Woo-hyun said the supply shortfall is expected to persist through 2030, citing the complexity of leading-edge processes and the long lead times for building new fabs.

Read that carefully, because it is easy to over-read. "Sold out through 2029" is SK Hynix's own contracted volumes, not the entire market — shops will still sell RAM. What it means is that the world's memory output is substantially spoken for years ahead, by customers with five-year contracts and deposits, before a single consumer stick reaches a shelf. Combined with Samsung's earlier statement that its 2027 HBM supply is already sold out, that is now two of the three big memory makers telling the same story.

The honest disagreement at the heart of this

Here is the part most coverage misses: Altman and SK Hynix agree that memory is scarce for years, but they point in opposite directions about what to do with it.

Altman's argument is about efficiency. In a shortage, he says, a chip working around the clock in a data centre serves more people than the same chip idling in a home PC. There is logic in that — shared infrastructure really does use hardware harder than individual machines do.

The counter-argument is about control. A home PC's RAM is not idle in the way the argument implies: it runs everything you do, not just AI, and it does so without a subscription, without sending your data elsewhere, and without depending on a connection. Efficiency and ownership are different values, and the shortage forces a trade between them.

Local AI vs cloud AI: the updated comparison

This is where the commercial question lands. If you want AI help, you now have two ways to pay for it — and the shortage changes the maths of both.

No specific subscription prices are quoted here because they change frequently — check the provider's live pricing page before deciding. The structural point does not need a price list: local AI converts a recurring cost into a hardware purchase made during a shortage, while cloud AI converts a hardware purchase into a recurring cost charged by someone facing the same shortage. Neither escapes it.

Who should still buy local hardware? People who use AI daily, who work with private material, who need offline access, or who were buying a PC anyway. Who should lean cloud? People who use AI occasionally, who do not want a hardware purchase now, or whose current machine already handles everything else they do. The worst reason to choose either is panic about a headline: Altman's quote is not a deadline for your purchase, and SK Hynix's 2030 guidance is not a promise that prices only rise from here.

The updated buy/wait list (part 3)

Parts one and two set the rules; this week's evidence sharpens them rather than changing them. Intel's guidance, covered in part two, still points to a soft PC market into the first half of 2027 — so there is no general rush to buy a whole new machine. SK Hynix's 2030 horizon strengthens the case for getting the configuration right when you do buy.

Two practical warnings carry over from earlier parts and matter more now that the story is louder. First, be sceptical of "shortage sale" urgency: retailers know a shortage story sells, and a discount from an inflated price is not a bargain. Second, check specifications before ordering — capacity, speed, generation and compatibility. The most expensive mistake in a shortage is still buying the wrong part in a hurry.

What could still change this story

Three things would loosen the market, and all three are worth watching rather than assuming. New fab capacity is coming — SK Hynix is pushing its Cheongju M15X fab toward full capacity and expanding HBM4 output — but new cleanroom space takes years to become finished chips, which is exactly why the company's own guidance runs to 2030. Demand could weaken: Intel already expects softer PC demand, and if data-centre spending also slows, memory could free up faster than forecasts assume. And the cycle itself has turned before, sharply, when stocked-up buyers and new supply arrived together.

The bottom line: buy carefully, buy the configuration you will live with, and ignore anyone selling you either panic or a precise price prediction.

Recommended Reading

If you are planning a PC build or upgrade, a good PC upgrade guide book is worth a skim before you spend — it helps you match RAM generation, SSD form factor and power requirements to your machine instead of guessing. Browse PC upgrade guide books on Amazon.

As an Amazon Associate, AI Trends Daily earns from qualifying purchases.

No affiliate links for the hardware in this article — component names are plain text, and no retailer prices are quoted, since prices change frequently. Check the retailer's live page before buying. The featured image is an original AI-generated illustration, not a photograph.

Sources: The Gamer — Altman on DRAM and local inference · Motley Fool — SK Hynix memory sold out through 2029 · TrendForce — SK Hynix Q3 2026, sold out to 2029, 14 long-term customers · Seoul Economic Daily — SK Hynix rejects memory peak talk