OpenAI boss calls it “terrible” that your PC keeps its own RAM and would prefer it to be in its data centers

As a guest on the Cheeky Pint podcast, in an episode posted online on October 29, Sam Altman revisited the desire of many users to run artificial intelligence models on their own machines.From a global distribution standpoint, he explained, “it is actually pretty terrible that everyone has a pile of DRAM, whose global supply is limited, sitting in their own computer and being used 5% of the time.” He argues for making maximum use of this memory in the cloud, given how constrained supply is, and wonders why “the intuition of many people” points in the opposite direction.

How much is the memory shortage costing you?

In October 2025, a 32 GB DDR5 kit—the live memory in recent PCs—was selling for around €120 at French retailers. The same capacity had exceeded €400 in May, and a reference 32 GB DDR5-6000 kit rose by a further €42 in a week in September, reaching €469.90. Even DDR4, the previous generation, was then costing more than €200 for 32 GB.

By mid-October, the median price of 16 GB of DDR5 had risen by a further 20% in three months. We never let a week pass without mentioning it in our columns: this surge is driven by the appetite of AI-focused data centers. Manufacturers are redirecting their production capacity toward HBM, a stacked, very high-speed memory reserved for accelerators (and more profitable, incidentally), at the expense of the memory modules sold to the general public.

Why is Altman’s departure being received so badly?

On social media, his remarks sparked a flurry of sarcastic comments, with one user asking whether he should now rent his lawnmower from OpenAI, since it is only used once a week.Others accuse him of having created the very shortage he invokes.

On October 1, 2025, Samsung and SK hynix, which together control about 70% of the global DRAM market, had signed preliminary agreements with OpenAI to supply Stargate, its giant data-center program. The targeted demand could reach 900,000 silicon wafers per month, enough to absorb nearly 40% of global DRAM production.

In March, Sam Altman already described a future in which intelligence would be sold like water or electricity, with everyone buying it from OpenAI “by the meter.” Running a model on one’s own machine could eventually reduce what people pay for subscriptions, even though OpenAI mainly sells access to closed models hosted in data centers.

His reasoning relies both on a data-center build-out that never slows down and on a memory module that is more useful for AI, day and night, than for spreadsheets, browsing, or gaming. OpenAI itself does publish some open models that can be run at home.

👉🏻 Follow tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.