ByteDance has spun off its AI-for-biology unit into an independent company, Anew Labs, and closed a $290 million Series A round at a $1.5 billion valuation, Reuters reported September 16, citing people familiar with the deal. ByteDance keeps a reported 56% stake in the newly independent company. Neither ByteDance nor the investors responded to Reuters' requests for comment, and the sources spoke on condition of anonymity because the round is not yet public -- this is sourced reporting, not a company announcement.
The round was led by HSG (formerly Sequoia China), IDG Capital and Hillhouse Investment, with 5Y Capital co-leading alongside Gaorong Ventures, Primavera Venture Partners, Boyu Capital, the state-backed Shanghai Future Industries Fund, and SBP Group as a strategic investor. Anew Labs runs operations in Shanghai, Singapore and San Jose. The stated reason for the spinoff, per Reuters' sourcing, is that AI-driven drug discovery runs on a different industry logic -- longer timelines, different regulatory exposure, different investor base -- than ByteDance's core apps business.
What makes Anew Labs more than a funding headline is that it already has something to show, and it isn't a drug. Protenix is a biomolecular structure-prediction model the team develops openly on GitHub under an Apache 2.0 license -- free for commercial and academic use, code and trained weights both. In its own bioRxiv preprint, the team describes Protenix-v1 as "the first fully open-source model that outperforms AlphaFold3 across diverse benchmark sets" -- Google DeepMind's own structure-prediction model, the one whose 2024 results won Demis Hassabis a Nobel Prize -- while using the same training-data cutoff, model scale and inference budget as a fair comparison. A follow-on model, PXDesign, designs protein binders from scratch and reports 20-73% experimental success rates across tested targets, 2-6x the rate of prior methods including AlphaProteo and RFdiffusion.
(A structure-prediction benchmark measures whether a model's predicted molecular shape matches reality -- it says nothing about whether a resulting drug candidate is safe or effective in a living body. That gap, between computational prediction and clinical proof, is the one every AI-drug-discovery company in this space still has to cross, no matter how the modeling benchmarks land.)
- Anew Labs raised $290M at a $1.5B valuation, with ByteDance retaining 56%.
- Protenix-v1 outperforms AlphaFold3 across diverse benchmark sets.
- PXDesign achieves 20-73% experimental success rates on protein-binder design, 2-6x prior state of the art.
- Anew Labs has 'a small number of' actual drug candidates in its pipeline.
Anew Labs isn't the only Big Tech AI-biology spinoff on the market this year, and the company it sits next to is a useful scale check. Alphabet's Isomorphic Labs -- also built on a Nobel-winning structure-prediction lineage, AlphaFold -- announced a confirmed $2.1 billion Series B in May 2026, bringing its total disclosed capital raised to roughly $2.6 billion.
Two Big Tech AI-biology spinoffs, a very different scale of bet
| Anew Labs spun off from ByteDance | Isomorphic Labs spun off from Alphabet/DeepMind | |
|---|---|---|
| Latest round | $290M Series A (reported, unconfirmed) | $2.1B Series B (announced May 2026) |
| Reported/disclosed valuation | $1.5B | Undisclosed; total capital raised to date ~$2.6B |
| Flagship model's license | Protenix -- open-source, Apache 2.0 | AlphaFold3 -- source available, more restricted commercial terms |
| Named pharma partners | None reported | Novartis, Eli Lilly, Johnson & Johnson |
The contrast is the actual story underneath the funding number. Isomorphic Labs -- also a Big Tech AI lab's spinoff into drug discovery, also built on a Nobel-winning structure-prediction lineage -- raised seven times as much in its most recent round, at a scale its own backers confirmed on the record, and already has named pharma partners running programs against its models. Anew Labs, at a tenth of the disclosed capital and with terms still unconfirmed by either company, is betting that giving its structure-prediction model away for free builds the same kind of ecosystem gravity Isomorphic is buying with partnership deals -- a genuinely different strategy, not just a smaller version of the same one.
- Diversifies beyond TikTok and ad revenue into a second AI-native business line, without ByteDance itself carrying the regulatory and clinical-trial risk directly.
- Are betting on a $1.5B valuation built almost entirely on modeling benchmarks and an unnamed drug pipeline -- the harder, slower, more expensive clinical-proof stage hasn't started for any disclosed candidate.
- Get a free, state-of-the-art structure-prediction tool regardless of how the funding round or the company's drug pipeline turns out -- the Apache 2.0 license doesn't depend on Anew Labs' business succeeding.
None of that makes the $290 million figure real until someone on the record says so. What is independently checkable, right now, without waiting on either company, is Protenix's code, its weights, and its benchmark claims against AlphaFold3 -- open on GitHub for anyone who wants to run the comparison themselves. That is the part of this story that doesn't depend on Reuters' sourcing being right.
- Reuters reported Anew Labs, ByteDance's spun-off AI drug-discovery unit, closed a $290M Series A at a $1.5B valuation.
- ByteDance keeps a reported 56% stake; HSG, IDG Capital and Hillhouse led the round.
- Anew Labs' concrete output is Protenix, a free, open-source protein-structure model it says beats AlphaFold3 on some benchmarks.
- PXDesign, built on Protenix, claims 20-73% experimental success designing protein binders -- 2-6x prior methods, per the team's own paper.
- Caveat: neither company has confirmed the funding terms, and the AlphaFold3 comparison is Anew Labs' own preprint, not independently replicated.