OpenAI plans to burn $85B in 2028 alone as Altman and CFO clash over IPO timing
Apr 6, 2026
Key Points
- OpenAI plans to spend $121 billion on computing power in 2028 alone, creating an $85 billion annual cash burn that would exceed virtually any public company in history.
- CEO Sam Altman wants to take OpenAI public by Q4 2026 despite projecting over $200 billion in cumulative losses, creating friction with CFO Sarah Fryer over what she calls "extraordinarily ambitious plans."
- The tension reflects a deeper unresolved question: frontier AI companies function as both capital-intensive infrastructure and software businesses, with no consensus on appropriate financial models or when spending trillions on marginal capability gains becomes irrational.
Summary
OpenAI plans to spend $121 billion on computing power for AI research in 2028 alone, translating to an $85 billion annual cash burn that year even after nearly doubling sales. These losses would exceed those of virtually any public company in history, according to Wall Street Journal reporting based on financial documents shared with investors.
The spending trajectory reflects an intensifying arms race. Both OpenAI and Anthropic are releasing model versions at accelerating cadence while pouring more resources into training runs. OpenAI's approach appears more aggressive than Anthropic's. Dario Amodei at Anthropic has signaled a more conservative stance—declining to spend recklessly to avoid insolvency—while OpenAI's strategy under Sam Altman appears more expansionist.
Internal tension over IPO timing
This spending posture has created friction inside OpenAI between Altman and CFO Sarah Fryer, according to reporting from The Information. Altman has committed the company to $600 billion in spending over the next five years and privately indicated he wants to take OpenAI public as soon as Q4 2026, despite expectations the company will burn more than $200 billion before generating cash. Fryer has voiced concerns about these "extraordinarily ambitious plans," and the tension has become structural: Altman has excluded her from some financial planning conversations, and she was recently reassigned to report to the head of applications rather than the CEO.
The unresolved model question
Underlying this clash is a deeper uncertainty about how to value and finance frontier AI companies. They function simultaneously as capital-intensive infrastructure plays (like railroads or utilities) and as software businesses with scaling economics. No consensus exists yet on the financial model or appropriate risk profile.
There is also the harder question of scaling limits. AI companies have benefited from stable scaling laws—each dollar of additional compute has produced predictable capability gains. But that curve is not infinite. At some point, spending trillions to extract marginal IQ points becomes economically irrational, like hiring an employee for $5 million a year to gain one point of intelligence. When that inflection arrives remains unclear, but it shapes the entire thesis around these IPO timelines and burn rates.