Sam Altman warns Samsung that AI demand may be double counted

Sam Altman has a message for the companies building the hardware behind the AI boom: some of the demand you’re counting might not exist twice.

The OpenAI CEO warned Samsung executives that AI-related memory and data-center demand may be double counted in market estimates. That is an unusual note of caution from the person whose company sits at the center of the spending spree. He paired it with optimism, though. Altman still expects strong long-term growth in AI demand.

What Altman told Samsung’s leadership

The meeting took place on October 31, 2026, at the Samsung Human Resources Development Center in Yongin, South Korea. Altman sat down with approximately 50 presidents from Samsung Group affiliates, according to a session detailed by Financial News.

His central point was double counting. In AI infrastructure, the risk works in a similar way. Demand signals can pass through several layers of the supply chain. If each layer reports its own needs, the same underlying demand can appear more than once in market assessments. Suppliers who plan capacity against that total could end up building for customers who are, in effect, the same customer.

Altman framed the concern against recent market volatility and an ongoing debate over whether a bubble is forming around neocloud infrastructure. Neoclouds are the newer cloud providers built mainly to rent out AI computing power, as opposed to the established general-purpose cloud giants.

He also offered a pointed read on who is actually generating demand. As of November 2026, Altman said only OpenAI and Anthropic are driving major demand growth in the AI sector.

The supply side is already stretched

High-bandwidth memory, or HBM, is the specialized memory stacked alongside AI chips so they can move data quickly. It is also hard to produce, and right now it is spoken for.

Samsung’s entire 2027 supply of HBM is sold out as of November 2026. SK hynix, its domestic rival, has its production fully committed through 2029. Altman highlighted these near-term constraints on AI expansion during his remarks.

OpenAI’s own ambitions help explain the stakes. The company signed a letter of intent with Samsung and SK hynix in October 2025. As part of the Stargate AI infrastructure project, Altman projected that demand could reach 900,000 DRAM wafer starts per month.

What this means for chipmakers and investors

For memory producers like Samsung and SK hynix, the warning is less about today’s order books and more about tomorrow’s capacity plans. Sold-out supply provides near-term comfort. The real risk sits in deciding how many new production lines to build on forecasts that could include duplicated demand.

The concentration point deserves attention too. If Altman is right that only OpenAI and Anthropic are driving major demand growth, then much of the sector’s outlook depends on the spending decisions of two companies. That creates exposure for anyone further down the chain, from neoclouds to memory fabs.

There is also a credibility angle worth watching. OpenAI benefits from tight memory markets in one sense, since it has locked in supply relationships. Altman’s comments can be read as both a candid market assessment and a customer managing its suppliers’ expectations.