Sam Altman Sees Near-Term AI Demand Swings, Long-Term Token Growth
He cited monthly token use by the top user rising from about 100,000 in early 2020 to 100 billion at the start of 2026.

OpenAI CEO Sam Altman said near-term demand for AI infrastructure may fluctuate as companies assess memory and data-center needs, while broader AI use could drive token consumption higher over time.
Altman made the remarks during a recent meeting at Samsung's Human Resources Development Institute with about 50 senior executives from Samsung Electronics and its financial and biotech affiliates. The group asked whether demand for AI, memory and data centers could saturate or slow over the next five to 10 years, putting the outlook for infrastructure spending at the center of the discussion.
Altman said the short-term path could be uneven and identified OpenAI and Anthropic as the only companies then driving major demand growth. He warned that the market could face supply shocks and a period of uncertainty as AI infrastructure demand develops.
For the longer term, Altman pointed to the way AI could expand into new uses. He said models could write custom software for users in real time and see broader use in robotics, spacecraft and scientific research. He cited monthly token use by the top user as a measure of the change: about 100,000 in early 2020 versus 100 billion at the start of 2026, roughly a millionfold increase. Those figures describe one top user's monthly use, not aggregate usage.
Altman named chip supply chains, financing, AI safety and social acceptance, and corporate transformation as constraints on expansion. He said chip supply and data-center infrastructure would pose the biggest limits over the next several years, and called for more investment in supply chains. His comments put memory and data-center capacity within the broader challenge of securing enough computing infrastructure.
Asked whether large investments could pressure profitability, Altman pointed to OpenAI's work to improve computing efficiency, including in-house chip development. He said long-term contracts had helped improve operational efficiency. He described the company's aim as building low-cost, vertically integrated infrastructure and selling tokens in high volumes at low margins. Altman said global demand for token use could support a large business even with that lower-margin approach.