OpenAI Expects to Burn $278 Billion in Cash by 2030 as Compute Bill Reaches $856 Billion
OpenAI has told investors it expects to consume $278 billion in net cash between 2026 and 2030, according to a company presentation reviewed by the Financial Times and reported by Reuters on September 18. The same document projects $856 billion in gross spending on computing power and infrastructure over that window — the company’s single largest expense category.
The numbers land in a different register than prior OpenAI financial disclosures. The company’s audited 2025 accounts showed research and development costs exceeding $13 billion in revenue. The new projections extend that trajectory five years out: roughly $55 billion per year in net cash consumed, and more than $170 billion per year in compute and infrastructure capex by the end of the decade.
What the Numbers Imply
The $278 billion figure is net — cash position change after revenue. The $856 billion is gross compute and infrastructure outlay. The gap implies OpenAI is projecting substantial revenue growth that offsets much of the gross spending, but still leaves a multi-hundred-billion-dollar hole to fill.
At current venture and commercial-credit scales, that gap requires either persistent revenue growth well above current trajectory, additional equity raises of a magnitude that would dwarf previous rounds, or debt financing against projected revenue. Likely all three.
The $856 billion compute projection places OpenAI on a spending curve that rivals the capital plans of the largest hyperscalers. For comparison, Microsoft, Google, and Amazon combined are guiding toward more than $500 billion in AI and cloud capex for 2026 alone. OpenAI’s number is cumulative over five years, but for a company without data center ownership it implies a sustained dependency on hyperscaler partnerships — primarily Microsoft Azure — or an acceleration of its own infrastructure buildout.
Context
OpenAI’s compute spending has scaled roughly in line with the company’s public statements over the years, from $30 million annually in 2017 to figures in the tens of billions by the mid-2020s. The new internal projections extend that curve steeply. GPT-5 series training runs consumed an estimated 2-2.5x the compute of GPT-4, with each successive generation compounding the infrastructure demand.
The FT presentation suggests OpenAI’s leadership is socializing these numbers with investors and partners ahead of further capital raises. The company converted to a for-profit structure in 2025 and completed a $40 billion round at a $300 billion valuation. If the 2026-2030 cash burn projections hold, that capital covers less than 15% of the projected five-year gap.