Many tech leaders try to justify the AI bubble based on a possible cure for a disease or on enabling nuclear fusion (or rogue autonomous agents hacking websites and creating civilizations).

For instance, Dario Amodei, the CEO of Anthropic recently said:

I think it will actually be possible to cure most human disease in 5-10 years, as crazy as it may sound to ordinary people and frankly to biologists as well (I used to be one!).[He also claimed this week that] today we announced the Claude-led discovery of a molecular machine that we suspect could represent a new gene editing mechanism [which was quickly disputed by scientists. Nobel Prize Winner Demis Hassabis said last year]: “we can cure all disease with the help of Al… maybe within the next decade or so, I don’t see why not.

These optimistic proclamations about AI have accelerated since the development of AlphaFold in 2018, an AI system that predicts a protein’s 3D structure accurately from its amino acid sequence. AlphaFold 3 subsequently moved beyond single protein structures to model DNA, RNA, chemical modifications, and how they bind together. Combined with our knowledge of DNA, curing diseases should be easy, or so the tech bros say.

The reality is that:

As of August 2026, not a single AI-discovered drug holds full FDA approval. Projections for the first such approval remain pinned to 2026 or 2027, and those dates have been quietly sliding for three consecutive years.

One reason is that the traditional 12-to-15-year drug development cycle involves mostly testing. The only way that AI will reduce that cycle is if drugs found with AI are able to pass tests faster and with a higher frequency of positive outcomes than traditional human discovery. There is no evidence that this will occur.

A second reason is that analyzing DNA data, thus using AlphaFold, is just a small part of curing diseases. So even if we know everyone’s DNA sequence, which we don’t, we won’t be able to cure most and certainly not all diseases because “an estimated 80% of chronic diseases and premature deaths are not driven by our genes, but by the way we live: What we eat, how much we move, how we sleep, how we manage stress and stay socially connected. It’s true AI can accelerate drug discovery and personalize nudges for our daily behaviors, but it can’t ‘cure all disease’ while ignoring human nature, free will, and how we live.”

Third, understanding the human body and diseases, and identifying a target for a drug, is more important than designing a drug. Some scientists break drug research into three separate problems: 1) Disease-to-mechanism; 2) Mechanism-to-drug: and 3) Drug-to-patient (delivery). AI has mostly been used for the second problem, but the first problem is the hardest part of drug research, and AI is having little impact there.

AI didn’t invent hype

Former President Richard Nixon declared a war on cancer in 1971 with hope of finding a cure within five years. Following the first sequencing of human DNA in 2000, then-director of Human Genome Project predicted “in another 20, 25 years we should be able to prevent or cure most cases of cancer, of diabetes, of heart disease, of multiple sclerosis, of asthma.” Those 25 years have passed, but those diseases persist.

Furthermore, “Cancer is not one disease. It is more than 200 distinct diseases, all with different causes, biology and mechanisms,” says a doctor at Harvard. “Scientists who study cancer rarely talk about ‘curing cancer’ because it’s nothing more than an empty slogan.”

 Many doctors concur. “Diabetes, Alzheimer’s, heart disease—we don’t have cures for any of those common diseases, To think that in the next five or 10 years new drugs are going to change everything, there’s no precedent for that. It’s unrealistic, and it sets the expectations for AI too high. It’s hype” says cardiologist Dr. Eric Topol.  And Eric Topol is no Luddite. If you follow him on x.com you will see that he is one of the most optimistic doctors about new cures, regularly posting data from recent medical articles.

Another scientist says:

Hundreds of years into modern medicine, our understanding of most human disease, and much of healthy physiology, is best captured by the parable of the blind men and the elephant; in this case, a really huge elephant.

So developing cures for diseases goes far beyond using AI to analyze DNA, DNA, RNA, chemical modifications, and how they bind together.

Other scientific fields are likely similar. A recent report from Google, Google DeepMind and the MIT FutureTech surveyed 637 scientists combined with analyses of 15 million Gemini conversations and an inventory of more than 2,600 specialized models. It found that 44% of scientists said, “over the past two years, their main research bottleneck had shifted downstream, toward later stages such as physical experimentation and data collection. 41% said their backlog of untested hypotheses had grown.” Outside of math and software, testing, particularly when combined with years of incremental design changes, is a major bottleneck in the development of new technologies.

Furthermore, even pre-testing activities have not benefitted from AI as much as many think. “89% of those who save time using AI spend more than a 10th of the time that they’re saving actually checking AI outputs,” says the co-lead author of the study. “46% spend more than a quarter.” While “a mathematical proof can be checked against formal rules,” “a predicted protein function has to be built in a lab and tested in a living system.”

New technologies start with the low-hanging fruit

A much simpler way to argue that AI won’t soon cure a disease or enable nuclear fusion is that new technologies always begin with the low-hanging fruit and curing diseases or enabling nuclear fusion aren’t  the low hanging fruit for generative AI. The first group of low-hanging fruit includes AI-assisted coding, video ads for YouTube, and AI-assistant chatbots. As those become solved, new low-hanging fruit will emerge.

 So one way to understand how long it will take for AI to cure a disease is to look at the low-hanging fruit, the rate at which its picked, and the rate at which revenues for these and other applications are emerging.  How fast are these applications succeeding? That is hard to say but the lack of profits for the AI labs, the emerging cash flow problems for hyperscalers, and the slow growth in revenues for both types of companies suggests that current AI applications are not growing at a speed that will enable diseases to be cured in the near future.

In the meantime, a bit more honesty about the AI bubble would probably increase the speed of AI adoption.