When Mira Murati’s Thinking Machines Lab raised what was reported as a $2 billion seed round at a $12 billion valuation, the number was so far outside the historical distribution that it broke the word “seed.” But the outlier is not the story. The story is what is happening to everyone else — the thousands of companies that are not founded by a former OpenAI CTO, and the steadily emptying space between them and the giants.

The Data

The median U.S. seed round in 2026 sits at roughly $3.1 million, but AI startups now command a median seed deal size of about $4.6 million — a premium of more than a million dollars over the broader market. On valuation, the gap is wider still: AI companies are raising at roughly 1.6x the pre-money valuation of non-AI comparables, with median AI seed pre-money valuations near $17.9 million. Capital is not flowing evenly into “startups.” It is flowing into a label.

The more telling number is what is vanishing. According to Carta’s State of Pre-Seed for Q1 2026, rounds between $1 million and $2.5 million — the historical workhorse of early-stage financing — fell from 24% of all pre-seed rounds in Q1 2023 to just 18% in Q1 2026. That six-point compression is the signature of a barbell forming: weight piling up at the extremes, the bar in the middle getting thin enough to snap.

At one end, a small cohort of companies with a credentialed team or visible traction raises $3-5 million (or, in the extreme, billions) at premium valuations and finds the round oversubscribed. At the other end sits a long tail of companies that cannot close a priced round at all and survive on SAFEs, angel cheques, or nothing. The comfortable middle — the solid-but-unspectacular seed that defined the 2018-2021 era — is precisely the segment that record-setting top-decile valuations are pulling capital away from.

Why It Matters

For founders, the barbell is a positioning problem before it is a fundraising problem. The worst place to stand is the middle: good enough to be “interesting,” not exceptional enough to trigger the premium, raising into the exact band that is structurally shrinking. The instinct to look like a reasonable, de-risked, median company is now a liability — the median is where the financing is evaporating.

For investors, the premium is a discipline test. Paying 1.6x for an “AI” label is only rational if the label maps to durable advantage rather than a wrapper around someone else’s model. The barbell rewards conviction at both ends — back the genuine outliers, or back the deeply undervalued tail — and punishes the index-hugging behaviour of crowding into the middle at inflated prices.

The Charaka View

A barbell market is a sorting problem, and sorting is exactly what blind, multi-lens assessment is built for. The danger in a premium-driven environment is anchoring: a strong logo or an “AI-native” framing pulls a single analyst toward yes before the unit economics are read. Manthan’s Analytical Council commits each lens independently before any synthesis, which is designed to surface the tension between a story that earns the premium and numbers that do not. Across our knowledge graph, the companies that command durable premiums share a trait the label alone never captures — a proprietary data or distribution position that compounds. The barbell is not telling founders to raise more. It is telling them to be unmistakably one thing or the other, because the market has stopped paying for the middle.

This analysis draws on TechCrunch’s reporting on AI seed valuations (Mar 2026), Carta’s State of Pre-Seed Q1 2026, Carta’s record-setting valuations data, and Pitchwise’s median seed round analysis. Human editorial oversight applied.

This analysis is informational and does not constitute investment advice, a research report, or a recommendation to buy, sell, or hold any security.

Charaka Notes by Manthan Intelligence. Subscribe