EXCLUSIVE: Vinod Khosla warns AI will upend jobs and healthcare sooner than expected | GSIF 2026

Okay, we are back. We have Vinod Khosla on the call now. Vinodji, namaste. Great to have you here. Thank you so much for making time. I know you have a flight to catch soon.

It's great to be here, talk to everybody.

Thank you, thank you. I'm going to quickly introduce your moderator, Matthew Soule, who is a partner at Adventure Capital. Venture capital firm in the Silicon Valley. He'll introduce Mr. Vinod Khosla, and then we'll get started with the session. We are running a little behind, so let's get started.

Thank you, Anurag, and nice to meet you all. So my name is Mathieu Soulet. I'm one of the partners at Cathay Innovation. We are a large venture capital firm managing €2.5 billion, and I'm based in San Francisco. As you can guess from my accent, I'm originally French. So it's a pleasure, I would say, today to do the fireside chat with Vinod. So maybe as a quick Set up today, we'll discuss 3 domains with Vinod Khosla. We'll start around AI and its impact on the world. The second domain will be around health, and how health, and also with AI actually, it's changing right now as we are speaking. And the last topic we'll address will be around energy and sustainability, and how we are looking at both existential risk and also like trillion-dollar opportunities. So the idea today would be for the fireside chat to have good ideas for the audience, so on founders, investors, we have a few diplomats in the room, and how we can make it better for the future. So thank you, Vinod, for participating. I guess I will start discussing on AI. Vinod, so on artificial intelligence, you are an investor in OpenAI, Replit, Cognition, and many others. AI is also often framed as a productivity tool, but increasingly we are seeing it as an accelerator in many other areas. AI is also multiplying what we call human minds. I guess you use the terminology with like multiplying human minds much like the steam engine was multiplying human muscle in the past. So I guess the first question I would have for you, Vinod Khosla, it would be today, are we underestimating or overestimating AI right now, and in which domain specifically? What would be the next frontier for you around AI?

Well, we are definitely, I think, as a society, underestimating what AI can do for us. Having said that, and AI is much further along than what is used by most people. So I would say most people today on average using 10% of the capability that is in AI today. So that, that's sort of a simple sort of answer. We are not using it. What's Even more interesting is the rate at which AI is advancing is huge. So my bet, a year from now, we will see AI more different from today than today is from 3 years ago. So we'll see more incremental progress in AI, uh, than we've seen in the last 3 years. So So not only are we underutilizing it, we are underestimating what the increase in performance will be. And everybody would benefit from getting to know it better and stay at the leading edge. Okay.

And as a follow-up question, maybe Vinod Khosla, which industries people think are safe from AI right now that actually could be in deep trouble?

Well, first, the how AI is deployed will depend on AI, but it'll also depend on politics. In the American Medical Association, you mentioned help prevent AI from practicing medicine. They will for sure try. Uh, Kaiser nurses are on strike today because AI does mental health therapy better than most therapists. And there was just a Nature Medicine paper 2 weeks ago that AI therapy sessions were— 75% of AI therapy sessions were rated as if they were done by the top 10% of human therapists. So there will be many forces trying to slow down AI. Many incumbents. Having said that, there's hardly an area where AI couldn't have a large impact. Maybe in housing and construction. That's a tough area. Energy is a complex area, but even there, in the slightly longer term, will— AI will help us design better materials, better fusion, better magnets. Lots of different AI science areas. In fact, I'm very excited about the work in plasma fusion control being done with AI. So almost all— most areas are impacted. Simple services like outsourced services, customer support, business process outsourcing, those things will be gone in 2 or 3 years. Literally gone. There are many other areas. And for this audience, I'd say I suspect over the next 4 years, the most exciting new area will be AI scientists. AI able to do hypothesis, run physical experiments itself, and then analyze the results, and then do a new hypothesis. So iterative science will become much, much better. And faster and leverage the best scientists.

Okay. And maybe as a quick follow-up, if we're looking at air impact on humanity, what will be the different impacts, how it will differ, I would say, from developed world maybe versus emerging markets? Will you have specific examples?

Well, so it's relatively simple technologically. To imagine why every person in the world shouldn't have free primary care. It shouldn't cost more than a couple of dollars per month to have primary care, and that's less than the health budgets of even the poorest governments. So everybody on the planet, 7.5 or 8 billion people, can have free primary care. Every child on the planet can have an AI personal tutor for free. Those are benefits that will be global. Whether you're in the West or the developing world, you're going to see those benefits. But if politically they are allowed to do it. And then there's Many other areas in the developing world where you could have huge help. An AI agronomist available to every small farmer in the less developed countries is a trivially easy thing to imagine. AI legal services for every human being, pretty easy to imagine. And most people never have access to their legal rights. Even though technically they have it. Yeah. So it's almost— or at the level of most services, they will be relatively inexpensive and declining in cost and become broadly accessible. There's other areas, of course. Entertainment will expand dramatically and get much cheaper.. You won't have to rely on Hollywood. Hollywood is mostly trying to bury its head in the sand like an ostrich as the tsunami comes. So I could talk to any area. The area I haven't mentioned, in the next couple of years, robotics will also have its ChatGPT moment. You'll see essentially all human expertise reduced to, I'll call it, a dollar an hour or dollar or two an hour. All human labor through robotics will be a dollar or two an hour, and that'll have the same impact wherever in the world you are.

Thank you. As we are looking at transformative AI companies, and I guess we are seeing sitting right now in the Computer History Museum in Mountain View in Silicon Valley. Do you think the most transformative, transformative AI companies will be built here in Silicon Valley, or maybe somewhere sometimes closer to the unmet needs where actually it could have a bigger impact?

I think there will be companies built everywhere, but Silicon Valley has a culture and it has nothing to do with the place. It is the culture it engenders where people tend to think more ambitiously and hence build more ambitious companies. I like to say most people are limited not by what they can do, but by what they think they can do. And Silicon Valley encourages you to think bigger and bolder and better. And that's very exciting. So a disproportionate number of large companies will be built in Silicon Valley, but they will be exciting companies in every part of the world. Okay.

So thank you, Vinod Khosla, for the AI section. We are moving now to the global healthcare. So global healthcare is strictly is estimated to be like a $10 trillion system globally, but is mostly concentrated in a few countries. Maybe the first question regarding AI and the impact actually on healthcare, do you think actually it will be deflationary potential of a few hundred billion dollars on that? And I think you wrote in 2012 that AI would replace 80% of what doctors do. Where do you think we are standing right now on primary care to mental health? Could you help us to figure out how healthcare is changing right now thanks to AI?

Well, what I wrote in 2012, I think I was too conservative. 80% is too low a number. And in 2016, I wrote a 100-page document called 20% Doctor Included. And that was by 2014. And today I would say 10% doctor included by 2030. So I was wrong on two fronts. So I— and we have just submitted a paper to JAMA, and I won't go into the details, the Journal of the American Medical Association. Suggesting that AI doctors already are superior to human doctors. Let's see if it gets accepted. Most of the audience probably knows the book, The Process of Acceptance. But that, that's where we are today in mental health. As I just mentioned, there was a paper in Nature Medicine about 2 or 3 weeks ago, uh, where we essentially proved that AI doing mental health therapy sessions are superior in almost all respects to human therapists doing mental health therapy. Uh, I would say the following. In this particular paper, AI therapy sessions and human therapy sessions were judged by human AI therapists. And the human AI therapists classified 75% of the mental— AI therapy sessions done by an AI therapist as in, as at the level of quality of the top 10% of human therapists. So everybody can get top 10% quality of therapy in mental health. So that's just the care provision. Today we have oncologists in process, physical therapists in process, AI, you name it, AI primary care in process. And we are seeing concordance between AI therapists and human therapists at 97%. Even the inter-human concordance doesn't get that high. So these are examples of the way therapy should be judged. How is it relative to human performance? That'll be true in most areas. AI will be superior to the median therapist, the median doctor, the median oncologist, the median gastroenterologist. So that is definitely true. AI's ability to do so— healthcare's four major categories is provision of labor services like an oncologist. That's a quarter of the spend in the US, at least a quarter of the spend is drugs. The number of drugs will explode. AI-designed drugs will have lower failure rates, so easier to design, and we'll see many more of them get through FDA trials successfully. So we'll, we'll see all that. We'll see drugs designed for one patient So every cancer we know is different. You could design a drug for one patient with AI. You wouldn't even have to run a trial because you don't only have a population of one. Another quarter of the spend is imaging and testing, and that's going to be impacted by AI. Of course, AI radiologists are much far superior to humans. So let me stop there. I can keep going.

Maybe one or two examples, if you want illustration. Where are the biggest opportunities according to you for underserved population? Do you have one or two examples of underserved population? How does it change?

Well, very much. I was in India recently and I proposed to the Prime Minister that we can develop an AI doctor available 24/7 to every Indian. And the poor Indians, especially in the villages, to see a doctor may have to travel a day or two. If they're not seen the same day, they will be— they'll probably sleep in the corridors in the hospital. Till there's somebody can see that. So there's a real shortage. That's probably the most interesting example in mental health therapy, which I mentioned. We looked at equity and inclusion in AI therapist sessions. It increased dramatically, and the more marginalized the community, the, the better inclusion was. So for transgender communities, inclusion went up by 150% for mental health therapy.

Maybe if we are looking against comparing different markets, does innovation for this market could be built locally, or can it be some, some in some ways exported for one country or one region to another?

It's pretty easy to export, export it from one region to another. The general models being developed are good enough. And in fact, the best models are likely to be either Western models— of course, China has its own situation and they will do well. But give you an example of what we are doing in a country like India. We are doing nonprofit provision of primary care and using the best Western models like GPT-5 and combining them with local Indian models that specialize in local languages, local accents, local voice conversations. There's 13 different scripts in India. We'll be providing primary care services in all regional scripts.

Okay. And last question. I think you were talking about it for India, but if you are today redesigning the healthcare system for a country like India or Vietnam, for example, what would be your priority? You mentioned the AI doctor like 24/7. Will you be— are there like priorities for you?

I think that's the clear first priority. All services are essentially free. And that can be done in a small fraction of even a poor country's current healthcare budget. Whether you're talking about India, Vietnam, or Malawi, it can be done for a fraction of that budget that they spend today on healthcare. And it will reduce serious disease downstream because primary care and disease management and early intervention will cause escalations to go down and reduce total healthcare spending. So that's obviously why.

Okay, maybe moving to energy challenges and climate. You've been investing in the space for a long time. So what are we still getting wrong? I would say climate innovation.

Though AI has had very limited impact on climate, but if AI can help control plasma fusion or the plasma in plasma fusion and we make fusion viable, which I am extremely optimistic about, that in the next 2 years people won't be talking about what is the source of energy on this planet. And AI will help get us there. There's, of course, better magnets to be designed, better control systems. AI can help with all of that. So there's that. Then there's better materials. Maybe one of my favorite projects is AI-based public transit. We are introducing AI-based public transit. What happens in public transit? And you're in the Bay Area, BART, which is the Bay Area Rapid Transit System, has a sales tax increase on the ballot. And if they don't get the increase, if it gets watered down, they will reduce service. A much better system because BART's a very large train, and 90% of the time It is half empty or more. Only during peak travel hours is it full. So the average carbon emissions of even electric systems is actually pretty bad when you compute total passenger miles and grams of emission from the electric grid per passenger mile. We have a system where autonomous vehicles run on these tracks. The carbon emissions are much lower. The service is much better because with small autonomous individual vehicles, you can do it point to point. And it's, of course, much safer and cleaner. So public transit is another area. I could go on and on.

Yeah, I guess, you know, just following on that, like discussing about specific magic bullets in energy, I think you mentioned or used the is the past regarding Chindia. So which emerging technologies, according to you, has the best chance to reaching a Chindia price threshold? And maybe you could define what is Chindia, I guess.

So I defined in 2010 a price I called the Chindia price, the price at which India and China will absolutely adapt these technologies. Because they are cheaper than their fossil alternatives. That's the way to do sustainability today. So even though India is adapting a lot of solar and China is, they're still building more coal plants than solar plants. And so they're getting to the cheapest source of power, not the greenest source of power, because Green is a luxury in most parts of the world, not in the West. But if we can make fusion power and my other favorite for power generation is super hot geothermal, that means geothermal above supercritical steam temperatures. So 400 degrees and above geothermal. As far as I know, there's no commercial projects except what we are attempting. Anywhere in the world. There's some research efforts for supercritical steam, but a supercritical well will get 10 times the power per well that a 200°C well will generate for the same amount of water flow. So we need to get there. That's true in many other things. I mentioned power. But today, I believe we have steel technology or production of steel that is cheaper than a new plant being built with coal-based steel. So I think we beat economic parity with old-style new plant construction, and the green comes for free in steel. Is not electric steel and some of the nonsensical stuff we see. The same is true in cement. We take carbon emissions from existing carbon plants, turn them into more cement. So with no more limestone input into a cement plant, you reduce emissions and the output goes up. So cement in the cost per ton of cement is cheaper than regular cement. So in a number of these critical areas, we've achieved that. I think electric cars have achieved parity, especially with the new price of oil. So jet aviation is the one area where there's no easy solutions and expensive solution. New solutions would require are putting all the old plants out of business. So take trillions of dollars of airplanes and obsolete them. That's not viable. And too often in sustainability, we look at the best sustainability technology. That's the wrong way to look at it. My general view is we should strive for the cheapest technology to also be the most sustainable. And that's achievable if we put our mind to it. And take somewhat higher technical risk like our fusion efforts were. So I do think there's a lot of areas that progress can be made. Yeah.

Maybe one last question on the climate one before the closing remarks. What would be the climate idea that most investors here like will reject right now, but you think could be a trillion-dollar outcome?

I just gave you 4 of them. Yeah, it is public transit, fusion, super hot geothermal, low emission steel, and low emission cement. Those 5 or 6 things are among only a dozen things that matter. Too many people work on the wrong sustainability problems. I would venture to guess 90% of the plans I see in sustainability are irrelevant to sustainability and low emission climate change, even if they are successful. The 10% that matter are only about a dozen technologies that I defined about 3 years ago when I wrote a blog on the 12 critical technologies in climate that matter. And most of the things don't matter.

Okay, so for maybe for last question for you, Vinod Khosla, for people in this room that are thinking, I would say, globally and not just locally, what should they stop doing immediately? What should they start doing tomorrow morning? And at the end, like, what excites you most nowadays?

Well, I would say to people, there isn't a magic bullet you should stop doing tomorrow. But staying at the forefront of AI, and it is a scary technology. AI has the popularity today outside of the Bay Area of same as ice. That's about the popularity. People are afraid of AI and job displacement with AI, and there's two classes of people, those who say don't worry, it won't happen. And, and then those who lie about it and say it'll be slow. My view, everybody should be at the forefront of technology. There's scientists in your audience. I just saw a case of somebody doing doing ME/CFS research. They had graduate students doing analysis for 2 years on data they had generated in the lab. On one weekend, the researcher replicated all of the results— years of graduate student work— with AI, using an AI scientist kind of approach, and then generated some new new insights that graduate students hadn't found from the same data. So stay ahead on the AI front. The second thing I would say is people like to work on projects that are successful or likely to be successful. I like to say not all hard problems are valuable, but most valuable problems are very hard. And people should work on hard problems because they result in large breakthroughs that actually solve real societal problems. I think that the risk of failure avoids us working on the bolder, bigger problems and looking for bolder, higher-risk solutions. Fusion is a good example. When we started Commonwealth Fusion in 2018, Every single person told me, without exception, including the Department of Energy, it was a really bad idea. Now there's a dozen reasonable fusion startups because we set a good role model. And I'm glad there's a dozen, but the chances that they all fail, I think, starts to approach low single-digit percentages. And I don't think 5 years from now we'll be debating what's the source of energy, and we won't even be talking about nuclear or natural gas or coal.

So I guess thank you a lot, Vinod, for your time in the fireside chat tonight. And I would ask the audience to give a round of applause for you. Thank you.

Thank you. Good luck, everybody.