What happened
The Verge reports that Altman spoke onstage at Semafor’s Silicon Valley & The World summit on Monday, November 16, 2026. “I want to be clear, I want people to have a lot of RAM,” he said. “I think everybody should be able to buy a lot of RAM for their own computer.” He added that “home tinkering is great.” The remarks address controversy over comments he had made the previous month about personal computers and data-center memory.
On the Cheeky Pint podcast, Altman had described it as “quite terrible” for everyone to have RAM in personal computers rather than data centers. Earlier in November, he followed up on X, saying he wanted people to have “so much ram” and that his original comments had been “taken out of context.” His November 16 statement makes the consumer-facing point more explicit: he supports people having substantial memory in their own machines, even while maintaining an efficiency argument for shared infrastructure.
Altman said that a few years ago OpenAI began saying the world needed to build more of the memory supply chain. He acknowledged that DRAM prices had risen sharply in recent months, estimating—while qualifying that he did not know exactly—that they may have increased fivefold to sevenfold. The Verge connects rising memory demand to AI companies buying memory for data centers and reports price increases affecting products such as phones and game consoles.
Altman also described memory as one of several constraints OpenAI has encountered at different times. Other constraints, he said, included rack assembly, data-center capacity, and “some random component.” That list matters because it places memory inside a wider physical delivery system: additional chips alone would not remove every limit on deploying more compute. The Verge does not describe a technical change in how OpenAI allocates memory, nor does it quantify the company’s demand relative to the wider market.
He argued that, if memory remains scarce, data-center memory used continuously can be more efficient overall than RAM sitting idle in an edge device. This is an argument about aggregate utilization, not a claim that every local workload should move to the cloud. A personal computer can be valuable for experimentation, privacy-sensitive tasks, offline work, or latency-sensitive software even if its memory is not used around the clock. The report describes remarks and an estimate, not a new OpenAI product, consumer purchase program, or commitment to expand supply.
What the archive adds
New Runtime covered Altman’s earlier argument that, during constrained DRAM supply, memory sitting mostly idle in personal computers is less efficient globally than memory concentrated in highly utilized data centers. This is a new clarification on a known story: Altman now explicitly says everyone should be able to buy a lot of RAM for personal computers, while retaining the shared-infrastructure argument. Earlier New Runtime coverage (November 1, 2026).
Trend signal
This develops the same shared-infrastructure tension captured in New Runtime’s shared-context infrastructure shift: centralized resources can be more continuously utilized, while local ownership still has value. Altman’s comments do not establish a new technical trend or show that either approach is winning; they make the trade-off unusually visible in the context of a supply shortage. The analogy is limited: the shift concerns project-scoped AI memory, while these remarks concern physical DRAM capacity.
Why now
Altman’s comments come amid rising DRAM prices and constrained memory supply. The Verge says AI companies have been buying memory for data centers, and Altman acknowledged that memory has constrained OpenAI at different times. The renewed remarks respond to the controversy around his previous comparison of personal-computer and data-center memory. They also arrive while the company is describing physical capacity, rack assembly, and components as linked limits rather than isolated purchasing decisions.
Why it matters
For AI builders, the remarks expose a practical tension rather than resolve it. Local RAM can support experimentation and workloads that benefit from keeping data and compute close to the user. Shared data-center memory can be utilized more continuously, but depends on available infrastructure and access to cloud capacity. During a supply crunch, both availability and price can affect decisions about where to run models and tools.
Altman’s position is not a procurement guide: it contains no recommendation about how much RAM a team should buy, no forecast for prices, and no new access terms for OpenAI infrastructure. Builders should treat the comments as context for a broader resource-allocation debate, not as evidence that personal hardware is obsolete or that data-center capacity is assured. The immediate takeaway is that memory is a system-level constraint, alongside racks and data-center capacity, not merely a PC specification.
Teams making near-term hardware plans should separate two questions that are often conflated: whether a workload benefits from local execution, and whether purchasing more memory is feasible at current prices. The Verge’s report supplies no threshold for answering either question. It does, however, underline that a cloud-first fallback can inherit another set of bottlenecks when suppliers, racks, and capacity are constrained at the same time.
Limits and uncertainty
Altman qualified the fivefold-to-sevenfold DRAM price estimate by saying he did not know the exact figure. The Verge’s account does not independently verify the estimate or specify which DRAM products, markets, or time period it covers. The remarks provide no detailed figures for OpenAI’s memory needs, utilization rates, or the relative efficiency of local and centralized memory.
The Verge also does not report a comparison of the cost, latency, energy use, privacy properties, or reliability of local and centralized AI workloads. It is therefore not possible to derive a general hardware-buying rule from Altman’s comments. They also announce no change to consumer RAM availability, pricing, or OpenAI’s infrastructure plans.
Reader takeaway
Altman now says plainly that people should be able to buy plenty of RAM for their own computers, even as he argues that shared data-center memory may be more efficient under shortage. For builders, the useful point is not a new hardware prescription: memory availability and utilization are competing constraints across local and cloud AI work.