Two Nobel prizes move AI for science into investable territory

AlphaFold's creators and two neural network pioneers won Nobel prizes. Value in AI for science sits with teams that own data and a measurable problem.
Machine learning collected two Nobel prizes in two days. On 8 and 9 October the Royal Swedish Academy of Sciences gave the physics prize to John Hopfield and Geoffrey Hinton for foundational discoveries that enable machine learning with artificial neural networks, and half of the chemistry prize to Demis Hassabis and John Jumper of Google DeepMind for protein structure prediction with AlphaFold2, with the other half going to David Baker for computational protein design.
The chemistry prize is the one capital should study, because it rewards a result, not a method. A model trained on decades of laboratory data solved a problem that had resisted science for 50 years, and its output now feeds daily work in biology. That changes how AI for science is valued. The asset is not the model alone but curated experimental data, as scarce as expert-written training data, joined to a narrow problem with a measurable answer and a path to a product.
Use is already industrial in scale. AlphaFold2, presented in 2020, has predicted the structure of virtually all 200 million proteins researchers have identified, and more than two million people in 190 countries have used it. Money followed the results. In January Isomorphic Labs, the Alphabet drug discovery company that Hassabis also leads, signed partnerships with Eli Lilly and Novartis worth close to $3 billion including milestones, which shows pharmaceutical and materials companies paying for model-driven discovery on industrial terms.
The open window sits where AI meets a scientific bottleneck with a clear test of success: protein and antibody design, enzymes, new materials, chemistry and lab automation. The strongest teams put machine learning researchers next to domain scientists and own a source of proprietary data, often from their own wet labs. Validation cycles run long, so revenue is best built around milestones and royalties with industry partners rather than software subscriptions alone. Recognition at this level also makes it easier to bring senior scientists from universities into startups.
Faster discovery has yet to show that it produces better drugs. That evidence arrives when candidates designed with these models enter clinical trials and reveal whether success rates rise along with speed. Access is the other open question, since how open the leading protein models stay decides how much room independent startups have. The size and terms of new partnerships between large pharmaceutical groups and AI-native biotech companies over the next year will show what industry is willing to pay.
Multi-year validation cycles fit the horizon Fanaura invests over, and the teams it favours here pair strong models with proprietary data and a clear path to a product that an industrial partner will pay for. AI for science has left the research paper and entered the term sheet.
Building an AI-native company? Apply in five minutes. →
More from Fanaura
Building AI for protein design, materials or lab automation with your own data? Send us the project.
Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.
Get the news by email
One short note each weekday. No marketing, and you can unsubscribe at any time.