The AI Bubble Pops… Or Did It? (Spoiler: Probably Not)

by

When the Wall Street Journal dropped a bombshell that OpenAI had missed its user-growth and revenue targets, the real drama wasn’t about OpenAI itself-those shares aren’t even on the market yet. No, the real chaos was in the dependents. By Tuesday, Oracle was sulking, CoreWeave was weeping, and Nvidia, Broadcom, and AMD were all playing the “I’m not sad, I’m just disappointed” game. SoftBank, meanwhile, had a meltdown so dramatic it made a toddler’s tantrum look dignified. The stock most tied to OpenAI’s spending? Oracle, which has spent 48 hours whispering, “The demand curve has flattened… and it’s not a joke.”

The Journal’s report was as specific as a witch’s recipe. CFO Sarah Friar, who’s been quietly warning her colleagues that OpenAI might not be able to afford future computing contracts at their current revenue rates, called the situation “ridiculous” in a joint statement with Altman. “We’re totally aligned on buying as much compute as we can!” they declared, while the market rolled its eyes like a teenager hearing “I’m fine” after a car crash.

Meanwhile, on the other side of the AI universe, Anthropic was having a grand old time. Its annualized revenue run-rate hit $30 billion, comfortably ahead of OpenAI’s $24-25 billion. The company spent the past fortnight locking in compute like a kid hoarding candy. Amazon offered $25 billion (and a few gigawatts of capacity), Google followed with $40 billion (and more gigawatts), and Wilson Sonsini was already drafting IPO plans. Insiders refused to sell at February’s $350 billion tender, and now they’re pricing the company at $800 billion or more. Bloomberg says the IPO window is October. Or maybe November. Or perhaps it’s a trick question.

The Pentagon Problem

OpenAI is shipping faster than ever, and the WSJ numbers say users aren’t compounding the way the company’s spending plan needs them to. It’s like trying to fill a bucket with a leaky faucet… and the bucket is a spaceship.

Which leaves the IPOs. Both companies are racing the same window-late 2026, possibly slipping into early 2027. Anthropic has the harder operational job: scaling into $165 billion of pledged hyperscaler capital while convincing public-market investors that a $30 billion run-rate justifies an $800 billion-plus valuation. OpenAI has the inverse problem: defending an $852 billion private mark while its supplier stocks bleed 5% on a single Reuters wire. Jordan Klein, TMT sector specialist at Mizuho, captured the timing tension neatly: “How new could update be as the round closed end March when the quarter would have ended. And it’s not even May 1.” The implication? Whatever’s happening was already priced into the $122 billion round-and may not be priced into the IPO.

AI labs have real revenue, real enterprise penetration, and demand strong enough that the most valuable private companies in the world are rationing access to their flagship products. None of that was true of Pets.com. But the comparison is partly right. The hyperscaler-to-lab-to-hyperscaler money loop-Nvidia invests in OpenAI, OpenAI commits to Oracle, Oracle buys Nvidia chips, repeat-is the kind of circular financing that gets named after the fact. SoftBank dropping 10% on a single Reuters wire is what happens when an entire equity story rests on one customer hitting a growth curve. The cycle is still expanding but its margin of safety is gone. The next leg will either compress the multiple by producing a model that obsoletes the field, or confirm-with a series of polite quarterly misses-that AI has finally become a normal sector trading on normal multiples. For crypto, a slow re-rate redirects risk-on capital without breaking the broader market; a sudden one takes Bitcoin with it. It’s like a rollercoaster with no seatbelts… and the track is made of jelly.

Read More

2026-04-28 23:06