News
Altman: Home RAM Sits Idle 95% of the Time, Data Centers Would Use It Better
On the Cheeky Pint podcast, OpenAI's chief said that from a global resource-allocation view it is terrible for home computers to hold scarce DRAM that is used 5 percent of the time. The remarks were mostly about local AI models, but with memory prices high they drew criticism.
Sam Altman, the head of OpenAI, said on the Cheeky Pint podcast that RAM in home computers is poorly used. According to Polish tech media, he believes that with DRAM in short supply, it would be better to move computing to data centers. The remarks came as memory prices are rising and consumers are paying more for hardware as a result.
What Altman said
The Cheeky Pint conversation touched on growing interest in running open AI models locally, on one's own computer, instead of in the cloud. According to Spider's Web, Altman said that from the standpoint of global resource allocation it is downright terrible for people to keep limited DRAM in their computers that is used only about 5 percent of the time.
His argument rests on sharing. In a data center, the same memory serves many requests and many users at once, so it does not sit idle in a rarely used computer. Dobreprogramy, in a piece prepared by the managing editor of Benchmark.pl, gives the example of computers with 32 or 64 GB of RAM.
What he didn't say
Spider's Web notes that the remarks were mainly about local AI models, not gaming PCs, though the argument also covers consumer memory. According to Dobreprogramy, Altman does not dismiss local AI: he acknowledges that it offers independence from providers, better privacy and use without an internet connection. He stresses that when memory is scarce, what matters is not just how much there is but how efficiently it is used.
Dobreprogramy's piece summarizes Altman's remarks without verbatim quotes and without giving the episode date. We did not check the original recording, and the remarks here come from Polish-language press translations.
The memory market
Altman also talked about the memory-making industry. For years, DRAM production was barely profitable: according to Spider's Web, he described it as spending $20 billion in capital expenditure to earn $19 billion. Now the largest producers can generate as much as $100 billion in free cash flow a year.
But the companies remember the old cycles of overinvestment and price collapses, so they are in no hurry to expand capacity. Altman said the invisible hand of the market is failing, since at such high prices new players should already be entering the market.
The bill for buyers
Spider's Web describes the effects on consumers: Apple has raised laptop prices twice, cheap smartphones are at risk of dying out, and the PC market is in its worst shape in over a decade. DDR5 and DDR4 modules are getting more expensive, and AI is driving demand, especially for HBM memory. That is the assessment of the article's author, not independently measured data.
Spider's Web is highly critical of Altman's remarks and points out that data centers are what is driving the memory crisis. Dobreprogramy reports them relatively neutrally and adds that it is up to users to decide how to use memory bought with their own money.
According to Spider's Web's Polish translation of the podcast, Altman said that from a global resource-allocation perspective it is downright terrible that everyone has sizable amounts of scarce DRAM in their computers that is used 5 percent of the time.
For Polish readers the takeaway is practical: the dispute is about where AI should run, locally or in the cloud, and memory is scarce in both places. According to Dobreprogramy, Altman's argument rests on the fact that private computers spend most of the day not performing intensive AI tasks.
