Internal memo: OpenAI projects $14B loss in 2026 after missed revenue and user targets, reviews compute spend

OpenAI’s problem right now is not that people stopped using AI. It’s that the business underneath the hype is getting harder to explain. In late April, reports said OpenAI had missed some internal sales and user targets for 2026, which immediately raised a bigger question: if even the category leader is struggling to hit its own plan, what does that say about the economics of generative AI? OpenAI’s finance chief, Sarah Friar, pushed back days later and said the company is still beating its plan at the highest level and sees a “vertical wall of demand” for its products. (bloomberg.com) ### Why did this story hit so hard? Because OpenAI has become the shorthand for the whole AI trade. When a report says the company missed internal targets, investors don’t hear “one startup had a forecasting miss.” They hear “maybe the biggest AI company is finding it harder than expected to turn usage into predictable revenue.” That is why AI-linked stocks sold off when the report landed. (bloomberg.com) ### What was the reported miss? The reporting pointed to shortfalls in new user acquisition and sales goals, especially in monthly targets during 2026. The pressure point seems to be enterprise and coding markets, where Anthropic has been gaining ground. That matters because those are exactly the segments investors expect to produce sticky, high-value revenue rather than consumer buzz alone. (bloomberg.com) ### So is demand actually weak? Probably not in the simple sense. Friar’s rebuttal is basically that the company is still growing fast, but the mix shifts around from period to period because the business is young and hard to forecast cleanly. Both things can be true at once — demand can be huge, and internal targets can still be missed if the company guessed wrong about where growth would show up or how fast deals would close. (bloomberg.com) ### Why does compute keep showing up here? Because OpenAI is not just selling software. It is also trying to secure enough chips, data centers, and power to keep training models and serving millions of users. That turns the company into something closer to a software-plus-infrastructure hybrid. OpenAI raised $40 billion at a $300 billion post-money valuation in March (bloomberg.com)Stargate, the giant data-center project backed by SoftBank, Oracle, MGX, and OpenAI. (openai.com) ### Why is that a tougher business? Because software investors love gross margins. Infrastructure eats them. Every extra user is not just another subscription — it can mean more inference cost, more GPUs, more networking, and more power. The catch is that AI demand can rise at the same time profitability gets worse. It’s a bit like running a wildly popular restaurant where every extra table also forces you to build another kitchen. Tha(openai.com). ### Where does the IPO angle come in? An IPO works best when the story is simple: growth is strong, margins are improving, and forecasts look believable. This story muddies all three. Bloomberg has already framed OpenAI as moving toward a possible mega-IPO, but public investors will care less about abstract AGI upside and more about whether revenue timing, compute spending, and competitive pressure are under control. (bloomberg.com) ### And the competition? It’s getting sharper. Anthropic is pressing in coding and enterprise. Big cloud platforms and model companies are also trying to own distribution, not just model quality. That means OpenAI has to win on product, pricing, and access at the same time its cost base is still unusually heavy. (bloomberg.com)del growing up. OpenAI may still be huge. Demand may still be real. But the market is starting to ask a more adult question — not “is AI important?” but “who can make money doing this at scale?”