OpenAI’s $750 Billion Problem Is Not Compute—It Is Cash Conversion OpenAI’s projected compute bill has climbed from roughly $600 billion to $750 billion through 2030, after Sam Altman had already floated $1.4 trillion of capacity commitments. The market should stop treating this as a pure technology story, because the binding constraint is becoming finance. The number needs parsing Projected spending is not the same as signed contractual commitments, and signed commitments are not the same as cash paid today. A meaningful share will be phased across years, while cloud partners build and operate much of the underlying capacity. That distinction kills the lazy take that OpenAI needs to find $750 billion in cash tomorrow. It does not kill the risk, because every reserved gigawatt creates a future revenue and margin hurdle. Growth is necessary, not sufficient OpenAI reportedly passed $25 billion in annualized revenue in February, versus $13 billion of actual revenue in 2025. That is exceptional growth, but it is still nowhere near the scale of the infrastructure being secured. More important, inference costs quadrupled in 2025 while adjusted gross margin fell to 33%. An AI platform can add users, revenue and workloads while making its funding problem worse if each new dollar of demand consumes too much compute. That is why CFO Sarah Friar’s concern matters: the danger is not weak demand today, but locking in long-duration capacity before durable unit economics are proven. The IPO test A potential IPO would move this debate from private-market storytelling to public-market disclosure. Investors will want to know how much of the $750 billion is firm, cancellable, take-or-pay, partner-financed or contingent on delivery. AWS, Azure and Oracle gain demand visibility from these deals, but also exposure to one customer’s ability to keep raising capital and converting usage into cash. Altman’s strategy is rational if scarce compute secures an unassailable product lead. It becomes dangerous if model pricing compresses, inference remains expensive, or competitors achieve similar performance with less capital. Bottom line: OpenAI does not need $750 billion today, but it must prove each incremental dollar of compute can produce durable, high-margin revenue before those commitments mature.