OpenAI CEO Altman cited rising compute and memory costs as significant headwind, questioning margin expansion.

OpenAI CEO Sam Altman acknowledged on Thursday that soaring compute and memory costs are creating significant financial pressure as the company pursues rapid expansion, calling the infrastructure burden "definitely a headwind."

In an interview with CNBC, Altman addressed the mounting costs required to maintain OpenAI's position at the frontier of generative AI. He offset this reality with a positive technical development: OpenAI's newest model achieves a 54% improvement in token efficiency specifically designed for agentic coding applications.

Altman confirmed that Microsoft remains among OpenAI's largest and most strategically important customers, putting speculation about partnership strain to rest. He pushed back firmly against reports of a potential U.S. government equity stake, stating, "There are a lot of inaccuracies there," and explicitly dismissing rumors of a 5% government stake.

While ruling out government ownership, Altman noted that OpenAI has implemented "many changes" following extensive discussions with Washington officials and expressed optimism that future collaboration will be "smoother." On the question of a 2026 initial public offering, Altman declined to speculate, simply saying, "I don't know."

On the geopolitical front, Altman warned briefly about mounting global competition, observing that Chinese open-source AI models are becoming "very good." His remarks reflect the dual pressures OpenAI faces: managing acute economic constraints from massive computing requirements while delivering successive improvements in model efficiency and capability.