DeepSeek disrupted the world’s artificial intelligence (AI) industry—and financial markets—when it rolled out not one but two AI models between December and last month. No less a pioneer tech investor than Marc Andreessen, co-founder and general partner with venture capital firm Andreessen Horowitz (a16z) crystallized the impact of DeepSeek by calling it a “Sputnik moment” akin to the Soviet Union stunning the U.S. by launching the world’s first space satellite in 1957.
Of particular worry to investors is the $5.58 million quoted by the Chinese AI startup as its total cost of launching one of its two AI models— a figure that market watchers cautioned was more low-ball hype than low-low reality—compared with the tens of billions and hundreds of billions of dollars that U.S. AI tech giants have committed toward expanding the technology.
A day after AI’s highest-profile company Nvidia suffered the biggest one-day drop in market capitalization, roughly $600 billion, the founding investor of a16z’s Bio + Health fund Vijay Pande, PhD, asserted that the U.S. remained in front in the global development of AI—at the moment.
“We’re in the lead for now, but could easily lose the AI race, with grave implications for the resulting new world order,” Pande posted on X January 28. To maintain that lead, he observed, the “U.S. Government should be working with innovators on how AI can best propel America forward.”
Earlier last month, Pande’s firm identified AI among areas in which it aims support development among biopharmas and healthcare companies, by partnering with Eli Lilly to launch the up-to-$500 million Biotech Ecosystem Venture Fund, designed to fund a broad range of innovations in human health.
“The Fund will take a long-term view to enable disruptive companies to realize the full potential of new biological science, engineering technologies, and AI capabilities,” a16z stated, adding: “The Fund will focus on advancing the development of new medicines, enabling novel modality platforms, and scaling emerging health technologies.”
Pande, who is also a general partner with a16z, recently discussed how AI’s uptake in healthcare will shape its adoption by biopharmas with GEN Edge. (This interview has been edited for length and clarity)
GEN Edge: In October, you and colleagues made a case for why technologists and not just medical professionals are key to reinventing healthcare. How much is this simply healthcare emulating what biopharma did itself in recent years, in terms of using more technology people as opposed to just relying internally on their own chemists?
Vijay Pande, PhD: I think there’s a couple of things. One is that I think AI and healthcare are having a real moment right now. You see it from the technology side, that AI can actually pass medical exams and do extremely well. And then given that technology, there’s the natural question of, how can that technology be applied to maximally help patients. When you think about the issues in healthcare of cost, quality, and access, AI can address each one of those.
GEN Edge: How can quality and access be achieved through AI?
Pande: There are human doctors, there are AI doctors. A doctor now is actually really valuable. And then obviously, you can then go talk to a human later. So that’s access. But actually, there are also many Americans who live quite far from doctors or have to wait in lines. And then quality is interesting because right now, doctors do very well on the medical exam. You can imagine in time, AI could be better than 80% of doctors and 90% of doctors.
People get excited about the superhuman part, but actually the irony is right now, especially with all the light on prevention, is that we can forget about whether AI is going to be a good brain surgeon or not, and that may or may not come. The prevention part is something where the scale aspect of AI is really intriguing. Not only can it do what it can do, but it can spin up an AI doctor the way you spin up a server. And so that’s where the doctor in your pocket is really interesting. It’s maybe more about making sure you eat the right things and do the right things and have that available all the time continuously.
GEN Edge: So it can lead you to water, in terms of healthier choices of food. But can it make you drink, so to speak?
Pande: That’s a very deep question. I think one of the things that we talk a lot about is, tech is really good at behavior change. You see people on their phones all the time. So, I don’t think we’ve perfected this, because enough behavior has not changed. But I think we see glimpses of where they can be coming.
I was on a podcast with the CEO of Oura [maker of the Oura Ring]. My colleague Daisy Wolf [PhD] said, ‘You know, all people should know, when you get an Oura Ring, and basically, you drastically limit or stop drinking alcohol after you see what it does to your sleep. I was like, ‘Oh, that’s interesting.’ Then I was like, ‘Well, it’s not going to affect me, and I don’t drink that much anyway.’ And then I got the Oura Ring, and drinking goes down because you just can see the effect on things. And it’s funny: It’s one thing to not feel good, but if you see it over time and have the record of it automatically, the weight of all that data weighs upon you.
GEN Edge: How does growing use of AI benefit biopharma as it relies more on technology to carry out drug discovery and development?
Pande: I break down biopharma into three stages; It’s target identification, understanding human biology to find the right targets, and hitting those targets, whether it be small molecules or antibodies or whatever, and then bringing those therapeutics into the clinic. In principle, there’s a fourth bucket, which is deploying them into patients.
For the first bucket, we see companies like [machine learning-enabled drug discovery and development company] insitro and others that are using AI to be able to decipher human biology. We know a ton about mouse biology, and that’s because we can do all these experiments on mice that you can never do ethically on humans. The mouse is not a great predictor of human biology.
AI has its limits and its flaws, but it’s becoming a much more accurate predictor of human biology now than an atom-wall. And given that 80% of drugs fail by the end of clinical trials, and most of them due to efficacy, the real problem is, it’s not that we couldn’t hit the target. It turned out that hitting the target didn’t change the disease. Having this AI model of human disease biology would have a huge impact in those results, so there are companies going after that.
GEN Edge: What examples of such companies can you discuss?
Pande: There are companies like Genesis Therapeutics or BigHat Biosciences or Inceptive that are developing AI approaches to hit targets. And then actually we’re trying to see companies apply AI to clinical trials. One of our portfolio companies, Formation Bio is doing this. There are several others. I’ve even heard pharma is doing this. This may start with things like generating the documentation and using generative AI for that. And I’ve heard people at FDA talk about their preparing for all these AI submissions that are coming. I’ve even heard internally that they realize they may need AI internally to be able to sort of have this arms race with the external AI coming at them, to be able to keep up.
And so that’s one thing. But I think where this gets more exciting is in everything else in clinical trials—patient identification, stratification. And the Holy Grail is, can you predict which drugs will do well in trials? And if you really understand human biology well enough, you should be able to do that, at least better than we do now. And if we can merely have only 70% of the drugs fail instead of 80%, that’s still tens of billions of dollars saved as we make progress to hopefully 0% fail as a goal.
Then finally, if you have this model of human biology, you can do precision medicine on the other side. You can now ask, ‘Well, for me as a patient, how will I respond?’ And the AI model of that would be very natural. That fourth bucket is, I think, right now an aspiration, but you can see how the same technology could help all the way through.
GEN Edge: How far along are pharma companies into that?
Pande: I don’t know for sure since I’m not on the inside, but I’m on the AI internal, there’s some pharmas that have developed their own AI divisions. Others are partnering. And I think this is a classic question for any new technology: Are the incumbents going to develop it internally or will they come from startups? If you look at the history of biopharma, M&A is typically how a lot of technology gets acquired. So, I wouldn’t be surprised if that happens in this case as well.
GEN Edge: We haven’t seen a lot of M&A in the AI space other than Recursion and Exscientia last year. Is it an issue of maturity of the technology before we see M&A?
Pande: I think the main issue also is that there hasn’t been a ton of M&A more broadly. But some of it probably is maturity too.
GEN Edge: You mentioned the FDA. To what degree are regulators encouraging AI? Or to what degree is it hindering AI extension?
Pande: I haven’t seen anything hindering, because in the end you have to go through trials anyway. When you talk to people at the FDA, they have patients’ best interests in mind as well. And I think that should be our North star: What is best for patients?
GEN Edge: Back in August, you asked a good question on the Raising Health podcast: Where is generative AI going? A lot of the excitement these days was about generative AI, where you don’t just understand a latent space, but generate something from it.
Pande: Generative AI can take a bunch of different forms. Generative AI for words is something that we see with ChatGPT all the time, and I think in clinical trial documentation that’s a no brainer. Generative AI for pictures is super cool; I make cat pictures for my daughters and so on. Making pictures is not that useful on the biology side, but I think generative AI for combinations of molecules is intriguing. And I think the paradigm is that typical machine learning is that you do screening whether it be virtual screening instead of screening real drugs.
The thing about virtual screening is, you make your libraries and you screen into a classifier. That can be cumbersome. The general approach is basically, ‘Find me the right drug. Find me the right example without having to do tons of screening,’ And that’s intriguing.
And I think there are companies that are definitely taking that approach. Where it might be most intriguing in my mind, is for something where there’s a relatively constrained problem and I would say chemical space is really vast, but antibody space is a little less vast in that you know it’s more constrained in terms of the protein backbone structure that has to be there. Certain parts have to be conserved and so. You can imagine generative AI for antibodies could be one of the first things that gets really interesting.
GEN Edge: Would that explain why you said the difference between the antibody space and why some companies are pursuing more chemistry focused AI and saying, ‘Gee, we’ll find many more interactions that way’?
Pande: Yeah, I think some of it comes from where their technology comes from for better or worse, also, the Inflation Reduction Act also is often on early startup people’s minds for what to do, and where that’s going to go. And who knows? I think with the new administration, there might be changes there too.
GEN Edge: You mentioned the new Trump administration. How disruptive to biopharma will it be? And what sort of changes should the industry or others embrace as a result?
Pande: We don’t know for sure, but my read is that there is, especially knowing people that are going into the administration, I think there is a desire to really help advocate for innovation and let innovation address a lot of these issues that I think are often much better handled by the private sector anyway. I view the role of government is to create a market with rules that are fair, but then let the market take it for there.
GEN Edge: Any real outstanding examples of companies applying AI other than companies in which you have invested and on whose boards you sit?
Pande: I have the luxury of investing in many things, and I picked the ones that I think are my favorites.
Let me think [pauses]. Recursion comes up a lot. And I think Recursion is similar to insitro in mindset of trying to decipher biology. Aviv Regev [PhD, head, executive vice president, Genentech Research and Early Development] at Genentech also, I think has a similar mindset and they’re all taking fairly different approaches in some ways. But I think they’re spiritual cousins. Tons of AI for small molecules and antibodies. Generate Bio[medicines] is also doing exciting things. Companies applying AI on the clinical trial side, I think that’s still early.
GEN Edge: Companies have talked about speeding up their pinpointing of patients faster, among processes they wish to get done faster. How much has the trial process itself been rethought via AI?
Pande: I think that may come in pharma first since they’re so good at running trials, but I think there’s a huge opportunity, and to some degree, Formation Bio is becoming this, to have a CRO [contract research organization] that’s a more modern CRO—not a cost-plus CRO, but a CRO that wants to innovate trials.
GEN Edge: How early in development are they potentially? Are companies grasping where AI can take them?
Pande: I think we’re seeing AI in all stages. You’ve probably heard me use this analogy before, but I think it’s still a good one: You look at how the Internet gradually changed everything. It didn’t change overnight. It changed over 20 years, and some people will have to break what everyone thinks are the rules.
People once felt very strongly that there was no way that dog food would be bought on the Internet because you couldn’t ship it for less than what it cost, especially a big 60-pound bag. But as shipping became more prevalent, basically the marginal cost of one more bag is not that much. And so, economics works out.
I think when AI shifts from being rare to being more omnipresent, there’ll be analogous economic shifts that will be there. But it’s still very early.
For that reason, I think AI in care delivery is maybe where they’ll be the first visible wins and it’s visible on a couple different fronts. Being visible, because as a patient you’ll see it, and this may be as you’re talking to an AI nurse, and this is an AI nurse that sounds like a great nurse. Or AI doing subclinical primary care. You’ll see this as a patient. I’ll see it in the P&L [profit and loss statements] of these companies as an investor, because the growth of these things has been really enormous. And then, hopefully, other companies or the country itself will see it in terms of decreased cost.
GEN Edge: Also speaking of numbers, you also raised the question about clinical trials being so expensive and 80% failing, and if we could raise the success rate from 20 to 30%. How differently would companies have to approach AI to deliver that increase from 20 to 30%?
Pande: I think as the AI companies that are using AI for target identification mature, they’re going to use those technologies for trials. And I think we’ll see that in probably the next five years.
The big reason why healthcare care delivery is more appealing is that we can see immediately the patient get better. Drug design, unfortunately, takes a while with the time of trials. So, once we see it in drug design, it’s going to be probably a lagging indicator.
