Baseten is worth $13 billion. Five months ago it was worth $5 billion. Five months before that, at its $300 million round, it carried a $5 billion tag — the jump is in the raise itself: $1.5 billion in, valuation to $13 billion, roughly 5x in a single re-pricing window.
The number underneath it is the tell: an unaudited third-party estimate of about $600 million in annualized revenue as of March 2026, up from $200 million at the end of 2025. That's a ~1900% year-over-year figure, and the word doing the heaviest lifting in that sentence is "unaudited."
So when PitchBook drops the line everyone will quote this week — nearly two-thirds of all AI venture dollars going to a small handful of companies — resist the reflex to read it as a scoreboard.
The sharper read came from a reply under that post: this isn't "AI is getting all the money," it's that a few frontier companies are. Everyone else is playing a completely different game, in a completely different market.
That distinction matters because the concentration stat is being misread as a moat. It isn't. It's a velocity problem.
Last issue we tracked the Carta data on capital concentration — the widening gap between the funded few and the starving many. This week the story moves: the concentration is now global, and the exit-and-distribution angle has shifted to how fast the top tier re-prices inference and physical AI.
Look at the clock, not the headcount. Baseten's 5x wasn't a slow accretion of enterprise logos; it was a five-month step function on numbers no auditor has signed. General Intuition is being priced at $6 billion pre-money "just weeks after" it raised $320 million at $2.3 billion — a near-triple, in weeks, on a round that's oversubscribed.
Weeks. Not quarters. The market is not queuing patiently outside these companies; it is bidding against itself in real time.
That's the thing the concentration framing hides. If you only count dollars-per-name, you conclude the winners are entrenching. If you count time-per-re-pricing, you see something more fragile and more interesting: capital is repricing entire layers of the stack faster than fundamentals can be verified.
Capital is repricing entire layers of the stack faster than fundamentals can be verified — concentration and diffusion happening simultaneously, and the second is the part nobody's pricing.
And it's repricing layers, plural — which cuts against the "same five names" story. Baseten is inference infrastructure, betting that inference costs will hit 85% of enterprise AI budgets. General Intuition is physical AI, training agents to move through space and time off hundreds of millions of hours of gameplay and button-press "action labels."
Different layer, different thesis, same tempo. The money is broadening across the stack even as it narrows to few names within each layer. Concentration and diffusion are happening simultaneously, and the second one is the part nobody's pricing.
Now widen the lens to Europe, because the pattern isn't a US quirk. It's fractal.
European AI startups raised a record $23 billion in H1 2026, up 130% year over year, per the HumanX–Crunchbase report. That's 55% of all venture capital across the region — a majority of the entire continent's VC now routing through one theme.
But 73% of that $23 billion went to just 38 companies raising $100 million or more — up from under half in recent years. The same top-heavy signature as the US, drawn on a smaller canvas.
Stefan Weitz of HumanX framed it as strength: "If the AI story was supposed to be a two-horse race, Europe didn't get the memo. $23 billion is flowing straight into the industries Europe already leads: robotics, healthcare, advanced manufacturing, defense."
Fine. But look at the tech.eu distribution data and the mood shifts. Over a recent 90-day window, the ten largest deals absorbed 47.2% of €23 billion. Average round size sat more than seven times above the median — €32.3 million versus €4.4 million.
That average-to-median gap is the statistical fingerprint of the whole story. A handful of mega-rounds drag the mean into the sky while the median deal — the actual, typical European AI company — closes below €5 million.
Two markets, one label. The frontier tier is being repriced weekly at eleven-figure valuations. The other 700-plus deals in that same quarter are splitting the leftovers at ground level.
Here's why the "moat" reading fails. A moat implies durability — that the leaders have locked in something competitors can't cross. But a 5x in five months on unaudited revenue, or a near-triple in weeks, isn't durability. It's momentum being mistaken for a fundamental.
Baseten's own bull case admits as much. If revenue doubles again in the next twelve months, today's $13 billion looks conservative. If growth stalls, the company faces a serious problem — and the hyperscalers, Amazon, Google, Microsoft, could optimize their own inference layers and crowd in.
That's not a moat. That's a coin flip priced at $13 billion, with the audited numbers still in the mail.
Which brings the 08/11 Carta thread to its next question. Concentration was the story then. Velocity is the story now — and velocity at the top is outrunning the audited fundamentals that an eventual OpenAI or Anthropic S-1 will eventually expose to daylight. When a company that raised at these tempos finally files, the reconciliation between "annualized," "estimated," "unaudited" and GAAP is where the repricing runs the other direction.
So what's the actionable read for a disciplined manager?
Stop hunting where the crowd is counting. The concentration stat directs everyone's attention to the frontier tier, which means that tier is being priced by a scarcity of attention, not a scarcity of opportunity. Thirty-eight European companies aren't soaking up 73% of the capital because only 38 companies are good. They're soaking it up because that's where the reflexive money knows to look.
The app layer and the non-mega tier — the 704 deals sharing 53% of the money, the median €4.4 million round — are being repriced downward by neglect, not by weak fundamentals. That's a different market, and it's the one the reply-guy was pointing at.
The tell that this is neglect and not judgment: Crunchbase forecasts that among 1,000-plus European AI startups funded since 2022 with $10 million or more, 59% will raise again within twelve months, 18% become M&A targets, 5% go public. And startups are already the most acquisitive buyers in the market — OpenAI alone has bought at least 19 companies, Anthropic five this year. Deep-pocketed unicorns flush with megaround cash are shopping the exact tier the headline capital ignores.
So the frontier names get repriced 5x in five months on faith. The tier below gets ignored on inattention — then quietly bought by the frontier names for the technology and teams the market underpriced.
The concentration everyone quotes this week is a distraction. The real signal is that valuation velocity is outrunning verifiable fundamentals at the top, and scarcity of attention — not scarcity of opportunity — is mispricing everything below it.
A sober manager doesn't need to win the auction for the next Baseten. They need to buy the median deal that the auction forgot, before the auction winner buys it first.