Skip to content
Bora Haber
Markets
  • US dollar50.44up 0.05%
  • Euro55.34down 0.18%
  • British pound66.55down 0.11%
  • BIST 10011,477down 3.08%
  • Gold (gram)6,041down 3.55%
  • Gold (ounce)US$3,725down 3.6%
  • BitcoinUS$62,686down 1.61%
BREAKING

Sam Altman: Most PC RAM sits idle, the cloud is more efficient

OpenAI CEO Sam Altman called the low usage of memory in personal computers a waste of resources at a time of tight supply, and pointed to cloud infrastructure.

Arda Sinyal

· 2 min read

Four RAM memory modules on a gray surface: two black with gold logos, two with mirror-finished gold and silver heatsinks.
Personal computer RAM modules (illustrative stock image). Temsili görsel · Fotoğraf: Andrey Matveev / Pexels

OpenAI CEO Sam Altman criticized the low utilization of memory resources in personal computers. At a time when DRAM supply is tight, Altman pointed to cloud infrastructure as the way to use RAM efficiently. His remarks came in a conversation with John Collison on YouTube's Cheeky Pint program, which covered AI's hardware needs and the squeeze in memory supply.

Altman's assessment emerged when the idea of running open models on users' own devices came up. Pushing back against the view that this approach would remove the need for data centers, Altman framed the issue around how the world's limited DRAM capacity is distributed.

Objection centers on low utilization

According to Altman, everyone keeping a significant amount of memory in their own machine creates an inefficient picture for global resource allocation when that hardware is rarely in use. Most memory in personal computers sits idle for much of the day, which he said amounts to considerable waste.

Giving an example of 5 percent utilization, Altman said that under tight supply, utilization of cloud infrastructure should be pushed to the maximum. The 5 percent does not mean only 5 percent of RAM capacity is filled; it refers to the share of time the memory is actually in use.

Local AI versus the cloud

In local AI use, the computations a model needs are performed on the user's own computer. Depending on the model's size and the method used, system memory or graphics card memory is set aside for the job. In the cloud approach, the user sends a request to remote servers and receives the result over the internet.

In the resource-sharing logic Altman advocates, handling requests that arrive at different times on shared infrastructure keeps hardware productive for longer. On the other hand, RAM in personal computers is not used only for AI; it serves many tasks, from games and web browsing to spreadsheets and other applications.

Background: Manufacturers cautious on investment

In the part of the conversation before his remarks on memory, the discussion also covered why rising prices do not quickly translate into new production capacity. Altman said DRAM makers have in the past faced low returns and sharp market swings despite large investments.

According to Altman, manufacturers remember the financial trouble they faced when they built too many factories. That experience leads them to be cautious about capital spending despite today's strong demand.

Technology guides

Take a quiz, vote in a poll

Video news

Comments

Sign in to comment.

More on this topic

See all