Google shuts down AlphaFold, undermining its own AI-cures-cancer pitch

Two weeks ago, Google quietly shut down AlphaFold, the AI protein-structure prediction tool developed by DeepMind and led by Demis Hassabis. A week later, Google removed Hassabis from his role directing DeepMind itself, reassigning him to what the article describes as a 'face-saving promotion' to a powerless position. The Financial Times later reported the real motivation: senior Google executives had grown frustrated with what they saw as Hassabis's lesser focus on the commercial demands of the company's AI business. The breaking point was Hassabis's decision to make AlphaFold's code and database freely available to the scientific community. That choice, while advancing research, meant the project generated little commercial return despite the scale of its investment.

AlphaFold's significance cannot be overstated. The tool was designed to solve a long-standing problem in protein science: predicting the three-dimensional structure of proteins from their amino acid sequences. That structure matters because misfolded proteins directly cause certain diseases and are believed to play roles in Alzheimer's, Parkinson's, and certain cancers. For decades, scientists struggled to solve this prediction problem reliably. The first version of AlphaFold won a prominent protein-folding prediction competition in 2018; its successor, AlphaFold 2, dominated the same competition two years later, helping Hassabis become one of three laureates for the 2024 Nobel Prize in Chemistry. The final iteration, AlphaFold 3, could predict more complicated structures and interactions involving proteins, DNA, RNA, and other molecules.

Yet despite AlphaFold's achievements, it had significant limitations. While it was a major technical breakthrough for the specific tasks it was designed to perform, it could not generalize to unrelated or even closely related fields of science. The improvements in successive versions came not from the model improving itself but from human developers imbuing it with new capabilities through more sophisticated architectures and expanded training data. AlphaFold did not fully 'solve' protein folding; there remain many proteins whose structures it cannot reliably predict, and even for those it predicts accurately, it cannot explain why or how they fold the way they do. In other words, AlphaFold was narrowly brilliant but commercially difficult to scale.

This became the fatal issue. Alphabet is spending $200 billion in 2026 alone on new AI infrastructure. Had Hassabis agreed to keep AlphaFold proprietary, Google would have enjoyed a monopoly on its use and a clearer return on its investment. But he didn't, and that sealed the fates of both the project and his leadership role. The article notes that Google is simply following Corporate America's playbook: Myriad Genetics infamously patented human genes to retain market monopoly on genetic testing until the Supreme Court shut the practice down in 2013. Cancer drugs remain proprietary, and pharmaceutical companies have used that fact to set prices at what the article calls 'extortionate levels.' Against that backdrop, Google's decision is unsurprising—but it exposes a contradiction at the heart of Big Tech's public messaging. Executives including Alphabet President Ruth Porat and DeepMind co-founder Demis Hassabis have repeatedly promised that AI will help cure cancer and solve humanity's greatest challenges. Yet the moment an AI project actually pointed toward that goal, Google killed it because it could not be monetized. The article concludes that Big Tech does not, in fact, care about curing cancer unless it boosts profits, and the industry's grand promises about AI's humanitarian benefits should be remembered as rhetoric rather than commitment.