Sam Altman says PC DRAM belongs in data centers, not personal computers

What Altman said

In an interview on the YouTube channel Cheeky Pint, OpenAI CEO Sam Altman argued that the world's limited supply of DRAM should go to AI data centers rather than personal computers. His core claim was that '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.'

The figure — 5% utilization of PC memory — was Altman's own estimate. He argued that a data center can keep the same memory busy far more of the day because many users share it, making centralized allocation more efficient.

Altman made the remark in the context of a broader discussion about the global memory shortage and how AI data centers are competing with consumer electronics for the same fab capacity. The interview was published on Cheeky Pint, a podcast-style YouTube show, and picked up within hours by Tom's Hardware, TechSpot, PCMag, and WCCFTech.

The pushback

The response on Hacker News was immediate and skeptical. Several commenters argued that the 5% figure was pulled from 'thin air' and that Altman's logic would apply equally to any personal computing resource — housing, cars, smartphones — that is not used at full capacity every minute. 'The whole point of personal computing is that you can do whatever you want,' one top-voted comment read. 'Centralization is more efficient, but also the goals of centralized systems are ultimately wrong.'

Others pointed to the business incentive: Altman runs a company that sells cloud API access. 'He wants me to have to pay him rather than being able to run open models locally,' one commenter wrote. 'His solution is to return us all to an undemocratic model of computing that has been in decline since the 1970s.'

Critics also noted that the DRAM market has been distorted by AI demand. Memory manufacturers have been prioritizing high-margin data-center modules over consumer DIMMs, contributing to rising DDR5 and DDR6 prices. Altman's argument, some said, frames a market reality as a design virtue.

The local inference context

Altman's remarks arrive at a moment when local AI inference is gaining traction. NVIDIA's RTX Spark laptops, launched in October, are marketed explicitly for on-device AI. Perplexity Computer runs fully local on Mac with Qwen-based models. The open-source community continues releasing small, capable models designed for consumer hardware — Nemotron 3.5 Bolt at 4 GB, Recursive Flint, and others.

Running models locally eliminates API costs, keeps data on the device, and works offline. For developers and enthusiasts building on those advantages, Altman's suggestion that the memory should be elsewhere reads as an attempt to steer the market back to the cloud.

The memory supply picture

Altman's argument rests on a real constraint: AI data centers consume a growing share of global DRAM production. Industry analysts have noted that memory makers are allocating more fab capacity to high-bandwidth memory and server modules, squeezing supply for consumer DIMMs. DDR5 prices have remained elevated through 2026, and DDR6 — expected in late 2027 — may face similar pressure.

However, the 5% utilization claim is Altman's own figure with no cited source. PCMag's coverage noted that 'Altman's 5% estimate is not backed by published research,' and Hacker News commenters challenged it as anecdotal. A data center's own servers often run at well under 100% memory utilization, and the comparison omits the value of low latency and data locality that local hardware provides.

What it says about OpenAI's strategy

Altman's comments fit a pattern. OpenAI has been pushing toward more centralized, API-based AI access — the Decisions API, the responses API, Codex Cloud — while deprioritizing on-device capabilities. The company's [ChatGPT and Codex shared usage allowance announcement](https://ailookout.news/blog/chatgpt-and-codex-will-share-one-usage-allowance-starting-in-december/) on the same day further ties users to OpenAI's metered ecosystem.

The DRAM remark, whether or not Altman intended it as a policy argument, puts OpenAI's CEO on record favoring infrastructure choices that centralize compute in cloud data centers rather than on end-user devices.