Analysts estimate OpenAI has amassed roughly $44B in cumulative losses

OpenAI’s problem is not demand. ChatGPT is huge, the API business is real, and enterprise customers are spending. The problem is that frontier AI eats capital at a speed normal software companies never have to deal with. That gap is why the old figure — roughly $44 billion in cumulative losses before profitability — keeps resurfacing as a useful way to understand the company. Recent reporting sharpened the pressure by showing OpenAI has missed some internal growth targets even as its compute deals keep getting bigger. (finance.yahoo.com) ### Where does the $44 billion number come from? It traces back to OpenAI’s long-range financial projections, first reported in late 2024 and still widely cited because they sketched the company losing about $44 billion from 2023 through 2028 before turning profitable in 2029. The exact framing matters — this is cumulative projected(finance.yahoo.com)ngry AI stack before the revenue base catches up. (finance.yahoo.com) ### Why can revenue be huge and losses still huge? Because OpenAI is not just selling software. It is also financing access to scarce compute — GPUs, data centers, networking, power, and the people needed to keep all of it running. Bloomberg-linked projections circulating in 2025 pointed to about $12.7 billion of revenue for 2025 an(finance.yahoo.com)he business model is scale-sensitive in an extreme way: a lot of gross demand, but even more spending needed to serve and expand it. (finance.yahoo.com) ### What changed this week? The newer wrinkle is not that OpenAI spends a lot — everyone already knew that. It is that recent reports said revenue and user growth missed internal targets, and that finance chief Sarah Friar has been working with other executives to tighten costs while the board looks more closely at compute agreements. Op(finance.yahoo.com) if growth slips even a little, giant infrastructure promises get harder to justify. (cnbc.com) ### How big are those infrastructure promises? Very big — and they have shifted. OpenAI announced Stargate in January 2025 as a project intending to invest $500 billion over four years in U.S. AI infrastructure, with $100 billion to begin immediately. Then, by early 2026, reporting said OpenAI was telling investors it was targeting (cnbc.com)er trillion-plus rhetoric. That is still a staggering number. (openai.com) ### Why is compute the whole story? Because compute is both the moat and the bill. Better models need more training and more inference capacity. More users also mean more inference cost every day, not just one-time R&D. So OpenAI faces a nasty loop: to stay ahead, it must buy more compute; to pay for more compute, it needs more users and higher-priced products(openai.com)toll road where every extra car also forces you to keep rebuilding the highway. (cnbc.com) ### Does this mean the business is broken? Not necessarily. The bullish case is simple: OpenAI already has real revenue at a scale most startups never reach, and if AI becomes core infrastructure for work, coding, search, agents, and enterprise software, today’s losses could look like front-loaded buildout costs. The bear(cnbc.com)pending curve stops looking visionary and starts looking reckless. (cnbc.com) ### Why does the IPO angle matter? Public markets are less forgiving than private ones. CNBC reported OpenAI was valued at $730 billion in a February 2026 fundraising round and is being discussed as a potential IPO candidate later this year. That means investors will care less about pure narrative and more about whether r(cnbc.com)ually want proof. (cnbc.com) ### Bottom line? The $44 billion figure matters because it captures the basic trade OpenAI made early: spend first, dominate first, and trust revenue to catch up later. That bet still might work. But now the company has to show that AI is not just a breakthrough product — it is a business that can absorb its own infrastructure appetite. (finance.yahoo.com)