What this covers
Vinod Khosla, the entrepreneur and investor behind Khosla Ventures, sits down to discuss his long-standing conviction that AI will reshape the global economy in ways most people underestimate. Vinod Khosla argues that within five years AI will handle 80% of 80% of economically valuable work, setting the stage for an era of abundance by 2040 where work becomes optional rather than necessary for survival. The interview ranges across his track record of high-conviction bets—from backing Juniper Networks against universal expert consensus on TCP/IP, to his early 2018 investment in OpenAI when the field dismissed it as a science project—and examines the mindset and principles that have guided those calls.
The conversation extends into concrete application areas: how first-principles thinking drives innovation in energy (geothermal, zero-emission cement and steel), medicine (moving from symptom detection to disease prevention a decade before symptoms appear), transportation (on-demand micro-vehicles fitting bicycle-lane widths), and content production (feature films for $100,000 instead of $100 million). Throughout, Vinod Khosla returns to a central claim: large innovation almost never originates from incumbents because experts extrapolate the past while entrepreneurs invent from first principles. The discussion also covers his approach to venture capital—prioritizing the consequences of success over probability of failure, treating founder partnerships as coaching relationships rather than board control, and favoring rapid strategic evolution over domain expertise. He addresses the darker possibilities—the potential for AI-driven disruption to exceed society's ability to adapt—alongside geopolitical dimensions, arguing that the West must advance AI faster than authoritarian competitors to avoid embedding foreign political philosophy through AI-provided services.
Khosla argues that AI and robotics will drive an era of radical abundance by 2040-2050 where work becomes optional, requiring society to restructure around impact-driven innovation rather than expert consensus, with success depending on whether Western democracies or authoritarian regimes dominate the technology.
- AI will automate 80% of economically valuable jobs within 5 years, making continued employment a choice rather than necessity
- Innovation historically comes from founder-led startups thinking from first principles, not from domain experts or large incumbents
- The geopolitical race to deploy beneficial AI (medicine, tutoring, entertainment) as soft power may be more consequential than AI safety risks
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
In 1996 Khosla backed Juniper to bet on TCP/IP for the public internet against universal expert consensus (telcos and Cisco's CTO insisted on ATM and refused TCP/IP above 155 Mbps), because the data showed TCP/IP usage climbing exponentially near the knee of the curve — the bet returned $7 billion on a $3 million investment (~2,500x).
“I talked to the CTO of Cisco who said they have no plans to ever do TCPIP above 155 megabits... But I looked at the data. TCPIP usage had been climbing and I said this will scale and every expert is wrong in the field.”
The OpenAI investment was a conviction bet of more than twice Khosla's largest initial venture investment in 40 years, made in 2018 (closing early 2019) when consensus held AI was 'a science project' and OpenAI wasn't even using transformer models — driven by the belief that the right team plus resources would figure it out and that it was 'too important not to try.'
“it was a very large investment more than twice the largest invest initial investment I'd made in four years of venture capital”
Khosla's conviction in AI's transformative impact dates back well before the transformer breakthroughs — a 2000 NYT interview said AI would redefine what it means to be human, 2012 TechCrunch blogs ('Do We Need Doctors', 'Do We Need Teachers'), and a 2016 NBER talk arguing the top 20 US occupations could be displaced — all based on tracking the rate of progress on human-performance benchmarks and the influx of top talent into AI.
“what I looked at was the best talent was going into AI... I looked at the rate of progress and the rate of talent influx. Those two things convinced me the moment for AI was soon.”
Large innovation almost never comes from incumbents in the field — across retailing (Amazon vs Walmart), media (Netflix/YouTube/Facebook vs NBC/CBS), transport (Uber, Airbnb), space (SpaceX vs Boeing/Lockheed), biotech (Genentech), and COVID vaccines (Moderna/BioNTech vs big pharma) — because experts extrapolate the past while founder-driven companies have permission to try crazy things.
“I can't think of very many large examples that where large innovation came from somebody who was large or in the business.”
In 1996, Khosla invested in Juniper Networks based on data showing TCP/IP usage was growing exponentially, despite every telco CTO saying they would never use TCP/IP (they were committed to ATM instead), demonstrating that experts extrapolate the past while entrepreneurs invent the future.
“in 1996 I invested uh I started a company uh called Juniper. We worked on a business plan in my office for six months. with the founder Praep Synindu. Why? Because the world was saying the internet would be a protocol called ATM asynchronous transfer mode. So every major company had given up on TCPIP for the public internet and were only going to use ATM because that Telos were specifying ATM. I talked to the senior management of almost every telco and every single one said to me they wouldn't use TCPIP. I talked to the CTO of Cisco who said they have no plans to ever do TCPIP above 155 megabits which is my home service to it's far below my home service for a public router. But I looked at the data. TCPIP usage had been climbing and I said this will scale and every expert is wrong in the field.”
A single location in Oregon for super-hot geothermal has the potential to produce 7 gigawatts of power at cost-competitive rates with natural gas, demonstrating the scalable potential of the technology.
“our location in Oregon, one location produced 7 gawatts of power. And I hope this year I can prove we can do all seven gigahertz competitive with natural gas.”
Khosla's public-transit startup designed self-driving two-to-four-person vehicles that fit in a bicycle-lane width yet provide ~10x the passenger throughput of light rail for the same street width, are hailed on-demand like Uber, are always cheaper and faster than today's public transit, and can be deployed via prefab raised infrastructure rather than years-long tunnel digs.
“we designed a public transit system of self-driving vehicles in bicycle lane width... Has 10 times the capacity of a light rail system or or a car system... it's on demand because the cars just you hail them like an Uber.”
Super-hot geothermal at 450°C (versus the typical US 200-250°C) yields 6-10x the power from the same well due to a multiplier effect — Carnot efficiency rises and the steam's heat content rises — making it cheaper than natural gas; the only barrier is that conventional drill bits collapse at that temperature, so solving deep-hot drilling unlocks it.
“If you increase the temperature to 450°, you get 10 times 6 to 10 times the power... carno efficiency goes up but the heat content also goes up... now it's cheaper than natural gas if you can get to 450°.”
Calling oneself 'founder friendly' is a real disservice to founders — like telling a five-year-old you'll always say yes — whereas the right approach is teaching founders to make decisions and lengthening the leash over time, giving them the final say while always challenging and debating them.
“Founder friendly is a real disservice to founders. It's like if you said to your kids... I will always say yes... we taught them how to measure themselves... we taught them control not gave them yes no permission.”
Even the strongest founders benefit from a debating partner who challenges them but does not make decisions — like a pro athlete with a coach — which is why Khosla refuses to sit on boards or take governance votes, instead doing regular dinners/quarterly sessions to challenge founders while always granting them the final vote.
“Great founders do well whether you like it or not... they can do even better if they have a debating partner... like even a pro sports player has a coach... you want people to challenge you and debate you but not make decisions.”
First-principles thinking, rapid learning, and hubris are the essential characteristics of great entrepreneurs — and the most important question when evaluating a startup is how much they've changed their plan over the last three months, signaling rapid evolution rather than deep domain expertise.
“I seldom look for deep expertise in their area. I do look for people who are thinking from first principles and learning rapidly... the question that's most important to me is how much have they changed their plan over the last 3 months.”
Founders overweight the dilution math (keeping 90% vs 88% of the company) and underweight strategic path choices, when in fact the strategic path can create a 500% swing in the value of their stake — so the real opportunity for value maximization is in building the company well, not in optimizing investment terms.
“The founders forget they're keeping 90%... Whether it's they give up 10% or 12%... Trivial compared to the 500% swing you can get by taking one path or a different path.”
Over the next five years, every professional will gain AI 'interns' — e.g. each physician getting five fresh-MD-equivalent assistants — increasing productivity; but these interns will eventually surpass the professionals' expertise, at which point the change becomes catastrophic and impossible to roll back.
“every MD, a practicing physician will get five fresh MDs as if they are fresh out of Stanford medical school that'll work for them... But that then turns into something much more catastrophic over time... at some point these interns grow up they will have way more expertise than the physician and then it's hard to roll this clock back.”
Most jobs on the planet are not really jobs but 'servitude' — repetitive survival labor like picking lettuce for 40 years or mounting tires on an assembly line — and freeing people from this servitude to reclaim their time would be a great thing.
“most of the jobs in this country or on this planet are not really jobs. There's what I call servitude. If you're a farm worker working for 8 hours a day picking letters and you do that for 40 years... those aren't jobs. They are servitude because you have to survive.”
Slowing down Western AI research is a mistake because authoritarian competitors won't slow down; the West must have the best AI to avoid being subject to Xi or Putin, since whoever provides the world's free doctors, tutors, and entertainment embeds their political philosophy globally.
“this is what the people who call to slow down AI research miss. Putin isn't slowing down... she and China definitely can do this.”
AI-powered medical diagnosis outperforms human physicians in a multi-center study of complex diagnostic cases: AI alone achieved 88% accuracy vs physicians at 73%; when AI and physicians worked together, accuracy dropped to 76% (worse than AI alone), suggesting that human oversight can degrade AI performance in domains where AI has superior information processing.
“I'll give you today's data of a multi-enter study at Stanford. They took the best academic institutions and judged complex AI diagnosis physicians only and these are academic learning centers with the best physicians. 73% accuracy. So 27% of the patients were getting the wrong diagnosis. That's like a lot. That's like one in Yeah, that's a lot. It's one in four. Think about it. And this complex means it was a serious thing. It wasn't just the flu. AI alone 88%. Accurate. Accurate. But then they gave the AI to the doctors. The doctors improved from 73% to 76%. The AI got degraded from 88 to 76%. That's funny. And this is a serious multic-enter study.”
Symptom-based medicine will disappear: instead of detecting disease from symptoms, we will detect disease 10 years before symptoms manifest based on what's happening in the body, and providing far more up-front care reduces downstream disease burden and expensive interventions like heart surgery and ER visits.
“I said symptom based medicine will disappear you don't detect disease from symptoms you detect disease based on what's going on in the body”
The capitated/Medicare Advantage segment of healthcare will be transformed by AI far more effectively than fee-for-service, because for a hospital system reducing cost means reducing revenue, whereas in a capitated system providers benefit from AI-driven cost reduction since they get paid a fixed amount per patient.
“we have a small part of our health care system... where payments are capitated... where you get paid a fixed amount to take care of a patient... That world will be so much more effective because they won't care about reducing costs which AI can do... Reducing cost for a hospital system is reducing their revenue.”
Capitalism exists mostly by permission of democracy; if democracy sustains, society will adjust to AI disruption more easily, whereas China holds the advantage of using coercive 'Tiananmen Square techniques' to enforce whatever it wants.
“we live in a capitalist system but people forget capitalism is mostly by permission of democracy... China has the advantage of using what I call tanaman square techniques uh to enforce whatever they want.”
Khosla sees a clear path to zero emissions by 2050 with most enabling technologies proven in the early 2030s — and crucially these solutions can be made cheaper than incumbents (e.g. capturing cement-plant CO2 into carbonates yields more cement per ton of limestone at lower cost; low/zero-emission steel at similar cost), so climate is solvable without anyone having to care about climate.
“I see our path clear to getting to zero emissions by 2050. Uh I think most of the technologies to that will be proven in the early 2030s.”
Medicine breaks roughly into quarters — medical expertise/physicians (~25%, going toward zero cost), pharmaceuticals (~25%, transformed by AI drug design where the key gain is raising the <5% probability of FDA approval from a lead candidate), diagnostics/imaging (~25%, where the cost is the technician not the machine, addressable by 'self-driving' MRI/ultrasound), and in-hospital care (~25%).
“A quarter of that spend is medical expertise... a quarter is pharmaceuticals... diagnostics imaging... is a quarter of the spend... a quarter is inhosp care.”
Almost certainly within the next 5 years, any economically valuable job humans can do, AI will be able to do 80% of it, with a few exceptions like heart surgery or brain surgery — meaning 80% of 80% of all jobs can be done by AI, which will make the world a very different place.
“within the next 5 years any economically valuable job humans can do AI will be able to do 80% of it with a few exceptions like heart surgery or brain surgery but by and large 80% of 80% of all jobs can be done by an AI”
The next 5 years (to 2030) will look like productivity improvements — GDP going up and accelerating — but by 2040 and beyond we enter an era of abundance so large it's hard to imagine, where the need to work goes away and people work because they want to, not to pay their mortgage.
“the next 5 years to 2030 will look like productivity improvements... and then if I look to the 15 years hence 2040 and beyond. We're going to be in a era of abundance that's so large it's very hard for people to imagine. The simplest way to say it is the need to work will go away.”
Movies and content will eventually be produced for $100,000 instead of $100 million using AI, and anyone who signed restrictive deals like the Screen Actors Guild agreement will go out of business because they locked themselves into an order-of-magnitude less effective content-generation paradigm.
“Anybody who signed up for the Screen Actors Guild thing will go out out of business, but that change will be slower outside that system. People will be producing movies for $100,000, not $100 million.”
Polite feedback to entrepreneurs is a disservice ('I prefer brutal honesty to hypocritical politeness in vain'); the only things that genuinely help a founder are those that change or challenge their thinking, and great founders love this while insecure ones avoid it.
“I prefer brutal honesty to hypocritical politeness in vain... most VCs give polite feedback to an entrepreneur, they're doing a disservice... the only things that make you feel better are things that change your thinking or challenge your thinking.”
Most people reduce risk to increase the probability of success; Khosla does the opposite — he starts with wanting high consequences of success and doesn't care about the probability of failure, because incumbents who 'can't afford to fail' can't take large risks.
“most people reduce risk to increase the probability of success I do the pos opposite start with I want high consequences of success I don't care about the probability of failure”
Among existential risks to humanity, sentient/unaligned AI killing everyone is less concerning than authoritarian regimes (China under President Xi, Russia under Putin) using AI for geopolitical dominance, which is why Western AI development should prioritize having 'the best AI' to avoid subjugation.
“Do I worry about sentient AI killing everybody on the planet? Yes. But more than I worry about President Chi doing nasty things? No.”
Coca-Cola successfully produced commercials using only AI last Christmas, most people did not know they were AI-generated, and the commercials performed well, demonstrating that AI content generation is commercially viable and breaking psychological barriers to adoption.
“Coca-Cola got a lot of flack last summer, last Christmas, because they produced some commercials with only with AI and people were nitpicking analyses and stuff. Most people didn't know it was done by an AI uh and they did great and I'm glad they did it. Broke the ice.”
The COVID-19 vaccine came from startups (Moderna and BioNTech), not from established pharmaceutical companies, despite Big Pharma being 'well equipped' to mobilize. This is further evidence that startups disrupt incumbents even in critical areas.
“Look, even the COVID vaccine came from two startups, right? Moderna and the biiointech. You'd think the pharma was well equipped to mobilize, not a chance.”
A multi-center Stanford study on complex diagnoses found physicians alone were 73% accurate, AI alone was 88% accurate, but when AI was given to doctors the doctors only improved to 76% while the AI degraded from 88% to 76% — illustrating that human involvement can drag down AI performance.
“physicians only and these are academic learning centers with the best physicians. 73% accuracy... AI alone 88%... But then they gave the AI to the doctors. The doctors improved from 73% to 76%. The AI got degraded from 88 to 76%.”
By around 2040 there will be a billion bipedal robots doing more total work than human labor today — robots that are smarter, never sleep, never complain, and aren't constrained by narrow specializations (the same robot that cleans your house could be your doctor).
“If you have a billion bipeedal robots by 20 in the 2040 time frame... Almost certainly they're doing more work than human beings do labor today... They'll be smarter and they won't sleep. They'll never complain”
Everyone on the planet will be able to get a free AI primary-care/multi-specialty physician for less than a dollar a month, but the hard problem is regulatory — the AMA won't let the AI write a prescription, so human physicians must approve the diagnosis and prescription.
“everybody on the planet for less than a dollar a month will get a free primary care free physician... the hard problem is the American medical association won't let the AI write a prescription... we have human physicians who approve the diagnosis”
Sentient AI killing humanity is a real worry, but it should be assessed within the full basket of risks humanity faces (asteroids, far worse pandemics that could kill 700 million rather than COVID's ~7 million), and it is a smaller risk than authoritarian leaders like Xi or Putin gaining AI dominance.
“Do I worry about sentient AI killing everybody on the planet? Yes. But more than I worry about President Chi doing nasty things? No.”
Khosla's public transit startup has won every single bid it's participated in so far, yet was not even invited to bid by most transit agencies, demonstrating the innovation gap between incumbent-invited solutions and first-principles startups.
“we have the most coolest self-driving for MRI machines...Turns out every single project we have bid so far, we bid we bid a lot. Four have been decided. We won every single one for a public transit system from a startup and the startup wasn't invited to bid on any of these projects because nobody knew about us. Yeah. Nobody invites a startup. San Jose invited 32 biders to build us transit system from the airport to the Google campus and then to the Apple campus in San Jose. Mhm. We bid over the transom because most of these bids are open and we want it outright.”
Khosla's cement company captures CO2 from production and converts it to carbonates, yielding more cement output from the same limestone input at lower per-ton cost, demonstrating that environmental and economic optimization can be aligned.
“We're producing cement today where we not only don't say like environmentalists do, let's go shut down uh cement plants every time. There's one in Certino that the environmentalists have been trying to shut down. My approach is very simple. They are emitting carbon dioxide. Let's capture the carbon dioxide and turn it into carbonates. We get more cement out of the same amount of limestone mind at a lower cost per ton.”
By detecting diseases based on biomarker changes rather than symptoms, medicine can shift from symptom-based diagnosis to preventive detection 10 years before symptom onset, dramatically reducing expensive acute interventions like emergency surgery.
“I've argued a piece I did called 20% doctor included in 2016 I said symptom based medicine will disappear you don't detect disease from symptoms you detect disease based on what's going on in the body So uh that's medicine. Yeah, I'm pretty excited about it...we are well on our way to discovering I'm 10 years before the symptoms manifest”
Being 'founder-friendly' in venture capital is actually a disservice to founders, analogous to telling a 5-year-old 'do whatever you want'; true venture assistance means teaching founders accountability and discipline, expanding their capability over time like lengthening a child's leash.
“Founder friendly is a real disservice to founders. It's like if you said to your kids, if you told your 5-year-old, 'Do whatever you want. Do whatever you want.' I will always say yes. Yeah. That's going to pay the price...And Yeah. Right. And in literally um at age 12 I thought of they can make 10% of the important decisions. By age 18 they have to make 90. And my job is to teach them how to make the leash longer.”
Most jobs are actually servitude, not real jobs—work that people do only for survival like subsistence farming or assembly line work—and AI's elimination of these survival-based jobs is actually wonderful because it gives those people their lives back.
“most of the jobs in this country or on this planet are not really jobs. There's what I call servitude. If you're a farm worker working for 8 hours a day picking letters and you do that for 40 years or you're assembly line worker at GM mounting a tire on a car for eight hours a day for 40 years, Yeah. Uh those aren't jobs. They are servitude because you have to survive. And what a great thing if we could give those people their lives back and the time those hours of their lives back.”
Within the next 5-10 years biological intervention will approach computer programming in precision — designing molecules and even drugs for one patient at a time targeting a single genetic anomaly — and single-shot genetic treatments for diseases like sickle cell will arrive within ~10 years.
“in the next 5 years or 10 years biological intervention will approach computer programming in being able to precisely program things design molecules... drugs for one patient at a time.”
When we invested in Square in 2010, Jack Dorsey had just been fired from Twitter and had four failed startups on his Wikipedia, so traditional wisdom would say don't back him.
“when we invested in Square in 2010, the following was the situation. He had just been fired from Twitter... he had like four star failed startups... traditional wisdom would say don't back him.”
Robotics will have its 'ChatGPT moment' in the next two to three years — a robot that learns and adapts rather than being programmed for tasks — and almost everyone in the 2030s will have a humanoid robot at home, likely starting narrow (e.g. kitchen cooking and cleaning), at a cost comparable to a car lease (~$300-400/month).
“robotics will take a little longer, but I think we'll have the chat GP moment in the next two to three years. What could that look like? A robot that doesn't need to be programmed. It learns.”
Children being educated today with the paradigm 'study hard, get into good college, get a good job' are being prepared for a world that won't exist; the education model needs fundamental restructuring around curiosity and learning rather than credentials and employment.
“A 5-year-old kid going to kindergarten today is being told by their parents go to school study hard get into good college and then you'll get a good job. That won't be the paradigm.”
Khosla predicted in 2000 (in a New York Times interview) that AI would require humans to redefine what it means to be human; in 2012, he published two TechCrunch blogs predicting AI would eliminate the need for doctors and teachers.
“But fast forward in the year 2000, I gave an interview which is in the New York Times, so it's on the record that said AI will have us redefine what it means to be human, which is this utopia thing we were just talking about. Will people need jobs? Um that was 20 year 2000 or 2001. 2012 I wrote two blogs. One was called do we need doctors under the premise AI will do all the doctoring expertise that we need. This is 2012 2012 it's in techrunch you can look up and the second was called do we need teachers idea that all tutoring will be done by an AI too.”
The 2030s will see a faster rate of demise of Fortune 500 companies than ever before, because most Fortune 500 companies should not be run even remotely the way they're running by 2040, and the transformative reinvention won't come from existing companies — someone new will reinvent each industry.
“one of my predictions is the 2030s will see a faster rate of demise of Fortune 500 companies than we've ever seen. It's actually pretty high anyway, but we'll see a rate accelerate dramatically.”
Capitalism exists 'mostly by permission of democracy,' so if democracy is sustained, society will adjust to abundance more easily; but China's 'Tiananmen Square techniques' for enforcement mean it can impose its preferred outcomes faster, making the next 15 years critically important for which political philosophy dominates the planet.
“We live in a capitalist system but people forget capitalism is mostly by permission of democracy. Yes. in our system. If democracy sustains um I I I think we will adjust easier.”
Khosla made a $3 million investment in Juniper Networks in 1996, which became a $7 billion return—2,333x the initial investment—making it rare in venture capital to get even a 1,000x return.
“We made $7 billion on a $3 million investment when almost no venture investment ever made even a billion.”
The binding constraint on humanoid robots is intelligence, not hardware — China has produced impressive hardware form factors (boxing, running), but they aren't learning robots and fail when the environment changes; the breakthrough needed is intelligence that learns and adapts without being programmed for each task.
“It's the intelligence. Clearly, we've seen out of China many many hardware form factors... but they're not learning robots. you change the environment and they don't do as well.”
AI videography technology that allows specification of precise camera parameters (lighting angle, pan, zoom) has reached capability in Khosla's portfolio where full movie production is only a matter of time away.
“we're building an AI videographer in our portfolio. you specify, you don't just generate video, you specify what the lighting angle is and what the how you pan and zoom the camera and those kinds of things. So the people who resort to traditional means will not do well. Uh people who do do movie production today we can't do a full movie that way but that's only a matter of time, right? Only a matter of time.”
Khosla made a $1 billion+ investment in OpenAI in 2018 (more than twice his largest venture investment in the prior four years) because he had been convinced by 2016 that AI breakthroughs were imminent based on two signals: (1) the rate of progress on AI benchmarks vs human performance was accelerating exponentially, and (2) the best talent globally was flowing into AI research.
“So it was a very large investment more than twice the largest invest initial investment I'd made in four years of venture capital. Wow. So it was a conviction bet.”
Khosla's investment philosophy prioritizes high-consequence-of-success over high-probability-of-success: he starts with 'I want high consequences of success; I don't care about the probability of failure' rather than 'I want to reduce risk to increase probability of success.'
“my view is very simple most people reduce risk to increase the probability of success I do the pos opposite start with I want high consequences of success I don't care about the probability of failure”
Dystopian AI outcomes (displacement of workers, AI doomism, authoritarian control) are not inevitable but rather choices societies make; the technology will produce abundance regardless, but societies must choose how to organize around it.
“the dystopian views will be a choice society makes the utopian view is going to happen anyway the world of abundance only a matter of when not if”
By 2040-2050, society will enter an era of such abundance that the need to work will go away entirely; people will only work on things because they want to, not because they need to pay their mortgage.
“We're going to be in a era of abundance that's so large it's very hard for people to imagine. The simplest way to say it is the need to work will go away. People will work on things because they want to work on things, not because they need to work on things to pay their mortgage.”
A novel public transit system using small, self-driving vehicles in bicycle-lane-width infrastructure can provide 10x the capacity of light rail or car-based transit, is cheaper than both, and costs less to operate than traditional transit, while being hailed on-demand like Uber.
“So and I think Whimo will have a great business for delivering people from transit routes to the last mile. There's only one question in a city when it comes to traffic. Can you increase throughput 10x for the same street width without redoing the street width? So we designed a public transit system of self-driving vehicles in bicycle lane width. So it's easy to insert in a city. Has 10 times the capacity of a light rail system or or a car system. So more capacity than light rail system in a bicycle lane width and it's on demand because the cars just you hail them like an Uber.”
Khosla attributes his longevity and success to being internally driven by curiosity and impact rather than money or others' opinions — he and his colleagues would do the same work unpaid, focusing on larger impact so returns take care of themselves, and he would choose more impact over more return every time provided LPs get a comfortable minimum.
“I'm internally driven. I don't care what others think of me... we do it for a different reason than making the most money because we focus on larger impact. The returns really do well or take care of itself.”
The Screen Actors Guild agreement signed last year locks existing studios into a paradigm that makes them order-of-magnitude less effective at content generation than AI-enabled producers, ensuring that incumbents will lose to new competitors using AI.
“They essentially signed existing traditional studios up to a paradigm that makes them order of magnitude less effective in content generation because we're building an AI videographer in our portfolio.”
Hubris is an essential ingredient of great entrepreneurship; founders need arrogance ('only I can do it') to attempt paradigm-shifting innovations that domain experts believe are impossible.
“that was very much that from you know my first startup was go on take on the computer business everybody big is dumb. Mhm. uh which is a little bit of an arrogant viewpoint but if we hadn't had that kind of hubris we probably wouldn't have attempted what we attempted so I think hubris is an important ingredient of great entrepreneurship yes we'll build an AGI right like that's hubris it's arrogance only I can do it”
Khosla 'instigates' (rather than 'invests in') certain society-altering technologies like fusion and public transit, sometimes starting the company from scratch to ensure the field exists and progresses, beyond just backing existing teams.
“in the more society societally impactful technologies, I will go through actually starting the company. Mhm. Like fusion, Bob was a senior fellow Commonwealth Fusion. Bob was a senior fellow at the MIT plasma fusion lab when I met him and he had ideas and and and I got him sort of like instigated so to say.”
Healthcare spending could be reduced because providers currently have an incentive to reduce costs for individual patients (cutting their own costs) vs. hospital systems (reducing revenue); under capitated payment models like Medicare Advantage, where a fixed payment covers all care, AI's cost-reduction would be maximized because cost savings equal profit increases, not revenue loss.
“Reducing cost for a hospital system is reducing their revenue. Of course, for a physician, reducing cost is cutting their cost down... But fortunately we have a small part of our health care system very small part where payments are capitated. what's called the capitated system or Medicare advantage where you get paid a fixed amount to take care of a patient roughly... That world will be so much more effective because they won't care about reducing costs which AI can do.”
Experts are terrible at predicting the future because they extrapolate the past; entrepreneurs invent the future they want and are not biased by experience. Therefore, in innovation, inexperienced people in a field are often superior to domain experts.
“Here's the first thing I would say in these 40 years in all the examples I've just cited for you, and I could give you more. None of them were done by somebody who knew the area. Mhm. Experts are terrible at predicting the future. They extrapolate the past. Entrepreneurs invent the future they want. Uh it's a bastardization of an Alen K quote but you know so you imagine the world and you try and make it happen and you're not biased by what is experience experience is a set of biases that prevent you from making mistakes and that's what experts do and that's why very rarely do experts innovate in their area.”
Khosla's selection criteria for founders emphasizes: people who think from first principles (not bound by conventional wisdom in their field), people who learn rapidly, and people whose plans change significantly over 3-month periods (indicating adaptation to new information), NOT deep expertise in the domain.
“So you know one of the things I always get asked is what do you look for in entrepreneurs? You know I seldom look for deep expertise in their area. I do look for people who are thinking from first principles and learning rapidly. Even in YC batch if I'm talking to a YC partner about some startup the question that's most important to me is how much have they changed their plan over the last 3 months. Yeah. Rapid evolution. rapid thinking, rapid learning, uh, is what matters.”
Khosla is 'bullish' on energy innovation broadly, with two primary vectors: (1) fusion at multiple startups, and (2) super-hot geothermal drilling at 450°C depth, which provides 6-10x more power from the same well due to both Carnot efficiency improvements and higher heat content of steam.
“I'm very bullish about energy. All sort many formats of energy. So uh I think there's two large formats. Fusion is one and there's multiple approaches to fusion. But I'm also very excited about super hot geothermal.”
Primary care in the future will provide 'multipality' (every AI primary care physician knows enough gastroenterology, cardiology, oncology, etc. to provide those services in addition to primary care), making the traditional specialist model obsolete as AI expertise is both broad and deep.
“everybody on the planet for less than a dollar a month will get a free primary care free physician initially called primary care but they'll be u do everything my son's company curi is called multipety primary care because the AI knows enough gastroantology to provide that in now the hard problem is the American medical association won't let the AI write a prescription”
Medical imaging and diagnostics can be automated: self-driving MRI machines eliminate the need for technicians (previously a bottleneck in cardiac MRI capacity), and self-driving ultrasound machines reduce the cost-per-scan from ~$25K equipment to removing the need for a $150K/year technician, suggesting that removing the human labor cost is more valuable than improving the technology.
“I do think uh there is diagnostics imaging and by the way we have the most coolest self-driving for MRI machines. So you don't need MRI machines. Now you need today technicians to run MRI machines. But if you have a self-d driver, you don't need that. Cardiac MRI is so specialized because the heart is moving um that you can't staff centers with cardiac MRI technicians. We turn an hour and 15 minute exam into 20 minutes, single button press. You can even do it remotely because so we did the same for ultrasound self-driving for ultrasound machines. An ultrasound machine might cost 25 grand, but a technician to do a cardiac MRI will cost you 150 grand.”
Climate change is 'mostly solved' at the technology level: fusion and super-hot geothermal can provide zero-emission energy; green cement and low-emission steel can be produced at lower cost than traditional methods; the path to net-zero emissions by 2050 is clear, and most required technologies will be proven in the early 2030s.
“I'm relatively optimistic. I see our path clear to getting to zero emissions by 2050. Uh I think most of the technologies to that will be proven in the early 2030s.”
Khosla's approach to founder coaching/advising emphasizes brutal honesty over hypocritical politeness, and he views his role not as a board member (with governance authority) but as a 'debating partner' who challenges founders' thinking while ultimately leaving decision authority with the founder.
“I almost never worked at a board. So I don't even go to boards now...my responsibility shifted from having Jack do better uh to governance for the public markets. It's it's not very fun.”
Khosla reframed a robotics company's business from 'automating Caterpillar machines' to 'renting you the driver by the hour,' which is a more compelling sell because it solves the customer's real problem (equipment downtime due to lack of operators) rather than a technological problem.
“Just yesterday I was talking to this founder. It's it's a very cool company self-driving for these heavy machines, bulldozers and loaders. That is cool. Right. And we had an interesting session and I reframed his business from automating big caterpillar machines to I'll just rent you the driver. You have trouble staffing drivers. Most of these are in mines are in remote places. Equipment multi-million dollar piece of equipment sits idle because you don't have an operator. I said just do rental by the hour and guarantee you will always have a driver available. Always have a this reframing the entrepreneur is pretty excited about. It's the same thing but it's not.”
Khosla provided a 2-hour presentation to convince the leadership team of a robotics company not to sell their company, but explicitly signed off saying 'you have my vote anyway, you can choose to use it or tear it up,' deferring the final decision to management.
“There was a company uh I'm forgetting the name called light something optical company uh the team wanted to sell. I only asked they give me two hours with the whole team to make a and I did over the weekend the presentation I gave to them on why they should not sell. But before that presentation, as long as I had the two hours, yeah, uh I signed the papers and said, 'You have my vote anyway. You can choose to use it or tear it up.'”
The dystopian AI outcome is a choice society makes, whereas the utopian world of abundance will happen anyway — it is only a matter of when, not if.
“the dystopian views will be a choice society makes the utopian view is going to happen anyway the world of abundance only a matter of when not if and within our lifetimes”
Within the next 5 years, any economically valuable job humans can do, AI will be able to do 80% of it, with very few exceptions like heart surgery or brain surgery.
“almost certainly within the next 5 years any economically valuable job humans can do AI will be able to do 80% of it with a few exceptions like heart surgery or brain surgery but by and large 80% of 80% of all jobs can be done by an AI”
Humanoid robots will be the dominant AI application in the 2030s because: (1) the world is organized around the humanoid form factor (doors, tools, spaces designed for humans), (2) hardware form factors from China are already impressive, (3) the limiting factor now is *intelligence* (learning, adaptation) not robotics, and (4) once humanoid robots exist with general intelligence, they will eventually exceed human capability in all domains.
“robotics will take a little longer, but I think we'll have the chat GP moment in the next two to three years. What could that look like? A robot that doesn't need to be programmed. It learns. It isn't programmed to do tasks. And it and it's a generalized, you know, maybe physical robot in our in our space. But the world is organized around the humanoid form factor. Mhm. And so it's probably a humanoid.”
Biological intervention will approach computer programming in precision: design molecules and genetic treatments for single-patient genotypes, with single-shot genetic treatments for diseases like sickle cell becoming standard within 10 years.
“The level of precision that's possible I think in the next 5 years or 10 years biological intervention will approach computer programming in being able to precisely program things design molecules. So we are starting to we have a couple of startups looking to do drugs for one patient at a time. So the whole drug only applies to one patient for one genetic Wow. anomaly.”
Drug discovery will be accelerated and improved by: (1) AI design of better lead compounds, (2) higher probability of success for drug candidates (currently <5% of leads reach FDA approval; AI can increase this), and (3) replacement of human trial phases with high-throughput testing using human genetics and organ models on robots.
“we are seeing an explosion in AI in drug design. Yes. Both small molecule and are you pretty bullish on that? I'm very bullish on that. Do you think it will um speed up the rate of drug discovery slashmake it cheaper or do you think it's more about it will help us discover undiscoverable things? probably a little bit of both.”
Large-scale innovation almost never comes from incumbent companies in the business—it comes from new startups. Historical examples: Amazon (not Walmart), Netflix/YouTube/Facebook (not NBC/CBS), Uber (not Hertz/Avis), Airbnb (not Hilton/Hyatt), SpaceX (not Boeing/Lockheed), Tesla (not incumbent automakers).
“retailing, you know, when we invested in Amazon, I was a client in 96. Uh nobody thought you could compete with Walmart. You'd think Walmart would in a way. Yeah. Look at media. Was it NBC or CBS? Or was it little things like Netflix and Twitter and YouTube and Facebook? And nobody even knew they were in the media business. Uber. Did that come from Herz or Avis? Car rental or taxi companies? No. Airbnb come from Hilton or Hayatt? Y they have more rooms than Hilton and Hayatt now. uh SpaceX or rocket lab instead of Boeing and Lheed you'd expect to be able to do space launches.”
Khosla Ventures does not calculate IRR (internal rate of return) and has not done so in at least the last 10 years, because the firm prioritizes building 'something significant' over financial optimization; they assume returns will follow from impact.
“We don't think of investments. We've never calculate never means not in last 10 years I can think of I can't think of a single instance where we calculate an IR. Mhm. Not one. Mhm. An investment firm that never calculates IRS because we want to build something significant and then the economics will work out.”
If all medical expertise became free via AI, the healthcare system would have to be completely redesigned, and this transition will not happen from existing healthcare companies but will require new startups to reinvent the system entirely.
“If all medical expertise is free, you have an unlimited number of primary care doctors, oncologists, gastroontologists, mental health therapists, uh you you name it. How would you redesign the health care system? Be crazy. Uh that transition won't happen from existing companies. almost somebody new will reinvent this.”
The 2030s will see a faster rate of demise of Fortune 500 companies than ever before, because most Fortune 500 companies should not run the way they currently operate even remotely.
“my predictions is the 2030s will see a faster rate of demise of Fortune 500 companies than we've ever seen. It's actually pretty high anyway, but we'll see a rate accelerate dramatically.”
Jack Dorsey was fired from Twitter, had four failed startups, and had no business experience, yet Khosla invested in Square (Dorsey's company) in 2010. Traditional wisdom would reject him; instead, Khosla backed him because he recognized Dorsey's ability to build—past failure predicted future success, not future failure.
“Jack Dorsey is great uh on paper, but when we invested in Square in 2010, the following was the situation. He had just been fired from Twitter. Twitter was a tiny company. just been fired and he had like four star failed startups. Uh his history on Wikipedia like he had four failed startups. So uh you know traditional wisdom would say don't back him.”
Khosla has extremely high confidence (>80%) that the world of abundance will happen, and >50% confidence it will happen by 2050; he would be shocked if it did not happen by 2060.
“What's your confidence level that this is the future high 80%. higher. I have very high confidence this world happens. Uh I think greater than 50% that it happens by 2050.”
Khosla will not sit on corporate boards after a company goes public because governance responsibility conflicts with his role as a founder's advocate; after taking Square public, he stepped off the board to maintain his advisory relationship with Jack Dorsey without governance burdens.
“I almost never worked at a board. So I don't even go to boards now. You won't show up to a board, do you even mean? I don't get on boards. Yeah, I see. Okay.”
The social contract will require fundamental redesign to accommodate a world where most jobs are automated and the need to work is gone, but Khosla believes this redesign is doable and doesn't dwell on it as a major obstacle.
“Is that going to require us to change anything about like the social contract to get there? like yes I'll come back to that but what I want to say is somewhere starting in 2030 the changes are going to be so disruptive that it's hard to imagine how it sorts out”
Khosla has high confidence (above 80%, greater than 50% by 2050) that a world of abundance happens within our lifetimes, and is essentially certain it happens before 2060 — timing is hard to predict but the direction is not.
“high 80%. higher. I have very high confidence this world happens. Uh I think greater than 50% that it happens by 2050.”
Khosla is 70 years old, feels 25, and hopes to work 80 hours/week for another 25 years (until age 95) as long as health permits, because the work is intrinsically motivating (learning + impact).
“In my lifetime, I at least look at 25 years. I'm 70. Easy. More. When you're talking about this dystopia, I think what was implied in what you just said was sort of a self-inflicted dystopia where we just, you know, humans don't decide to organize correctly to take advantage of this amazing abundance.”
Khosla gives a talk titled 'Plausible Tomorrows' with a dozen predictions about the future that sound implausible unless explained in detail, suggesting that future breakthroughs (robotics, energy, transit, medicine) are not speculative but are grounded in solvable technical challenges.
“I was mentioning this talk I did. It's called plausible tomorrows. So you can Google that too. Um, I have a dozen predictions that look really implausible unless somebody sits down with me and I walk them through each and they say, 'I see no reason this won't happen.'”
Keith Rabois returned to Khosla Ventures, a 'generationally great' re-recruiting that happened because Khosla Ventures offers something different: the ability to challenge founders without governance burdens and focus on impact over returns.
“I had um I had Keith on this podcast and what you're describing is a big part of I think this you know what drew him back to Kosla. Yeah. By the way on that topic I'm curious like I that I read that as one of the sort of like generationally great sort of recruiting or re-recruiting um you know moves that has happened”
I've been doing venture capital for 40 years and have gone through every large innovation cycle, but I've never seen a cycle like this one.
“I've been doing venture capital for 40 years, so I've gone through every large innovation cycle. Uh I've never seen a cycle like this”