Altman courts Samsung as DRAM shortage threatens AI buildout to 2027

OpenAI CEO Sam Altman met roughly 50 senior Samsung executives in Yongin, south of Seoul, this week, seeking to secure memory chip supply as a DRAM (dynamic random-access memory, the workhorse memory in servers and PCs) shortage deepens across the industry.

Bloomberg first reported the meeting. The agenda covered AI's impact on industries such as finance and biotech, plus potential business opportunities over the coming decade. Samsung released no concrete commitments. No allocation figures followed the talks.

The meeting lands at a strained moment for memory supply. Memory companies have warned the shortage will persist into 2027 and beyond. Consumer electronics is expected to suffer most, as capacity shifts toward AI infrastructure projects.

How much memory does AI actually consume?

The numbers are stark. In January 2025, the Wall Street Journal reported TrendForce data showing up to 70 percent of memory produced worldwide in 2026 will go to data centers. That leaves consumer device makers fighting for the remainder of a fixed pie.

Altman himself framed the squeeze days earlier. On the Cheeky Pint podcast, hosted by Stripe co-founder John Collison, he argued that individuals running local inference (running AI models on their own hardware rather than in the cloud) make the problem worse.

"A lot of people want local inference, and somehow, they assume that the problem is solved now, because we are going to have open-source models," Altman said. "[They think] 'We're going to be able to run them on our computers and so we're not going to need all the stuff in the data center.'"

He then challenged that intuition directly: "But, 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 five percent of the time. You want maximum utilization in the cloud if you're trying to deal with this very supply-constrained environment. And somehow many people's intuition is the opposite of that."

The claim deserves scrutiny. Altman's utilization figure of five percent is an assertion, not a measured statistic. No sourcing accompanied it. It also serves OpenAI's commercial interest in cloud-based inference. A caveat belongs here, as with every promise the company makes about supply.

What does the Samsung deal actually cover?

In October 2025, OpenAI signed a letter of intent (LOI, a non-binding preliminary agreement) with Samsung Electronics and SK Hynix for memory chip supply. The companies stated OpenAI's Stargate memory demand was projected to reach 900,000 DRAM wafers (the silicon discs from which chips are cut) per month.

That figure lacks a timeline. The companies gave no date for when the demand must be met. They provided no further details on the agreement or anticipated DRAM allocation. An LOI is not a purchase contract. The Yongin meeting suggests the hard bargaining has yet to conclude.

Is a deeper chip partnership coming?

In September this year, Harrison Kim, general manager of OpenAI Korea, said Samsung was working with OpenAI on the development of next-generation chips. OpenAI quickly softened the claim.

A company spokesperson said the partners were planning to expand their collaboration "in the context of our existing relationship, including Samsung Electronics' adoption of ChatGPT and the LOI announced last October." That statement denies nothing. It also confirms nothing beyond what was already signed.

The pattern is consistent. OpenAI announces projections without timelines. Executives describe chip development; spokespersons walk the language back to adoption of ChatGPT and an existing LOI. The gap between the two formulations is where the actual state of the deal sits.

Why the shortage outlasts the headlines?

The structural picture favors memory makers. Data centers may absorb 70 percent of 2026 output. Consumer electronics makers will bid for what remains. Shortages through 2027 give suppliers pricing power and give AI firms a reason to lock in volume years ahead.

For OpenAI, the arithmetic is unforgiving. A claimed 900,000 wafers a month is a recurring demand, not a one-time purchase. Securing it requires suppliers willing to prioritize AI infrastructure over consumer accounts. That is what Altman was doing in Yongin. Whether 50 executives constitute a decision-making forum or a courtship ritual remains unknown.