Sam Altman Calls PC RAM a “Quite Terrible” Use of Scarce Memory, as OpenAI’s Deals Target Up to 40% of Global DRAM

BossMac Suba
By
BossMac Suba
Boss Mac Suba is the driving force behind Back2Gaming.com, one of the most enduring, dedicated PC gaming hardware and video game reviews publication. With over a...

Sam Altman Says Your PC’s RAM Belongs in the Cloud, While OpenAI’s Memory Deals Target Up to 40% of Global DRAM Output

OpenAI CEO Sam Altman has a problem with the memory inside your PC. Speaking with Stripe co-founder John Collison on episode 39 of the Cheeky Pint podcast, released October 29, 2026, a hopefully uninebriated Altman argued that RAM installed in personal computers sits idle most of the time, and that while memory stays this scarce, it should go to cloud data centers running at full utilization. We went through the full transcript, and the argument deserves scrutiny for what it says and for who is saying it.

What Altman said: About 29 minutes into Cheeky Pint episode 39, answering a question on the compute shortage, Altman said many people assume local inference on open-source models will reduce the need for data centers. He countered that “from a global allocation perspective, actually, it’s quite terrible” for everyone to keep a share of the world’s limited DRAM in their own computers, used about 5% of the time, and that maximum cloud utilization is the better answer in a supply-constrained market. He did not propose any limits on consumer RAM.

The Context: A Shortage Altman Knows Well

The remark came out of a broader discussion on why high memory prices have not produced new supply. Altman told Collison that progress on that front was going badly and that, from his perspective, the invisible hand is currently failing. He explained that DRAM makers historically spent about US$20 billion in capex to earn about US$19 billion through repeated boom and bust cycles. He estimates each of them is now on track for around US$100 billion in free cash flow in a year, yet they remember nearly going bankrupt from overbuilding fabs, so they keep expansion conservative.

Altman then offered local AI as an example of how counterintuitive the supply chain has become. In his framing, a home PC running an open-source model ties up memory that a data center would keep busy around the clock.

The Biggest Buyer Lecturing the Shoppers

OpenAI is not some random consumer in this market. In October 2025, Samsung and SK hynix signed letters of intent to supply OpenAI’s Stargate initiative, with OpenAI’s anticipated demand reaching up to 900,000 DRAM wafer starts per month. TechInsights analysts put total global DRAM production at about 2.25 million wafers per month in 2025, which would place Stargate’s ceiling at up to 40% of worldwide supply.

PC builders are paying for that demand.  As of this writing, PH pricing for a 32GB DDR5-6000 kit is around $400.

When the CEO of a company holding letters of intent for that much memory calls the RAM in a gamer’s tower a misallocation, the people paying those prices have every reason to object.

FigureValueSource
PC DRAM utilizationAbout 5% of the timeAltman, no data provided
DRAM makers, historical economics~US$20B CapEx to earn ~US$19BAltman
DRAM makers, current cash flow~US$100B free cash flow each per yearAltman’s estimate
OpenAI Stargate DRAM demandUp to 900,000 wafer starts per monthSamsung and SK hynix letters of intent, October 2025
Share of global DRAM outputUp to ~40%TechInsights estimate
32GB DDR5-6000 average price in the Philippines~$400PCHUB pricing
Data center cost per gigawatt (5-year TCO)~US$70B now, US$20B targetAltman

The 5% Figure Has No Source

Altman pretty much asspulled this number and gave no study, measurement or definition behind the 5% number. Utilization is also the wrong metric for hardware people own. Memory in a PC is available to its owner on demand, works offline, keeps files and prompts on the machine, and comes with no subscription, rate limit or terms-of-service change attached if talking about local AI use. Idle capacity is what ownership looks like and for most users, larger RAM is more number more better duh.

The same DRAM also does far more than local AI. My benchmark for desktop usage sees RAM not for 24/7 at-load usage but rather for on-demand capacity for multi-tasking whether its for desktop apps or games, more RAM is more better for a more responsive system at the very least.

Altman Argued the Other Side Earlier in the Same Hour

Earlier in the conversation, Altman said OpenAI wants to help people build lives they define for themselves, rather than live on a basic income while a small group in a few square miles of San Francisco makes the decisions about the future. Later, discussing AI agents that operate computers, he said he gets annoyed when he cannot have an AI automate tasks on his own computer, and that he wants it to feel like his agent doing what he asks.

Pooling the world’s memory in cloud data centers puts the hardware, the pricing and the terms of access in the hands of a small number of operators. That sits uneasily with the decentralized future he described minutes earlier.

What Altman Says He Wants

Altman did not call for restricting consumer hardware, and his stated long-term goal is more supply. He said OpenAI has built its own chip tuned for performance per watt, is learning to build its own data centers, and has a robotics program aimed at constructing more of them. He also said he would be much happier if the five-year cost of a gigawatt of AI compute fell from about US$70 billion to US$20 billion.

Cheaper compute through more supply is a goal PC owners share. Calling the memory they already paid for a waste of the world’s resources is another matter, and the backlash shows how it landed. As a publication that tests hardware on our own benches, our position is simple: owning your compute is worth defending, and the costs of the AI buildout should not be pinned on the people who still build their own machines.

Share This Article
Follow:
Boss Mac Suba is the driving force behind Back2Gaming.com, one of the most enduring, dedicated PC gaming hardware and video game reviews publication. With over a decade of experience in IT and more in doing reviews for things he love since 2002 from Jpop albums, anime series, games in the early days of the internet.My stance on AI-generated writing and content

Tondo since birth. To this day.

457 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *