OpenAI anticipates a substantial cash burn of nearly $280 billion by the end of 2030, as reported by the Financial Times. This projection indicates that the company's expenses are expected to significantly exceed its revenues over the same period. While revenues are forecast to grow from $36 billion this year to $350 billion in 2030, the associated costs, particularly for cloud compute, are a major factor in the expected deficit.
HSBC analysts have elaborated on OpenAI's financial situation, estimating that the company will need to raise at least $207 billion by 2030 to cover its operational costs and maintain liquidity. This figure accounts for massive data center rental bills, including a cumulative $792 billion in compute costs between the current year and 2030. The estimate incorporates $250 billion from a Microsoft deal and $38 billion from an Amazon deal, bringing total contracted compute power to 36 gigawatts.
Despite projections for significant revenue growth, driven by an S-curve user adoption reaching 3 billion by 2030 (44% of the world's adult population outside China) and an increase in paying customers to 10%, OpenAI is still expected to operate at a loss. HSBC's model forecasts a cumulative free cash flow of approximately $282 billion by 2030, which, even with additional injections from Nvidia and other facilities, leaves a funding gap of $207 billion, plus a $10 billion cash buffer for safety. This financial strain highlights the immense capital required to scale AI infrastructure.
The cost concerns are further emphasized by reports suggesting OpenAI's inference spend on Microsoft Azure alone was nearly $5 billion in the first half of 2025, with total inference compute costs exceeding $12.4 billion over seven calendar quarters. This spend appears to significantly outpace reported revenues, raising questions about the sustainability of the current business model for OpenAI and other general-purpose LLM vendors, implying that either costs must drastically decrease or customer charges must rise substantially.