Memory prices were already a hot topic for PC builders when a clip of OpenAI chief executive Sam Altman started circulating at the end of October 2026. This article separates what he said from what people say he said, explains the numbers that keep getting repeated, and ends with practical points for anyone planning a gaming PC.
Key takeaways
- Altman said on the interview channel Cheeky Pint that it is “quite terrible” for everyone to hold the world's limited DRAM in their own computer.
- The “5% of the time” figure is his claim. The outlets that reported it did not verify it.
- Nothing he said changes how much memory your games need today. The real effect for gamers is on prices.
What Altman actually said
Tom's Hardware and PC Gamer both reported on October 30, 2026 on remarks from an interview with the YouTube channel Cheeky Pint, about 29 minutes into the video. Neither article gives the date of the episode, so we do not either. The topic was running AI models on home computers, called local inference. Altman argued that, in a supply-constrained market, it is “quite terrible” to have a bunch of the world's limited DRAM “sitting in their own computer” and “used 5% of the time”. He added that you want “maximum utilization in the cloud”.
Two details matter. First, he was making an argument about allocation of a scarce resource, not announcing a product or a policy. Second, Tom's Hardware notes that OpenAI's main business is closed models that run in data centers, so his view comes from a company with a stake in the answer. That is context, not a verdict on whether he is right.
Why gamers are angry
Memory has become expensive, and the industry links this to demand from AI data centers. In its September 30, 2026 press release, market researcher TrendForce says suppliers are prioritizing advanced capacity for high-performance server products, which keeps the market undersupplied, and that PC makers keep buying aggressively because they expect supply to stay tight in 2027. Against that background, a CEO of a leading AI company saying PCs should hold less memory landed badly. PC Gamer called his complaint about scarcity “a bit rich” because OpenAI helped create it. That is the outlet's opinion, and we note it as such. On November 1, Elon Musk replied on X with “The people deserve to have their RAM too!”, as reported by Tom's Hardware.
The “40%” number: what it is and what it is not
You will see claims that OpenAI could take up to 40% of the world's memory. OpenAI did not state that figure. It comes from a Tom's Hardware article dated October 1, 2025, whose author divided up to 900,000 DRAM wafers per month by an estimated global capacity of about 2.25 million wafer starts per month. Samsung and SK hynix had said that anticipated OpenAI demand could grow to 900,000 wafers monthly under preliminary agreements for the Stargate project. So “up to 40%” is a media calculation from a ceiling, not a purchase that has happened. Treat it as an estimate.
What is fair in his argument
A server shared by thousands of users can be busy around the clock, while a home PC is idle for much of the day. If memory is scarce, an arrangement where the same chips serve many people has an obvious efficiency logic, and it is honest to say so. Whether you agree with it depends on what you value, because efficiency is not the only thing people want from a computer.
What is debatable
This part is our analysis, not fact. A single percentage cannot describe what PC memory does. A computer uses spare RAM to keep programs and files ready, so idle memory still makes the system feel fast, and a game needs a lot of it exactly when you play. Also, the 5% figure has no published method behind it, so it cannot be checked. Finally, running a local model, using a game, or editing video are choices made by the owner of the machine. Whether those uses are worth the memory is a question about priorities, and different people will answer it differently.
What it means for your next upgrade
We do not predict prices, and nobody can promise one. What we can report is dated and sourced. TrendForce's September 30, 2026 release forecasts conventional DRAM contract prices up 10–15% quarter on quarter in Q4 2026, with the pace of increases expected to moderate. A later TrendForce item from October 20 cites about 20%. These are contract prices between manufacturers and buyers, not shop prices, and forecasts change. Tom's Hardware also reports that Lenovo said in June it did not expect prices to return to pre-2026 levels soon, if ever, and that Samsung and SK hynix have warned the shortage could last into 2027 and beyond.
- If your games run well today, there is no technical reason to rush. Altman's remarks do not change game requirements.
- If you must buy now, check what your games actually need with our Can I run it tool and see what frame rates to expect with the FPS calculator. Buy the capacity you need rather than the biggest kit.
- If you can wait, you take the risk that prices stay high. Sources disagree on timing, so check current prices in your own market before deciding.
More hardware stories will appear in the Hardware section, and our games pages show what the titles you play require.
FAQ
Will RAM prices go up because of OpenAI?
Nobody can say for certain. TrendForce forecasts that contract prices for conventional DRAM keep rising in Q4 2026, and it links this to AI server demand in general, not to one company. Any specific price forecast is a guess.
What did Sam Altman actually say about memory in PCs?
On the Cheeky Pint channel he said it is “quite terrible” for everybody to have a bunch of the world's limited DRAM in their own computer, “used 5% of the time”, and that you want “maximum utilization in the cloud”. The 5% figure is his claim and was not verified.
Should I buy RAM now or wait?
If your current memory is enough for the games you play, there is no urgency. If you need an upgrade, buy the capacity you need and check prices in your region. Sources expect a tight market into 2027, but forecasts change.
Do I need a lot of RAM to run AI models locally?
Usually yes. As a rough guide, a model needs memory close to its size: a model compressed to 4 bits per parameter takes about half a gigabyte per billion parameters, plus overhead. Exact needs depend on the model and the software.
Sources
- Tom's Hardware, Oct 30, 2026: OpenAI CEO says it's 'quite terrible' for your PC to have RAM
- PC Gamer, Oct 30, 2026: Sam Altman argues that it's bad you want DRAM in your computer
- Tom's Hardware, Nov 2, 2026: Elon Musk mocks Sam Altman's claim about PC RAM
- Tom's Hardware, Oct 1, 2025: OpenAI's Stargate could consume up to 40% of global DRAM output
- TrendForce, Sep 30, 2026: 4Q26 memory contract price forecast
- TrendForce, Oct 20, 2026: DRAM prices seen rising 20% QoQ in 4Q26