Isomorphic Labs seeks $40B+ and has never disclosed a valuation

Isomorphic Labs, the Alphabet drug discovery company, is reportedly in early talks to raise money at a valuation of at least $40 billion, and it has never disclosed a valuation for any round it has announced. Both of its published financings, the $600 million first external round dated 31 March 2025 and the $2.1 billion Series B dated 12 May 2026, state the amount and stop there, so the $40 billion figure rests entirely on unnamed sources.

Bloomberg reported the talks on 8 October. Its own headline says “at least $40 Billion”, and one of its sources said the round could reach as much as $50 billion, as Tech Funding News quoted it. Bloomberg’s page returns a 403 to us, so we’re working from the outlets that quote it instead. Semafor ran the $40 billion number alone, while SiliconANGLE led on “up to $50B”.

Event	Date on the primary	Amount	Valuation stated
First external round	31 March 2025	$600 million	None
Series B	12 May 2026	$2.1 billion	None
Reported talks	8 October 2026	Not specified	$40B to $50B, per sources

The gap between a reported number and a stated one is the whole story here, which is the same trap we unpacked in how to read a round announcement. SiliconANGLE also placed the $2.1 billion round “in March”, but Isomorphic’s announcement page is dated 12 May 2026. That March date belongs to the $600 million round a year earlier. Thrive Capital led both.

The Series B page names its backers in full. Thrive led, Alphabet and GV returned as existing investors, and MGX, Temasek, CapitalG and the UK Sovereign AI Fund came in new. That last name matters, because a sovereign fund now sits in a company whose worth nobody has published.

What the February report actually claims

SiliconANGLE wrote that in February “Isomorphic stated that IsoDDE was more than 50% better at finding binding pockets than the closest competing tool”. That claim isn’t in the February post, though. The IsoDDE write-up, dated 10 February 2026, gives no percentage for pocket finding at all. It says blind pocket identification reaches “performance levels approaching experimental techniques like fragment-soaking”.

The headline accuracy number is narrower than the coverage suggests, because the conditions matter. IsoDDE “more than doubles the accuracy of AlphaFold 3 on the most difficult systems”, meaning one bin of an external benchmark, the 0-20 similarity bucket of Runs N’ Poses. That’s the subset least like the training data, not the task in general. On antibody-antigen structures it beats AlphaFold 3 by 2.3x and Boltz-2 by 19.8x, but only in the high-fidelity regime where DockQ exceeds 0.8.

Runs N’ Poses isn’t Isomorphic’s benchmark. Škrinjar and colleagues built it with 2,600 high-resolution protein-ligand systems released after the models’ training cutoffs, and published it on bioRxiv. Their conclusion was unflattering to the field.

We demonstrate that current cofolding approaches largely memorise ligand poses from their training data, hindering their use for de novo drug design.

Škrinjar et al., Runs N’ Poses, bioRxiv

Measuring a model on that benchmark is therefore a stronger move than measuring it on an in-house one, which cuts in Isomorphic’s favour. The point is that the result is conditional, and the conditions vanished in the retelling. If you want the wider picture on model-led biology at Alphabet, our write-up of AlphaGenome Atlas covers the sibling effort at Google DeepMind.

The revenue side is also a potential figure

Isomorphic’s pharmaceutical deals get quoted as “nearly $3 billion”, and its own 7 January 2024 page says the Eli Lilly and Novartis collaborations “have the potential to be worth nearly $3 billion to Isomorphic Labs, excluding any royalties”. But the upfronts were far smaller: $45 million from Lilly and $37.5 million from Novartis, so the rest is milestone money. A third collaboration, with Johnson & Johnson, was announced on 20 January 2026 with no figure attached.

One verifiable claim in the February post does survive intact. IsoDDE recapitulated a cryptic allosteric site on cereblon that Dippon and colleagues found experimentally and published in Nature, using only the protein sequence as input.

So what to watch is whether a closed round finally carries a number. Isomorphic’s announcements page listed nothing about the talks as of 10 October, and its most recent entry is a 7 October note on joining the Virtual Biology Initiative. Sir Demis Hassabis is listed as founder and chief executive, with Max Jaderberg as president.