Is OpenAI the Poster-Child of AI, or the AI Bubble?
You would think that the company that brought us ChatGPT – the one that supposedly aces every math test thrown at it – would be able to do the math on its own business model and cash-flow forecasts. But apparently not.
You would think that the company that brought us ChatGPT – the one that supposedly aces every math test thrown at it – would be able to do the math on its own business model and cash-flow forecasts. But apparently not. According to a bombshell HSBC research note that leaked into the wild this week, OpenAI is staring at a $207 billion funding hole by 2030 even under absurdly heroic growth assumptions. That’s after counting every dollar of projected profit, every committed partner check, every debt facility, and every penny of cash on the balance sheet. In plainer English: the company has already signed contracts for more compute than the entire planet can plausibly pay it back for. The numbers are so eye-watering they almost feel like satire.... OpenAI has locked itself into at least $288 billion of cloud commitments with Microsoft ($250 bn) and Amazon ($38 bn) alone, with another rumored $300 bn Oracle deal floating around that somehow never quite makes the earnings calls. Add in CoreWeave, Nvidia chip pre-buys, AMD, Broadcom, and the rest, and credible estimates now put total infrastructure promises between $1 trillion and $1.8 trillion over the next decade. Once the GPUs are bolted into racks, HSBC thinks the annual “rent” bill hits roughly $620 billion – more than the current GDP of the Netherlands. To put that in perspective, HSBC’s model assumes OpenAI grows to 3 billion users (44 % of the non-Chinese adult population), converts twice as many of them to paid as today, grabs a third of a $386 bn enterprise AI market that may or may not exist, and still scoops 2 % of global digital advertising on the side. Even with that fairy-tale hockey stick, the company is still $217 bn short (including a prudent buffer). This is the same OpenAI whose CEO, Sam Altman, spent much of 2024 and 2025 telling anyone who would listen that: “Energy will stop being a constraint in a few years” “We’re going to raise $7 trillion for chip fabs and data centers” (yes, trillion with a T) “Intelligence is shortly going to be basically free and abundant” “The cost of intelligence is dropping 100× every 18–24 months, maybe faster” Pick any one of those quotes and try to square it with a cost structure that is not collapsing but exploding by orders of magnitude. Sam’s public narrative is that we are on the cusp of an intelligence explosion so profound that GDP will be measured in tens of trillions of new wealth. The private contracts his company actually signed tell a different story: a quasi-fixed rental bill that looks more like WeWork on Blackwell GPUs than the arrival of digital superintelligence. The Great Monetization Mirage Outside the hyperscalers (Microsoft, Google, Amazon, Meta) who can theoretically print dollars by renting their own spare capacity back to themselves, nobody else is making real money selling frontier AI yet. Anthropic is burning cash. Mistral is burning cash. xAI is burning cash. Inflection, Adept, and Character all sold for pennies on the cap-table dollar. The enterprise “hockey stick” everyone keeps putting in pitch decks stubbornly refuses to curve upward in the real world; pilots are everywhere, signed seven-figure deals are rare, and billion-dollar deals are mythological. Consumers, meanwhile, are happy to use ChatGPT, Claude, or Gemini for free forever. The moment you try to push them to $20/month Plus tiers, conversion rates collapse. HSBC’s “10 % paid” assumption would already be heroic at 800 million users; stretching it to 3 billion and then doubling it again feels like fan fiction. Reality vs. the Religion There are two AI markets right now. Market #1 is the thematic one: trillion-dollar market caps, $7 trillion chip-raising dreams, TED-talk standing ovations, and politicians promising to re-arrange the periodic table if that’s what superintelligence requires. It trades on slides that say “$15 trillion to global GDP by 2030” and “AGI by 2027-ish.” Market #2 is the one where companies have to pay actual power bills, sign take-or-pay cloud contracts, and figure out how to turn tokens into dollars before the cash runway turns into a smoking crater. In Market #2, the HSBC note lands like a precision-guided munition. The gap between the two markets has become so wide that serious people are now openly discussing whether Microsoft, Amazon, Oracle and the rest will have to “show flexibility” on hundreds of billions of dollars of contracted capacity. Translation: someone, somewhere, is going to have to eat the mother of all write-downs, stretch the payment schedules into the 23rd century, or quietly pretend that some of those headline contracts were more aspirational than binding. The Stargate That Ate the Future Industry insiders have taken to calling the coming wave of AI mega-data centers “Stargate” projects – multi-gigawatt campuses that make today’s hyperscale regions look quaint. The assumption baked into every one of those projects was that OpenAI would be the anchor tenant whose insatiable…
Read on MPC Markets Mosaic
