Why Are OpenAI and Anthropic Both Racing Into Biotech?

Have you ever wondered why, almost every week lately, another AI company announces it's getting into biotech?

Nvidia and Eli Lilly are building a lab together in San Francisco; Isomorphic Labs, spun out of DeepMind, is doing drug discovery; Sam Altman has personally invested in the longevity company Retro; and Anthropic is steadily ramping up in life sciences. Alex Colville is close to all of this. He is the co-founder of age1, the world's first venture fund focused on longevity, which he started together with Laura Deming. He did a PhD in genetics at Stanford, studying the biology of aging, and has managed more than $100 million in longevity investments. This article chases just one question: why do AI companies want to do biotech, what can they actually help with, and where will they get stuck? I'll first lay out what he said, then add my own thoughts.

Why AI companies are all crowding into biotech

Alex starts with a number: the two weight-loss drugs semaglutide and tirzepatide together bring in more revenue than OpenAI and Anthropic combined, and they are still growing fast. Many people assume this kind of business will be captured by tech companies, but it hasn't been. His guess is that companies like OpenAI, Anthropic and Google already carry astonishing valuations and need equally astonishing revenue; looking around, they find that a single drug in the world earns more than their entire company, which of course makes it worth a look.

The second reason is the moat. Drugs are patent-protected and the industry is tightly regulated, he says, unlike most tech products, which over time drift toward commoditization and get cheaper and cheaper. He cites an observation Marc Andreessen makes often: prices in regulated industries keep rising, while prices in unregulated industries get pushed down relentlessly. So in his view biotech is highly resistant to AI — it won't be disrupted by Anthropic inside a single year. The reason is simple: there are too many hard problems, and even throwing all the intelligence in the world at them still takes that long.

On Anthropic specifically, he says the company published a market map showing life sciences as one of the areas with the greatest impact. The team has also said publicly that this could be the company's second-largest market direction, at least top five. Not long ago Anthropic released a new model aimed at life sciences and mentioned preclinical programs in areas such as aging and rare disease. His reaction was one sentence: when Anthropic says publicly that this is a priority, that means "it's on."

What I find rare about this section is that it doesn't dress up the AI companies' motives as noble. They come because here is the world's biggest revenue, the deepest moat, and an industry that AI cannot wash away for now. Seeing that clearly actually makes it easier to judge what they'll do next: not go wherever is most romantic, but dig in wherever the business is hardest to replace.

Two things AI has already helped with

So what has AI actually helped with? Alex says he sees two of the most obvious changes.

The first is generating molecules. For certain classes of drugs, designing molecules is now far faster and much cheaper than before, no longer requiring humans and labs to go back and forth at every step — though it is still a long way from "no humans, no experiments." He mentions companies like Nabla Bio that can already do zero-shot generation of antibodies and small molecules, meaning it can produce candidate molecules in one shot without being fed a pile of examples first. But he also says that at this stage, most of what gets done is aimed at known targets (the protein or gene a drug acts on), producing new versions of existing drugs — just very fast. The next step, he thinks, is inevitable: even entirely new targets will get molecules generated in one shot like this.

The second change is more mundane. Everyday work is now 30% to 95% faster, he says — writing regulatory documents, researching competitors, hunting for partnership opportunities, or getting up to speed on a new target. His judgment: a year and a half to two years ago, starting a drug company was a heavy undertaking; now it has been almost "democratized."

My own read is that one of these is about "making things" and the other about "doing things more efficiently," and both happen in the front half of drug development. They make trial and error cheaper and let small teams get off the ground, but they haven't touched the industry's hardest part. What that part is, he goes on to explain.

The costliest, slowest stretch: where AI hasn't helped yet

Some people have imagined a future with millions of "zero-employee" biotech companies: with enough tokens, just like writing software with AI, you could do research and design in one breath. Alex thinks that is still far from reality. His reasoning: the real bottleneck comes after molecule generation. To turn one optimized molecule into an approved drug, you must prove it is safe and effective, and this stretch accounts for a disproportionate share of the cost and time in drug development. Molecule design itself is hard and expensive, but next to that stretch it is small. This has become "the last frontier," he says: pick the thing most likely to succeed, then run the whole process as cheaply as possible. Even if one person designs a molecule with AI, every such company still needs millions of dollars afterward — and this is precisely where AI has the least leverage today.

I think this is the soberest part of the whole conversation. AI's narrative always revolves around "being smarter," but in the end drugs don't compete on smarts, they compete on being proven. Proof needs patients, time, regulators and money, and none of that shrinks automatically because a model gets better. That also explains why AI companies are willing to enter, yet nobody dares say they will make drugs cheaper tomorrow.

In longevity, there's still little money and no product

Back to the field he invests in. Longevity is the area people most readily associate with AI, but its reality is plainer than many imagine. Alex estimates that all companies targeting aging put together are worth only about $15 billion, with roughly one-fiftieth the number of companies and the investment dollars of oncology. Those funding headlines that look big are almost invisible on a bar chart next to AI financings. So his conclusion: every dollar has enormous leverage right now, and we are still in the earliest innings.

More important, the field still has no commercial product proven to work in humans. He expects the first batch to appear within the next decade, aimed at delaying or preventing certain aspects of aging — not some dazzling new technology, but the first proof point. He calls this the longevity field's "ChatGPT moment," except it won't give feedback as instantly as the first time you use a large model. His grandfather was one of the first American soldiers treated with penicillin and, after a bout of pneumonia that nearly killed him, lived into his nineties. He says he has been waiting for longevity's "penicillin moment," and estimates it is ten to fifteen years away. Just as people today take statins at a certain age to prevent cardiovascular disease, he is waiting for a "statin" for something other than the cardiovascular system.

The nearest thing to people is a drug for dogs. The company Loyal is making a once-a-day oral chew aimed at extending dogs' lifespans, and it could become the first life-extension therapy approved by the U.S. Food and Drug Administration (FDA). He says both the efficacy and the safety approvals have been cleared, with only the manufacturing process left, expected to be completed around the end of next year, though uncertainty remains. The human version, he thinks, is still several years off.

Further out is a bolder second generation, and this is where AI will play a big role. One category is replacing old cells and tissues with new ones over the next fifty years; another is epigenetic reprogramming, which doesn't change the cell type but only dials the cell's "age" backward. That is what he himself studied in the first half of his PhD. He considers NewLimit the furthest along in this category, with more than $600 million in cash on hand. They started with the liver, using mRNA technology to make a therapy in aged mice that let the animals withstand alcohol damage the way young mice do, and they will move on to immune cells and kidneys. But he stresses that no single approach will solve everything: the extracellular matrix (the supporting structure outside cells) also accumulates damage, and rewinding the age inside cells alone may not be enough — multiple routes have to move forward together.

I think this section is refreshingly down to earth. Outside there is a lot of noise; inside it is a few companies, a few mice, and one drug for dogs. AI's job here isn't to replace that long validation, but to let more of the possible routes be tried out faster.

His view on AI companies and biosecurity

Biology has something other industries don't: the same set of capabilities can both cure and be used to harm people. Alex says he is very glad that founders of nearly every frontier lab genuinely care about biosecurity, about the threats of pandemics and bioterrorism. He thinks that if you put the typical big-company CEO in those labs' positions, they probably wouldn't care as much. He doesn't think the labs always get it right or do enough — after all, it is hard to be right all the time in that position — but the caring is real, and he is grateful for it.

He is candid about the reputational layer too. These labs know very well that once they are seen as curing cancer, they become hard to restrict or shut down. He admits that hiding behind this halo may be a little unsettling, but he believes that their genuinely caring about this is a huge benefit for humanity. He also says that apart from bioterrorism, AI's uses in biology are barely controversial — something truly good for humanity, unlike other applications where good and bad are hard to draw a clean line between.

A few thoughts of my own

Looking back, the things Alex described are really several layers of the same story. Why AI companies come: the biggest revenue and the deepest moat. What they can help with: molecule design and everyday efficiency. What they can't help with: the costliest, slowest human validation. And on a frontier like longevity, money is still scarce and the first product hasn't even arrived. Put those layers together and a very plain conclusion emerges: AI has made "thinking it up" cheap, but it has not made "proving it" cheap.

My own view is that this means different things for spectators and for builders. Spectators get pulled along by "company X enters biotech" headlines and feel that drugs are about to get cheap and people about to get long-lived. Builders should ask one more question: is this company solving the idea problem, or the proof problem? AI is accelerating the former; the latter still depends on time, patients and money.

Of course, what Alex offers is judgment, not prophecy. Whether the first longevity product will appear within ten years, and whether AI can push down the cost of validation, both remain to be seen. But the ranking he gives is useful: if he could bet on only one category, he would bet on the one that can be proven first — delaying and preventing aging — and only five years further out would replacement and reprogramming have their turn.

Ideas will keep getting cheaper. Proof won't.