NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

$7.7 billion in, and not one molecule has reached a human

Isomorphic Labs raised a billion Series C on 17 November, bringing total funding to .7 billion in 20 months. Its reported 0 billion valuation rests on programmes that have not yet reached a single patient.

2026-11-20 · 4970 words · NeuroAI
$7.7 billion in, and not one molecule has reached a human

The first paragraph of the press release is a row of investor names.

Greenoaks, Sequoia Capital, Andreessen Horowitz, Alphabet, Thrive Capital, MGX, GV, CapitalG, Temasek, the UK Sovereign AI Fund.

Ten names, one round.

Ten names on one line

On 17 November, Isomorphic Labs announced a $5 billion Series C. Greenoaks and Sequoia co-led, a16z came in as a new face, and Alphabet put in more.

The other six were existing holders, and every one of them followed.

Lead investors get written up everywhere; whether the old shareholders stayed is usually buried in the last paragraph. All of them staying means the people who have seen the data and know where the pipeline is stuck still chose not to sell.

The company was spun out of DeepMind in 2021 and is headquartered in London. Its founder, Demis Hassabis, shared the 2024 Nobel Prize in Chemistry. The technical foundation comes straight from AlphaFold, the system that predicted protein structures.

Twenty months, $7.7 billion

Three rounds, close together.

March 2025: $600 million, the first outside money. May 2026: $2.1 billion, Series B. 17 November: $5 billion, Series C.

Six plus 21 plus 50 is $7.7 billion, inside 20 months.

Each round was larger: six to 21 is 3.5 times, 21 to 50 is 2.4 times. Only six months separated the last two. The valuation was not disclosed.

That is not the rhythm of executing a research plan. It is the rhythm of taking a window — before a competitor reaches the clinic, before appetite cools. Funding windows are not taps. When one opens, you fill up.

An $82.5 million down payment

Isomorphic has deals with Eli Lilly, Novartis and Johnson & Johnson.

In early 2024, Lilly agreed to $45 million upfront with up to $1.7 billion in milestones. Novartis followed with $37.5 million upfront and up to $1.2 billion in milestones. J&J joined in 2026; the terms were not disclosed.

For the two disclosed deals, the upfront payments total $82.5 million and the milestone ceilings total $2.9 billion.

The upfront is 2.8 percent of the headline.

That ratio is normal. A drugmaker pays a small sum for an option and the large sum only if the work validates, capping the loss at the upfront. So the phrase "nearly $3 billion in partnerships" describes $82.5 million of actual cash. Put that $82.5 million next to the $5 billion round and it is 1.65 percent — the round is roughly sixty times two upfronts. Milestones are not receivables. They are possibilities.

Fifty percent against twenty-three

The company has disclosed one comparison: on a protein-ligand structure prediction benchmark, its engine IsoDDE scored about 50 percent, against 23.3 percent for AlphaFold 3.

More than double.

That gap deserves unpacking. Structure prediction asks how a molecule and a protein fit together; molecular design asks whether a molecule exists that binds, is drug-like, can be manufactured and can go into a person. The first is reading a picture, the second drawing one. Fifty percent says the first gate clears well. It says nothing about the other three.

Hassabis's own phrasing is that the engine is generating molecules "with unprecedented precision and speed".

Twenty molecules, zero patients

Hassabis has said the company has more than 20 programmes and expects its first clinical trial to start before the end of 2026.

Today is 20 November. Forty-one days remain.

As of the announcement, no Isomorphic drug has entered human trials. Every disclosed programme sits in preclinical. The company says internal and partnered work in oncology and immunology has produced positive preclinical results.

Preclinical to Phase I is the stretch money cannot compress.

Phase I does not turn on algorithms. It turns on ethics committees, investigators, recruitment and toxicology. Whether an oncology patient signs a consent form has nothing to do with a benchmark score; when an ethics board meets has nothing to do with the size of a round. Starting a first case within 41 days is the hardest deliverable on the 2026 calendar.

Whose $50 billion is it

The company did not disclose a valuation. Bloomberg reported a target of about $50 billion.

Fifty billion is ten times the round.

One side is revenue, the other probability. Both can be called $50 billion; one is priced off earnings, the other off odds.

Where the money goes

The stated uses: iterate IsoDDE, advance internal and partnered pipelines, build AI-first translation and clinical infrastructure.

The third is the one people read past.

Building clinical infrastructure means owning the trial. The old model was to design molecules, license them out, and let the drugmaker run the studies. Owning it means hiring clinical staff and absorbing the failures yourself: a move from selling blueprints to putting up the building.

That is also where the valuation logic shifts. Pure design caps out at licence fees and milestones. Owning the clinic caps out at a drug. The price is inheriting the cost and the timeline of the layer that is least responsive to software.

A drug is the slowest product in this industry: toxicology, three phases, review. Stop at any gate and everything spent is stranded. Isomorphic has taken that distance onto itself and been marked up for it — the market is betting it can walk the route with fewer failures.

The last step has no accelerator

What AI can accelerate here is clear: target finding, molecule generation, structure prediction, binding affinity. All of it is work on paper, and more compute makes it faster, measured in months.

What it cannot accelerate is equally clear: ethics review, enrolment, long-term safety follow-up. Measured in years.

So the real bet is not on the model. It is on time — that a molecule reaches a human, and comes back out the other side, before the money runs out. The $7.7 billion bought that interval.

In 41 days, the first sentence of that answer is due.

Disclaimer: This article is based on public information and is for reference only; it is not investment advice.

Related coverage

Cite this article

NeuroAI (脑机网), "$7.7 billion in, and not one molecule has reached a human," 20 November 2026, https://neuroai.site/a/na-capital-77yi-1120

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Capital & Markets” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories