Sam Altman: “Pretty Terrible” – He Says Your PC’s Memory Belongs in a Data Center

TVS News AI, editorially supervised, presents Sam Altman: “Pretty Terrible” – He Says Your PC’s Memory Belongs in a Data Center.

With these words, Sam Altman criticized the use of RAM in private computers, according to Frandroid. According to the outlet, Altman considers it hardly sensible for people to equip their computers with working memory when, in his view, it is actually used only 5% of the time.

“Pretty terrible”: With this phrase, Sam Altman criticized the use of RAM in private computers, according to Frandroid. According to the outlet, Altman considers it hardly sensible for people to equip their computers with working memory when, in his view, it is actually used only 5% of the time.

Altman is therefore addressing, above all, the many PC owners who buy or install additional working memory but rarely use it to full capacity in everyday life. The conflict between private hardware and centralized computing power thus becomes tangible: What lies unused in your computer for months or years, in the OpenAI chief’s vision, should be pooled in large data centers and made more readily available.

The statement, however, is not a finished product plan. It comes from a podcast that Frandroid published on October 29, 2026. What may sound like a radical restructuring at first glance therefore raises, above all, a question of distribution: Should working memory remain part of your PC – or should it henceforth live in a data center that you access remotely?

Why Altman criticizes RAM usage in private computers

The central thesis

According to the source, Altman’s criticism is not directed in principle against the use of RAM. It is directed against the proportion of time that RAM in a private PC is actually used. The cited 5%, from his perspective, makes a clear difference between what the devices possess and what they are used for.

This is a provocative counterproposal to an everyday experience: Anyone who installs more working memory expects to need it for demanding programs, several applications open at once, or large files. In the statement by Altman reproduced by Frandroid, however, the ownership of working memory is judged primarily on the basis of its actual use.

“Pretty terrible” – this is how Sam Altman assesses the fact that, in his view, private PCs use their working memory only 5% of the time.

This does not mean that existing RAM is automatically useless. The fundamental question is whether private components should continue to be installed in every computer simply because they are technically available there – or whether the same capacity could be organized centrally.

OpenAI plans to procure up to 900,000 DRAM modules per month

The opposite: central concentration instead of private provision

According to Frandroid, OpenAI is already pursuing an opposing model. The company plans to procure up to 900,000 DRAM modules per month for its own data centers. The figure illustrates how great the demand for centrally provided working memory is in OpenAI’s business.

OpenAI is therefore relying not only on data centers in which processors and other central resources are pooled. Working memory is also becoming an element of the company’s own infrastructure. With his criticism, Altman is thus advocating a departure from the individual PC as the sole place for storing and computing data – toward concentrating large storage capacities with a provider.

The source provides no details on whether the cited 900,000 modules have already been fully ordered or how the working-memory modules will be used within the data centers. Only the reported scale is certain: The amount of DRAM procured each month far exceeds the capacity of an individual private computer.

What Altman’s proposal would change for your PC

Your working memory would no longer automatically be your own component

Following his argument, it would make more sense to keep RAM capacity out of private computers and instead make it available in the data center. This would shift the question: What would matter would no longer be how much working memory your PC has, but how much of it you can use through the central infrastructure.

This is particularly relevant where computing power and data storage are already centrally available. OpenAI is building data centers in precisely this environment and, according to Frandroid, wants to use up to 900,000 DRAM modules per month for them. The two sides of the report therefore fit together: Altman criticizes private memory reserves, while his company is building resources for a centralized alternative.

The source does not provide any specific information about costs, access times, or possible use by private individuals. Nor does it name a concrete launch date for such a consumer service. Altman’s statement should therefore be understood above all as a thought experiment: Can working memory be treated more as a shared resource if it is otherwise mostly left unused?

The real debate: Ownership or demand-based use?

5% usage against 900,000 centrally provided modules

The report makes visible a contrast that extends far beyond individual computers. On one side are private PCs with working memory that, according to Altman, he uses only for a small fraction of the time. On the other side is OpenAI, with a planned monthly procurement of six figures for centralized data centers.

For you, the practical consequence is limited but instructive: Additional RAM expansion should not be measured solely by how much memory a device has. What matters is how much you actually need and use. The available sources provide no recommendation on whether buying more makes sense and no details about an end-user product. But they clearly show why OpenAI and Sam Altman want to distribute resources differently: centrally, in a planned way, and according to actual demand.

More value for you

You can find the complete context for Sam Altman’s podcast statement and OpenAI’s planned DRAM requirements in the Frandroid source. It also explains what the 5% RAM usage refers to and which monthly quantity of DRAM modules is mentioned for OpenAI’s data centers.