Is There an AI Bubble? Yes, and It Doesn't Matter

Is there an AI bubble in 2026?

An AI bubble exists, and pretending otherwise requires ignoring the tape. Ian Kilpatrick put the question to the room directly on Session 009, and Justin Novak's answer holds the whole post: locally, right now, there is a bubble. On the long-term horizon, we are very, very early.

Both halves matter. A bubble is a statement about prices, not about whether the technology works. Getting that distinction wrong cost people fortunes in 1999, in both directions.

“right now, locally, there's a bubble, but on the long-term, mid-term time horizon, we're very, very early.”

Justin Novak · 20:41

What does the AI bubble actually look like?

The froth is concrete and countable. Justin's benchmark from the previous era: Y Combinator used to hand founders 100 to 200K for 7.5% of the company, and a million-dollar seed round was a lot. Today he watches companies raise $100 million seed rounds, pre-product and pre-revenue, with just a team.

Credentials alone now price in the hundreds of millions. Leave a research role at OpenAI or Anthropic and a nine-figure seed round follows your name, before a product exists. That, in Justin's words, is indicative of a bubble, at least in private markets.

The exclamation point is the rumored Anthropic IPO at $2 to $3 trillion. Nothing that size has ever gone public. When the biggest IPO in the history of the world is the ambient background news, prices are running hot.

How is this like the dot-com bubble?

A bubble does not mean the technology is fake. Justin's parallel is the dot-com era: Pets.com went public without making money, valuations were simply overinflated in the late 90s and early 2000s, and yet the internet produced tons of incredible businesses on the other side. The prices were wrong. The thesis was right.

One structural difference came from Alec Saluga: in the dot-com bust the major players were public, while today's AI giants are still mostly private. The froth lives in venture rounds rather than public tickers, which changes who absorbs the losses when prices correct, not whether the technology survives.

“So I think we're in a bubble, but that doesn't mean it's fugazi.”

Justin Novak · 22:28

Where is the real value in AI?

Proof of value showed up the same day as the discussion. Moderna announced a personalized melanoma treatment, tailored to the individual patient, that passed Phase 3 trials, and the stock roughly doubled on the news. A week earlier, people at a party were laughing at the idea that AI-era medicine might touch cancer. Justin told both stories back to back.

Infrastructure tells the same story. Databricks, which provides cloud and infrastructure for AI training, raises at valuations backed by real revenue. Anthropic's growth rate and run rate are, in Justin's words, complete gangbusters. Justin's summary sentence deserves quoting in full: is there a financial bubble in private markets? Absolutely. Is there real value creation happening? Absolutely.

What should a business owner do about the bubble?

Operators and investors face different risks here. The bubble threatens people holding inflated paper: late-stage investors, employees valuing options at peak marks. An owner using AI to cut costs or ship faster holds none of that paper. If the bubble pops, the tools do not get worse. Historically they get cheaper.

Ian goes one further: he is hoping the bubble bursts. Hype leaving the room means the tourists leave with it, and the people who see the real value get a longer runway to build. AI4NTP (AI for Non-Techy People) exists on that side of the trade: the technology compounds for its users regardless of what the valuations do. Use the tools, skip the paper.

“I'm kind of hoping that there is a bubble, and it bursts, and people kind of lose interest, because then guys like us get a longer runway to build awesome things.”

Ian Kilpatrick · 23:33