Big pharma spends millions of dollars and many years hunting for the molecules that might treat a disease, and most of that hunt is a search problem hidden inside a laboratory. Nova turns that search into a public competition on Bittensor. Miners submit molecule candidates and search strategies, validators score them against a reference model, and the best submissions get rewarded. The compute is effectively cost-neutral to Nova, because each miner brings their own machine and only the winner gets paid.
The search space is the constraint, not the money
Five combinatorial reactions generate on the order of 61 billion possible molecules, and brute-forcing that space on standard infrastructure would take roughly 175 years. Nanobody space is larger again, by orders of magnitude. Throwing money at faster machines does not fix this, so the real work is inventing more efficient ways to search enormous chemical spaces and to know where to look. Once a search turns up promising candidates, they go to a physical lab for validation, the real gold standard, because a virtual benchmark only matters if the molecule behaves in the real world.
Why do it in the open
The case for decentralizing this is not idealism but efficiency, because doing it in the open buys Nova several things a closed lab structurally cannot:
- Adversarial pressure. Miners constantly poke holes in Nova's state-of-the-art models, and every submission becomes a signal about what works and what does not.
- A shared learning loop. Closed labs have no incentive to share data, so groups all over the world quietly relearn the same lessons in parallel, and Nova's loop is built to remove that redundancy.
- Interdisciplinary ideas. Abstracting discovery into a general search problem lets outsiders contribute. One striking example: a miner who borrowed an optimization strategy from energy-supply arbitrage and used it to beat Thompson sampling, a well-regarded published technique, across several targets, including a hard-to-drug oncology target.
The prize, and the incumbents
The reason serious money is chasing this is that the payoff on the other side is close to unbounded. The comparables are stark:
- Insilico Medicine has listed publicly, raised well over a billion dollars across its life, and carries clinical-stage assets.
- Isomorphic Labs raised $600 million, then a $2.1 billion round, one of the largest in biotech history.
- The two leading GLP-1 drugs now generate more revenue than the largest AI companies combined, a measure of the market Nova is aiming at.
Against an average of roughly $2.6 billion and ten years per approved drug, the pitch is that improving the very first step of the funnel, discovery itself, changes the economics of everything downstream. Rather than tweaking known compounds, which tend to be hard to patent because the surrounding chemical space is already claimed, AI can find genuinely new points of entry, which opens up both novel intellectual property and diseases that have no treatment today.
Where Nova is now
Nova launched in March 2025 and now runs three incentive mechanisms, one for small molecules, one for the search algorithms themselves, and one for nanobodies, with a fourth planned. The team is deliberately global, with partners in China, Canada, and the United States, and a recently announced partnership with OnePot AI, an AI-driven robotic lab, is meant to close the loop between a virtual prediction and a physical result. The first assets in development include a triple reuptake inhibitor aimed at a set of large indications and a nanobody program against an oncology target. There is a real tension underneath all of it: drug timelines and the attention economy of a token do not naturally sit well together, and the hardest ongoing work is contextualizing slow, real scientific progress for a community that would prefer results overnight. Science benefits from openness, replication, and adversarial pressure, and a global, decentralised competition supplies all three more cheaply than a closed lab.