Astromech Raises $20m at a Reported $3.8bn Valuation

Astromech Sold Roughly Half a Percent of Itself for $20m. That Is What Prices the $3.8bn.

The Colossal spin-out has raised $60m in total and reports a valuation 63 times that. Between the two sits a candidate gene list.

Astromech, the evolutionary-biology AI company spun out of Colossal Biosciences by Ben Lamm and Harvard geneticist George Church, closed an additional $20m on 20 August. Total funding to date is $60m, and the company "reports its valuation at $3.8bn." Bob Nelsen, co-founder and managing director of ARCH Venture Partners and an existing Colossal investor, led the round; PEAK6, NeoGenesis Capital, Builders VC and CAZ Investments also participated.

The arithmetic nobody runs

Twenty million dollars against a $3.8bn post-money is a sale of roughly half of one percent of the company. A round that small does not price a business in any ordinary sense — at any other company it would read as a seed extension. The gap between money in and paper value is 63-fold: $60m raised in total, $3.8bn claimed.

What the common framing gets wrong

Coverage saying Astromech "hits a $3.8bn valuation" treats a 0.5% insider sale as a market verdict. It is not: it is a number agreed between a company and an investor who was already on the cap table of its parent. Note also the phrasing in the announcement itself — the company reports its valuation, which is not the same as a lead investor independently marking it. A widely repeated prior mark of $2bn from March is not stated in the primary account and is not asserted here.

What actually exists

The disclosed technical asset is a model trained on Colossal's multispecies genome bank, producing a cross-species map of longevity-associated genes, with evolutionary outliers such as elephants, bowhead whales and bats. The company describes itself as preparing forecasting pilots in health and biosecurity. There is no disclosed revenue, headcount, peer-reviewed publication, clinical programme or commercial product. Church has been publicly candid about the core limitation: comparing living species by genome can fail to connect DNA differences to functional impact — which is the hard part of the whole enterprise.

The pattern

This is a clean specimen of how AI-for-biology is being priced in 2026: valuation set by tiny insider rounds, scientific ambition stated in the present tense, and the deliverable between them a list of candidate genes.

Why It Matters

Headline valuations in AI-for-biology are increasingly set by rounds too small to constitute price discovery. Anyone reading them as evidence that the science is working is reading the wrong signal.