Transcription
Next, we have a speaker who I respect a lot. He's the founder of Exponential View, and I tend to believe in exponential views. He's written a book of that ilk and has lots of other accolades, like the World Economic Forum advisory stuff. So, I won't, I won't talk about the other accolades, but he is a former founder, so he appreciates that, and I think he understands the exponential view. So, I'd love to hopefully get a view on where the world is going. So, let's start with 2035.
2035. We'll hopefully still have our CEO summit. Mostly, hopefully most of you have IPOed and gone away, uh, and there's a new batch, uh, but maybe there won't be a need for a new batch. But tell us what 2035 is like. I mean, I think there still will be. The 2035 is a great time frame because it, it's 10 years. It's really, really difficult. As Dwars said this morning, first thing, 2050, 2060, 2070 almost feels easier because you can pick science fiction off the, the shelf. Um, and the reason I think this 10 years is, um, is interesting is because, um, it's long enough for there to be generational change across, uh, the economy. So, people who are vice presidents of big firms will be in the C-suite, uh, in, in 10 years' time, and they will have gone from agitating the use of ChatGPT to being the people who run the organizations.
Um, it, it's also really long enough for, uh, the, the impact of, of commitments today, uh, to be felt and somewhat in the dim and distant past, whether it's Stargate, uh, or whether it is, uh, you know, the, the big data centers being built in the Middle East. So, what does it actually look and feel like? Um, energy and, uh, intelligence at that point have become substantially cheaper than they are, uh, today. We can look at the cost curves of the decline of tokens. Um, and, you know, we saw 150x, uh, decline in about a year and a half for GPT-4 quality tokens. So, you'd be looking at, um, if this even makes any sense, right? Five or 10 cents for a billion tokens of, of inference in today's measures. But we may have a different architecture by then. So, perhaps that's not exactly the right way to, to look at it.
In terms of, of energy, we will start to see a real consistent, um, downward pressure on the cost of, uh, of electricity, predominantly because solar, um, will really continue its exponential growth. I mean, one thing to note with solar is that in 2024, we added 600 gigawatts of capacity, which is about 35%, um, of what was the installed base. But we're actually, uh, demand constrained. So, the Chinese factories have got 1.2 terawatts of manufacturing capacity out there, and at some point, we'll start to pick up that demand. We will have, I'm sure Bob will have delivered the first, uh, working fusion reactor, but we'll only have one. From, I think we'll be well past solar. Well, so, you and I disagree on the solar and fusion thing because I think what will, will happen is that the, just the sheer economics of solar, um, being decentralized will, will being available today. Yeah, I'm only kidding. But yeah, but we, but we'll be so, so that period of time, I think you get a moment where the key inputs in the economy, which is brainpower and fundamentally energy, whether it's embedded, are much, much cheaper than they are today. And that should trigger, um, a change in the base level of economic growth, and that has been 3 to 4% for much of the last 60 or 70 years. Maybe we'll get an extra two or three points, maybe an extra four points. I don't think we'll get the 30%, but we'll get, we'll get something significant. So, we are talking about a world where intelligence is free, which is every little AI agent has all the brainpower and knowledge of everybody in this room, and of course, much, much more. So, intelligence is virtually free. Mhm. Uh, and I'll come back to my other question, which I asked earlier this morning, which is, when do we get a 10-person company doing a billion dollars in revenue? Mhm. But intelligence is free, and if fusion and solar, and solar is for a period, a different kind of electricity. So, solar for production of other goods and services is probably a good use of electricity. Intermittent electricity, not dispatchable. In that world, what's valuable?
Well, you, you know, we always shift, um, the sense of what ends up being valuable in, in a world. Uh, and the jobs that many of us, us do, or our parents might do. My dad was an accountant. Didn't have any value 150 years ago. And David Autor, who's an MIT economist, shows that 60% of the job categories that Americans, um, have today didn't exist at the time of, of, of World War II, the end of World of the War. And, and so, you know, the historical record always tells us that that ultimately, when technologies are as dramatic as these ones will be, our polit, our social arrangements, the political economy starts to change. So, it's hard to know exactly what jobs will emerge, but there will still be. I mean, one thing that is, is, is sort of easily noticeable is the idea of status, right? Unless we change the way in which people relate to, uh, comparing ourselves to each other, status goods will still be important. And we can manufacture status mimetically, right? We can just manufacture it through a TikTok meme or whatever replaces that. So, there will always be this, I think there will always be this sense that that you can construct a, a sense of value. But if you, if like me, I use Venode, I use O3, and I'm not as smart as you. And even in my main line of work, um, I would say that more often than not, O3 gets to where I would get to. It just gets there five or 10 times faster. And we're already facing that in 2025. And I don't feel any less capable. I actually feel much more capable because my backlog of things I want to do in my life is so huge. So, I think the other thing that becomes valuable personally is, do people, are people able to create ambition? And are we able to create, um, a sense of ambition in a world where we have access to so much capability?
So, in this world where intelligence is free, and the physical world is subject to robotic AI that manipulates the physical world, and electricity is free, or energy is free, what throttles the rate of growth? Well, the thing that is the, um, is ultimately going to be the blocker is the law of physics. Um, so, the way I think about this is that, you know, energy, uh, enables us. Intelligence allows us to be smarter about that enablement. And the way we can collectively organize, whether it's as companies or agent swarms, is a force multiplier. So, energy is always at the, the foot of, of all of this. And pragmatically, in the US and in Europe, energy is the thing that that gates, will gate this. And we can see this from the curves, right? So, algorithms are improving in their efficiency by a factor of 10 or 100 every year. Chips are improving in, you know, FLOPS per dollar or FLOPS per watt, anything from 40 to 70% a year. The price of energy through solar, 10% a year. The addition of grid connections across the US, not even as fast as that. So, you can see ultimately, there's this kind of weird, you know, you've got these geometric curves on the one hand, you've got these linear curves around the energy system. So, over the next 10 years, pragmatically, I think that entrepreneurs are going to be super creative about making sure the compute fabric gets the power it needs. But there's a kind of systemic issue that that will become the gating, gating.
Let me challenge you a little bit. Of course, we were talking earlier about Vaclav Smil. How many people have seen his work? Uh, a few hands. He's one of my experts. I say experts are experts in the previous version of the world, not the version we are trying to create as entrepreneurs. Yes. And so I have a very different answer. Yes. Grids are very hard to set up unless you solve one problem: high-temperature, room-temperature superconductors. What will do that? We have lots of scientists working on it. But I suspect in the next five years, we will multiply the number of scientists working on high-temperature superconductors. Is Bob Mumgaard here? Yes. By a thousandfold because we'll have AI scientists working on high-temperature superconductors. And once we have high-temperature superconductors, the current existing grid is more than enough, right? More than enough. So, we solve it discontinuously with a change in assumptions about what can and can't be done, which is what entrepreneurs do so well. I, I completely agree with you on Vaclav. I think he has the world's best rearview mirror. So, if you want to know where you've come from, by the way, I always disagree with Vaclav partly because Bill, Bill Gates loves him. Yes. And I have to just disagree every time. He, he is a, he's also a fantastic historian. So, I mean, I do recommend, recommend his books, and I, uh, absolutely agree with you. I got so angry reading him, I couldn't finish him. Yes, I, I know that. But, you know what? I'm going to say just there, he's got this big book, Energy and Civilization. The first 300 pages, which are about the first, you know, up to about the year 600 AD, are really fantastic. So, I agree with you that a discontinuous, um, innovation like superconducting, um, mag, uh, superconductors could really transform the basis upon which we make our assumptions. Um, but I'm being quite, I'm being practical about the speed with which institutional and some social organizations can ultimately change. I mean, even if we invented those, maybe Bob has a few of them in his back pocket, and he can pull them out. I think it would still take years to change the laws to allow us to make those changes in, in situ. So, so in a way, that's why 10 years, 2035, is kind of interesting because it's a sort of ugly teenage amount of time, which is neither short-term nor, nor long-term.
So, you, I want to come back to the question of how these two cultures clash. But we've seen electrification happen historically. I hate history because it's not, it doesn't foretell the future. We've seen engines, and first steam engines, and then, uh, electric motors, and things, microprocessors, then the internet and mobile phones, and now AI. Mhm. So, we've seen these platform chip transitions, and many people look at them and say, how long did they take to permeate into society and everyday use? Is this time different? Well, I mean, I think it's, of course, it's different. I mean, like you, the history is only so useful in in showing some patterns. I think what's hugely different with, with AI. I mean, one thing that is really noticeable is that it's a massive demand pull. It's a massive demand pull by enterprises and consumers. And that's not what electricity was like in the, like 1920s and 30s, right? There wasn't quite the same demand pull. I think the other thing that's really, um, that's really different is that we actually have a, a culture and mechanism which is entrepreneurship, venture capital, um, and networks to share that know-how to people who really want to make, want to make a difference. So, my view is that it'll be much, much shorter than any previous, uh, transition that we, we saw. But that doesn't mean that it's instantaneous. It just, it still takes time to get a big company to, to do all of this. It may be two years rather than five, but it's not, it's not two days. So, I really do feel that this time is, is, is really different because there are kind of fundamental tailwinds, um, from entrepreneurs, capital that can deliver these products, the demand from consumers, um, and also the, you know, the fact that the big platform companies have really lent into this. I mean, back to, you know, when, when you, um, were building networking companies, right? 25 years ago, I was a telecom investor at one point in my life. Yes, you've got a couple of, of okay deals, I understand in history, historically, but, you know, you'll remember at the time that AT&T and MCI and France Telecom were not promoting the internet. They were trying to slow it down. And what we have today is all of the big companies, the big tech companies from Salesforce to your Microsoft and so on, are absolutely doubling down on this as, as customers and as infrastructure providers. So, I think that creates opportunity.
Let's, let's be clear. I mean, I think you bring up a really good example, one of my favorites in my life. Um, in 1996, I talked to the senior management of every telecommunications carrier in the US. Every single one of them, no exceptions, other than one little thing called UUNET. Yes. Uh, said they would never adapt TCP/IP in the public network. Yes. Never. I called Cisco, and their CTO, Ed Kozel, I said to me, they will never do a router above OC-12, which is less than the service I have to my house today. Mhm. Right. But, so this was the worldview, and telecom wasn't going to change. So, I gave a keynote talk at the Goldman Sachs telecom conference in the year, around the year 2000, sometime, and I said Nortel, Lucent, Alcatel, Newbridge, which was the ATM company, when they were at their all-time high stock prices, uh, I declared at this Goldman Sachs conference at lunch that they would be going bankrupt sometime when they're at their all-time high stock price. Talk about conviction, right? Um, within three years, they were all, all gone from roughly $80 a share to two or $3 a share. So, my, the reason I tell this story is these changes happen rapidly. And because we had done TCP/IP at Juniper, we ended up with a 2500x return on a $3 million investment. You can compute the numbers. Um, so my point is, change happens quickly. Yes, it, it happens really quickly. But it's the Hemingway quote, right? How did your business fail? Slowly at first, and then all at once. And there's something else about that period, which was a telecoms bubble. And when I've gone off and talked to public market investors and, you know, talked to, you know, events like that, people bring up the telecom's bubble with respect to what's going on in the public markets today with Nvidia. And there's a chart where they overlay Cisco's share price and Nvidia, and they follow each other, and then Cisco's drops and doesn't return for a long time. And they push back and say, look, aren't we seeing the same, the same thing? And, and of course, here, we don't believe that. Here's one reason why I think it's completely different, which is back in the telecom's bubble, the, uh, rate of revenue growth for internet communication services, data services, was running at about 17 to 20% per year, but the rate of CAPEX growth was running at about 80% per year. Whereas today, we're, we're flipped, right? So, for everything that Microsoft and Google, Meta are putting into their data centers at those rates of growth are dwarfed by the rates of growth of, of consumer and business demand for AI services. Right? The, the revenue demand is growing much, much faster than the capital that is being poured, uh, into, uh, into the infrastructure. So, I, you know, I mean, I, I feel it's a completely different picture. I think it's going to happen very, very quickly. But just that one caveat is that big companies still move a little bit slowly, even when they need to move quickly.
But, but again, I want to challenge this, and I want to encourage all of you to challenge this view. I was 25 or something when I started Sun. Mhm. DEC was the number two computer company in the world, Digital Equipment Corporation. We had fewer people than their local sales office in Santa Clara. It took five years before DEC was not making computers. They were mostly doing customer support and harvesting their installed base, and we were the dominant, right, number one computer company. IBM was number one, but IBM also became a support and services company. My point is this change happens rapidly. If entrepreneurs drive the change. Yes. Because somebody has to say, "I'm going to go disrupt this," and then it happens. And if Bob Mumgaard hadn't said, "I'll go tackle fusion," he, he was just a senior fellow, I think, at the MIT Plasma Fusion Lab when we met, and he said, "I'm going to go do this." And then the world changed on fusion because everybody then, including the Department of Energy, said, "Not in the next 50 years," right? So, entrepreneurs make these changes happen. Bringing it to today, in the last two weeks, I've talked to Sam Altman. We'll talk to him again tomorrow, and he sees this exponential capability. So, I'll ask him, but he'd probably say by the end of next year, we'll see more progress from today to then, 18 months, than we've seen in the last five years from GPT-2, GPT-3 days, right? So, there's this world had the exact same conversation in the last two weeks with Larry Page, and he's even more optimistic than that than me. I'm conservative in that crowd, um, but, so, there's this world. Then I talk to boards of directors of various large companies, a number of, uh, healthcare companies. They've asked me to speak to their board, and they're, in my view, rearranging deck chairs on the Titanic. Yes. Right? So, there's this, "Oh, we need to make these marginal sure adjustments," and I'm going to introduce AI in this little customer support function, instead of saying, "How do we reinvent?" So, my view, in the next 15 years, we will see many, if not most, of the Fortune 500 go under. So, they won't adjust rapidly. Yes, you're right. But they won't be relevant in my worldview. That's a pretty large statement to make. I, I mean, I'm so sympathetic to that. I, there's, there's so much in that, uh, question, comment. Um, so, a couple of thoughts. One is that if you, if you look at the use of coding assistants, um, which are, you know, the preeminent use of AI in, in, in business, uh, today and industry today, and you can see Google has announced one today. OpenAI announced theirs last week. Everybody, you know, loves Cursor and Replit and Windmill. Um, the, the total number of, of users of those services, based on some similar web data, is about three to four million per, per week. Now, they're meant to be about 35 million professional developers. So, I find it quite remarkable that we're already at like 10% market share in just like, in just a year, but it, it is still only 10%. But I wager that something else is happening there, which is that there are lots of people who do not count as professional developers who are now moving into that space. So, the effectively, the TAM is expanding. That's the Replit model. That's the Replit model. That's sort of my forecast last year at this audience when I said there'd be a billion programmers on the planet is what I said last year. Yes. I mean, you can certainly, you can certainly see that, um, see that that term expanding significantly because people will want to do whatever coding, coding turns into. Um, on this question about how quickly can companies run into, uh, you know, run into trouble, the capabilities are expanding really, really fast, and you can see it through the, the, the, the measures. There's a, a research group called Meter. Meter have got one particular benchmark, which is how long can an AI agent undertake human work without intervention. And that benchmark is doubling every seven months. And so, when we extend that benchmark out, we can say by about 2030, uh, you'd expect to be able to get 200 hours of professional software development done for less than $50, right? So, that's pretty remarkable because presumably we're paying our developers more than $50 for every five weeks they work or three weeks they work. And, and that, that curve is, is an exponential curve because it's doubling every, every seven months. And you don't see reasons why I see more reasons why that would shrink than reasons why that would get, that would get slower. And, and so I think that then you find this situation with, with large companies where they are absolutely exposed because doing process re-engineering, which is like the management consulting term for it, is, is almost impossible in, in these companies. And so, what it does is it creates an opportunity for someone who is, is built from the ground up to go really quickly. And a great example of that, I know you hate history, but let's go back to history, is Henry Ford. So, Henry Ford was not the first car manufacturer, nor was he, in fact, the first car manufacturer to use electricity because famously, he used electricity for the electric, uh, moving assembly line. The first car manufacturers all bought into electricity really quickly. The way today's insurance companies have bought into AI. They got single pendant lights in their workshops so people could work an extra hour. And you can imagine at their board meetings with a like a McKinsey chart going, "Tick, we're doing AI," but they hadn't done AI, right? They had, "We're doing electricity." They hadn't done electricity. They were, they kind of bought into it. But it was Henry Ford who believed in it. And the point about entrepreneurs today in the room here, the next vintage in the next year or two, is they will all believe in it. They will believe in AI, and they will be able to build that system from, from the ground up. What we're seeing in big companies today at the moment is, with a few exceptions, um, is using AI to pare back costs, but nobody has ever built a great company by cutting costs at 15%. I've talked to multiple boards of healthcare companies because it's a tough area, and I, I have a lot of discussion. Then I say, if all your physicians were free oncologists, everything you do, right, is set up to have less and less access for the most expensive traits, your neurologists, your oncologists, your gastroenterologists. If all these things cost zero and less than the nurse you have, how would you redesign the healthcare system? And the answer is blank stares, every single time. And so they're all saying, "I can make billing a little more efficient," or "have scribing AI," or "have," right? But, but show how, let's look at how difficult it is, um, in that, in that environment, and how they'll defend themselves. So, take, take self-driving cars, right? You take Waymo's, um, and whatever comes, you know, as, as Tesla's rolls out. Those cars, um, hit pedestrians less often than human drivers do, and certainly less than, you know, Uber drivers do. And so that will have an immediate impact on, um, emergency room admissions in, in health systems, right? And these are profitable, pretty profitable lines. So, if you're a health insurer, or you're operating a health system, you're not just worried about in three or four years' time having a neurologist that is available on an iPhone. Actually, you're at risk from the fact that self-driving vehicles are going to reduce the amount of injuries on, on the roads. So, so they, and at the same time, I think what they do have, though, is they have a certain amount of political clout, and they have this regulatory moat. And I expect them, and I can only imagine that they will fall back to that moat and really argue from the perspective of, you know, regulation, consistency of care, um, you know, what happens with older people who may not want to move to the, to, to digital systems. I, I think it only slows it down. I don't think it kind of prevents. Yes, it's a matter of time. It's a matter of time.
I want to go to questions, but I want to give you the following perspective. In this group of people, there's people working on an AI primary care doctor, an AI mental health therapist, which should be free, an AI oncologist in this group. And I could go on, an AI nurse or care management, uh, an AI accountant, an AI chip designer. In fact, more than one AI chip designer, an AI structural engineer, an AI salesperson, an AI marketing person, just in this group. These are like literally profession by profession. All of these in two or three years will clearly be superior to their human counterpart. But let's go to questions. That's what gives me optimism and worry about how discordant this battle between the old world and the new world will be. And I think it'll be one of the major issues between 2030 and 2035. But open it up to questions in the back there. Uh, I'm interested in what you said, uh, about major health systems and relying on regulations. Do you think that makes non-US markets potentially leaders in fields like deployment of AI doctors? Just curious. That's, it's a great, great question. I think they're regulated. Um, I mean, I can't speak to markets I don't know. Uh, but, you know, in the UK, with the, the National Health Service, uh, it, there are certain advantages in the sense that we have singular sources of data. And I think you've got a company, you've backed a company out of the UK that was able to to build from some of that, uh, but there's just much, yeah, and, but there's just much less money around in those other markets.
Can I ask, come back to your 10-man unicorn, billion-dollar company? So, we said, you know, when would that happen? And, you know, YC has talked quite a lot about, uh, how the teams are much smaller, and they're hitting $1, $5, $10 million ARR with much, much smaller teams. But I, I still think that if your company is growing really quickly, there will be benefits to hiring humans because you still need humans to manage the workflows of, of the agents. But only 10. I, I said 10. Yes, only 10. Well, maybe a department head, a VP of marketing, a VP of finance. Yes. But, but I also, I just, I just wonder about about whether you, you still, in the margin, will benefit from adding that 11th person because he, or she, can manage another 5,000 agents, right? And, and that is really, um, where someone who's super aggressive and says, "I need to localize. I need to have physical presence in a load of markets." I may go beyond the 10. Mhm. Uh, yes, your comments on the speed of potential collapse of the Fortune 500 is chilling for those of us who are selling to enterprise. Makes me think very differently about the LTV of those contracts and how I might approach pricing and packaging. Do you have any advice? I, I mean, there are some, uh, bits of the market that are clearly going to, going to grow, right? We, we know that, uh, demand for compute, amount of demand for data, is, is going to grow. So, as long as you veer into there, I would also say that even these big companies, um, will spend a lot of money to fend off their death. So, you, you can also see that, uh, that working. The other thing is you could always pivot to restructuring, uh, because there'll be a lot of that.
One last question somewhere in the back. Who has, who has the mic over here? I have here. Can we get a mic up front here? Okay, got two. Hi. Um, so building on this idea of labor disruption and optimism, there's obviously going to be, you said, you know, doesn't help you if there's riots in the streets. Um, how are you guys thinking about new types of job creation, new types of work that is going to come about? I think a lot about creative industry for what we're doing, all that kind of stuff. But what kind of new and exciting jobs are giving us optimism about this future where there is a lot of labor disruption? I think the, the reality is that, uh, what, what under any scenario, there's going to be a period of time when new jobs have not yet been created and other jobs have been, been lost. And so, you have to, you have to navigate that. It's really hard to predict beyond the trivial things you can get from a Google search for what those, what those jobs might be. But I would say that there is, um, that there will probably need to be some measure of a new social, social contract. So, one thing that's happened in the US over the last 70 years has been the share of national income that's gone to labor has dropped from 70% to the 60% that Darkh mentioned this morning. It was 66% eight years ago. So, it's really fallen off, um, quite quickly. And that is a leading indicator for pitchforks in the streets. So, I think this, this new MAGA bond, um, that Brad talked about, is, is a really interesting idea. Like a good, another version of that is the Alaska, the Alaska Wealth Fund, which pays like a $1,500 to $25,000 dividend to every citizen in Alaska based on the oil wealth. Um, and I can imagine that the way that you, I'm not sure whether it's politically tractable, but as a system, and that's why the MAGA, MAGA fund is such a great indicator, um, that essentially says, we know that this shift is going to move to further, further and further towards capital. So, we need to make sure people have some skin in the game and some way of participating starts to provide some, some insulation, um, which, which doesn't directly answer the question as to what will the jobs of the future, uh, be, except back to the point I made, which is that status is always something that humans value. So, we will may well create things that live around that. Um, I will end, since we are out of time, with the following comment, and I'll again remind you, Chattam House Rules. When I asked Larry Page what happens by 2035, he said, "None of the things you worry about, how will corporations adapt or not, are is going to matter." Thank you all. Thank you.