YouTube accuses OpenAI of burning billions

The story here is AI economics — not whether people like ChatGPT, but whether the business underneath it can ever look normal. That question got sharper over the last few months as new reporting sketched out just how much OpenAI expects to spend on compute, while the company and Microsoft rewired a partnership that had become both a growth engine and a constraint. The basic tension is simple. OpenAI is growing fast. But growth in AI is expensive in a way normal software growth usually isn’t. ### Why do people keep saying OpenAI is burning cash? Because the core product is costly to run. Every model training cycle eats huge amounts of chips, power, networking, and data-center capacity. Then inference — actually serving answers to users — keeps the meter running every time someone uses ChatGPT or an API call. That is why OpenAI’s finances keep getting described less like software margins and more like industrial buildout economics. (cnbc.com) ### What changed recently? Two things. First, investor-facing reporting in February said OpenAI reset expectations and told investors it was targeting roughly $600 billion in total compute spend by 2030, after earlier talk around a much bigger $1.4 trillion infrastructure ambition started to look too loose. Second, on April 27, OpenAI and Microsoft announced a revised partnership that lets OpenAI serve customers across any cloud provider, even while Azure stays the primary cloud partner. (cnbc.com) ### Why does the Microsoft relationship matter so much? Because Microsoft has been more than an investor. It has supplied a substantial share of OpenAI’s financing and compute, and OpenAI itself flagged that dependence as a risk in investor materials circulated this year. That is the catch with hyperscale AI — if one partner provides the money, the cloud, and some of the distribution, that partner is also a bottleneck. The new agreement loosens that a bit by making Microsoft’s license non-exclusive and allowing OpenAI to use other clouds. (cnbc.com) ### Is OpenAI actually bringing in serious revenue? Yes — and that is what makes this story interesting instead of fatal. OpenAI hit $10 billion in annualized revenue in 2025, and earlier this year it was reported to be aiming for $12.7 billion in 2025 revenue. So this is not a zero-revenue moonshot. It is a fast-growing company trying to outrun an even faster infrastructure bill. (cnbc.com) ### So why are people still worried? Because revenue growth alone does not settle the question if the spending curve stays steeper. In April, reporting said OpenAI had missed some internal revenue and user-growth estimates, raising concern about its ability to fund future compute agreements. Basically, if the company keeps signing giant capacity deals, small misses start to matter a lot more. (cnbc.com) ### Is this just an OpenAI problem? Not really. OpenAI is the clearest version of a broader AI problem. The whole sector is trying to find out whether generative AI behaves like software — high margin once built — or like utilities and telecom, where scale demands constant capital spending. OpenAI just sits at the center because it has the biggest brand, the biggest visible partnerships, and the clearest appetite for compute. (cnbc.com) ### What would make the economics look better? Higher-priced enterprise usage, more efficient models, cheaper inference, and packaging that pushes customers toward predictable recurring spend instead of novelty traffic. The recent Microsoft reset helps too, because more cloud flexibility should improve OpenAI’s leverage when sourcing capacity. But none of that changes the core fact that frontier AI still runs on very expensive physical infrastructure. (cnbc.com) ### Bottom line? OpenAI is not collapsing. It is doing the harder thing — trying to turn a hit product into a capital-intensive platform business. That can work. But the bill is now big enough that “growth” by itself is no longer the whole story. (cnbc.com) (openai.com)