Latest / Faster, Please! โ The Podcast / ๐ The business of AI: My interview with technology analyst and investor Azeem Azhar
Transcript
- 0:00SPEAKER_01: Welcome to Faster Please,
- 0:01SPEAKER_01: the podcast, the audiovisual companion of my Faster Please newsletter.
- 0:06SPEAKER_01: I'm your host, Jim Pethokoukis.
- 0:08SPEAKER_01: My guest today is Azim Azhar.
- 0:11SPEAKER_01: Azim is a longtime angel investor in tech,
- 0:13SPEAKER_01: and he's the founder of Exponential View,
- 0:15SPEAKER_01: a research platform focused on helping leaders understand emerging technologies
- 0:20SPEAKER_01: and their impact on society.
- 0:22SPEAKER_01: He's a digital fellow at the great Stanford Digital Economy Lab,
- 0:26SPEAKER_01: executive fellow over at Harvard Business,
- 0:28SPEAKER_01: and the author of the Exponential Age book.
- 0:31SPEAKER_01: He also recently co-authored the state of the AI economy.
- 0:35SPEAKER_01: It's really one of the first serious attempts to measure the size and trajectory of
- 0:39SPEAKER_01: the AI economy from the demand side, know what customers are actually paying for.
- 0:44SPEAKER_01: Azeem, no surprise,
- 0:48SPEAKER_01: we're going to talk about where the AI economy stands, where it might be headed.
- 0:53SPEAKER_01: Great to chat with you again.
- 0:56SPEAKER_01: I first wanted to get a sense of what kind, I think you're an AI optimist, correct?
- 1:00SPEAKER_01: Do I got that right?
- 1:02SPEAKER_00: Am I an optimist?
- 1:03SPEAKER_00: I'm going to use your terms, Jim.
- 1:04SPEAKER_00: I'm probably more on the Acela line than I am on the San Francisco consensus start of the world.
- 1:13SPEAKER_01: Well, good.
- 1:14SPEAKER_01: That's perfect because I have a quote here from Demisa Sabas of Google DeepMind.
- 1:18SPEAKER_01: He recently posted this on Twitter and I wanted you to think about it and then answer.
- 1:24SPEAKER_01: Hassaba said, This is a pivotal moment in human history.
- 1:28SPEAKER_01: Artificial general intelligence is probably only a few short years away.
- 1:33SPEAKER_01: The magnitude of the technology's impact will be unprecedented.
- 1:37SPEAKER_01: Perhaps ten times in the Industrial Revolution,
- 1:40SPEAKER_01: we could reach a point where resources are no longer the limiting factor
- 1:44SPEAKER_01: for human progress, leading to an amazing new era of abundance.
- 1:49SPEAKER_01: What do you make of that sort of thinking?
- 1:52SPEAKER_01: And is that the kind of thing that would factor into your sort of broad investment thesis?
- 1:59SPEAKER_00: I always wonder how something like that will play out in Peoria or Atlanta or
- 2:06SPEAKER_00: in Milton Keynes in the UK and the places where ordinary people live.
- 2:12SPEAKER_00: I think what Demis said has been true for nearly 200 years in the Western world.
- 2:18SPEAKER_00: The moment that we started to have sustained economic growth starting in
- 2:22SPEAKER_00: the 18th century,
- 2:24SPEAKER_00: that world looks like a world of abundance compared to any of the humans who lived
- 2:27SPEAKER_00: in the 300,000 years before then.
- 2:30SPEAKER_00: And what did we learn about the modern economy,
- 2:33SPEAKER_00: industrial capitalism,
- 2:35SPEAKER_00: is that It may look like abundance before the inflection,
- 2:40SPEAKER_00: but when you're in it,
- 2:42SPEAKER_00: it really looks like a day-to-day grind where you still have to fight for the things
- 2:47SPEAKER_00: that you care about.
- 2:48SPEAKER_00: There are still temporary shortages.
- 2:50SPEAKER_00: There are certain types of scalability.
- 2:51SPEAKER_00: There are scarcities that have disappeared.
- 2:53SPEAKER_00: People don't die of smallpox anymore.
- 2:55SPEAKER_00: So there is a scarcity of people dying of smallpox.
- 2:58SPEAKER_00: So we have made progress.
- 3:00SPEAKER_00: But the notion of abundance in general, I think defies the
- 3:10SPEAKER_00: You know, you always need energy to get something done of time.
- 3:14SPEAKER_00: You can't, you know, you and I can't both eat the same banana, Jim.
- 3:18SPEAKER_00: I mean, if we started at opposite ends, I promise you I'd eat more than half of it.
- 3:22SPEAKER_00: But these are still,
- 3:25SPEAKER_00: So I'm always a bit skeptical about that.
- 3:27SPEAKER_00: But I also agree that what AI could do is find a way of taking us to a new level
- 3:35SPEAKER_00: of growth above this last 300 years historical average of 1.6 to 2% per year.
- 3:42SPEAKER_01: I think the sort of, again,
- 3:43SPEAKER_01: the realist looking at like where we are today,
- 3:48SPEAKER_01: that certainly infuses,
- 3:49SPEAKER_01: you know,
- 3:50SPEAKER_01: really this great report that you put out called The State of the AI Economy.
- 3:54SPEAKER_01: And it's, I'm gonna, this is a very incomplete way of describing it.
- 3:59SPEAKER_01: But these are the two headline numbers that you use,
- 4:01SPEAKER_01: so I'm going to use them too,
- 4:02SPEAKER_01: where you say the generative AI economy,
- 4:05SPEAKER_01: Has generated about $110 billion in sales over the last year.
- 4:09SPEAKER_01: And on an annualized basis, the revenue run rate exceeds $175 billion.
- 4:15SPEAKER_01: And to be clear,
- 4:16SPEAKER_01: these are numbers about companies and consumers buying AI capabilities,
- 4:23SPEAKER_01: applications.
- 4:25SPEAKER_01: These are from the model makers, the cloud providers.
- 4:28SPEAKER_01: We're not talking about the chip makers or anything like that.
- 4:33SPEAKER_01: One way I interpret that,
- 4:35SPEAKER_01: those two numbers,
- 4:36SPEAKER_01: those revenue numbers,
- 4:37SPEAKER_01: from the perspective of someone who lived through the 1990s internet boom.
- 4:43SPEAKER_01: So we're talking about actual money that actual customers are handing over.
- 4:48SPEAKER_01: This is not about valuations or companies buying from each other.
- 4:52SPEAKER_01: We're talking about real people and businesses.
- 4:56SPEAKER_01: I think people and businesses are paying for stuff If you're a bubble worrier,
- 5:03SPEAKER_01: this is good news because real stuff is being paid for,
- 5:07SPEAKER_01: real money is being handed over.
- 5:09SPEAKER_00: Yeah, real money is being handed over.
- 5:11SPEAKER_00: It's about three times faster than the internet boom or mobile apps.
- 5:17SPEAKER_00: There is some venture capital in there that is being spent.
- 5:20SPEAKER_00: So we know that the average venture backed company in America spends,
- 5:26SPEAKER_00: I think the number that Ramp,
- 5:28SPEAKER_00: which is a fintech that tracks this stuff, spends about $11 per employee per month.
- 5:34SPEAKER_00: There are about $4 million.
- 5:36SPEAKER_00: People employed by venture backed companies.
- 5:39SPEAKER_00: So arguably 4 billion of that 110 billion is series A,
- 5:44SPEAKER_00: series B seed funding on pre profitable companies.
- 5:48SPEAKER_00: And then you also do have the fact that the big frontier labs like OpenAI
- 5:52SPEAKER_00: and Anthropic are spending significantly Thank you very much.
- 6:12SPEAKER_00: We're going to change the rules of American capitalism and the American economy
- 6:16SPEAKER_00: for this particular sector and say risk capital no longer counts as real revenue.
- 6:21SPEAKER_00: You're still at a number that's around 85,
- 6:24SPEAKER_00: 90 billion dollars, which is still the fastest growing sector we have ever seen.
- 6:29SPEAKER_00: So, yes, I mean, I think that this is real, real money.
- 6:33SPEAKER_00: But I would say that.
- 6:35SPEAKER_00: That doesn't mean we're out of the woods.
- 6:37SPEAKER_00: This is an early shoot.
- 6:40SPEAKER_01: Let me jump in then and say, two things.
- 6:44SPEAKER_01: First, you mentioned,
- 6:46SPEAKER_01: I want to get to the out of the woods,
- 6:48SPEAKER_01: but just quickly, a lot of times we're grasping for historical analogies.
- 6:53SPEAKER_01: And you mentioned that this is happening much faster than prior IT waves.
- 6:59SPEAKER_01: So when I hear that data point,
- 7:02SPEAKER_01: it sort of informs this question,
- 7:04SPEAKER_01: which is,
- 7:06SPEAKER_01: Is the reason this thing is scaling and that this money is being spent,
- 7:10SPEAKER_01: is this reflecting companies buying these tools because they're finding,
- 7:18SPEAKER_01: because they started to buy them,
- 7:19SPEAKER_01: they're finding them useful, so they're buying more of the services.
- 7:23SPEAKER_01: Is it that because they're putting these things in productive use
- 7:26SPEAKER_01: or is this still really, hey, what's this all about?
- 7:29SPEAKER_01: We don't want to miss out.
- 7:31SPEAKER_01: We at least better try to make use of them.
- 7:33SPEAKER_01: Like, where are we?
- 7:35SPEAKER_00: I think we've got past the stage,
- 7:38SPEAKER_00: certainly in the US,
- 7:39SPEAKER_00: where this is about random experimentation and giving everyone a co-pilot.
- 7:45SPEAKER_00: Companies are starting to get much more sensible and strategic and they're learning
- 7:49SPEAKER_00: about how to deliver this.
- 7:53SPEAKER_00: What we don't have yet is Binders full,
- 7:58SPEAKER_00: to use a Mitt Romney phrase, binders full
- 8:02SPEAKER_00: of huge ROI successes from every sector.
- 8:07SPEAKER_00: And I'm not sure when history would tell us we would get those binders.
- 8:11SPEAKER_00: But what we normally know with technology is that there is a There's a J curve of investment.
- 8:16SPEAKER_00: You know, you invest up front and then you start to see the returns.
- 8:20SPEAKER_00: I think one of the things that's really difficult to tell right now is that
- 8:24SPEAKER_00: a company that is being really successful with generative AI will be running in
- 8:30SPEAKER_00: the red very heavily now because they will have been successful with a pilot
- 8:33SPEAKER_00: and they will run a dozen or more new projects in parallel,
- 8:38SPEAKER_00: all of which they are investing in.
- 8:40SPEAKER_00: And a company that has no clue about what it's doing and is really poorly run
- 8:44SPEAKER_00: and is just throwing money into the fire.
- 8:47SPEAKER_00: We'll have exactly the same sea of red in the same number.
- 8:49SPEAKER_00: So you can't really tell from that data one from the other.
- 8:55SPEAKER_00: What we have to just have some faith in is what happens when we talk to CEOs
- 8:58SPEAKER_00: and what they tell us.
- 8:59SPEAKER_00: And do they sound like they are men and women driving off a cliff at high speed?
- 9:05SPEAKER_00: Or do they sound like people who are really trying to fix the car on the road
- 9:09SPEAKER_00: and realize it's got potential?
- 9:10SPEAKER_00: And in general, they sound like the latter.
- 9:13SPEAKER_01: I recently did a chat with Eric Brynjolfsson,
- 9:18SPEAKER_01: who to me sounded long run and even medium run optimistic,
- 9:22SPEAKER_01: but a little disappointed that there are not those binders full
- 9:26SPEAKER_01: of high return applications.
- 9:30SPEAKER_01: Are you disappointed at sort of the pace that we're seeing of the productive use
- 9:37SPEAKER_01: of this technologies that ends up transferring down to the bottom line and we see,
- 9:41SPEAKER_01: you know, big, big Profit gains and eventually big productivity gains.
- 9:47SPEAKER_00: You know what?
- 9:47SPEAKER_00: I'm more disappointed in how I disrespected history two or three years ago.
- 9:52SPEAKER_00: Because what we're actually learning is that companies are going through
- 9:55SPEAKER_00: the similar type of process that Paul David describes around electrification.
- 10:01SPEAKER_00: Which is that you can put the light bulbs in the workshop.
- 10:04SPEAKER_00: That is the co-pilot.
- 10:06SPEAKER_00: But to move to the moving assembly line requires a lot more work and a lot more expertise.
- 10:12SPEAKER_00: And that's what I see when I talk to company bosses, both in the US and in Europe.
- 10:17SPEAKER_00: They can point to a handful of Growing examples which look like they are AI native
- 10:25SPEAKER_00: and a large set of examples which is just hanging light bulbs
- 10:29SPEAKER_00: in your existing systems.
- 10:32SPEAKER_00: So the disappointment I think for me is just that I should have had much more sense
- 10:37SPEAKER_00: to look at what the economists and the academics had done
- 10:39SPEAKER_00: when they'd studied general purpose technologies previously.
- 10:42SPEAKER_00: I am constantly surprised by the sophistication of apps that I see Even large,
- 10:49SPEAKER_00: slow-moving companies.
- 10:50SPEAKER_00: I spent some time with a government contractor in McLean in Virginia,
- 10:55SPEAKER_00: or is it McLean up in Virginia?
- 10:58SPEAKER_00: And they're quite a stodgy business and they really had some impressive apps.
- 11:02SPEAKER_00: What it brought back to me was there's a depth of expertise that is building,
- 11:07SPEAKER_00: but it is still really difficult to Transform a firm.
- 11:12SPEAKER_00: It's very easy for a management consultant to say it's really hard to do it in real life.
- 11:17SPEAKER_00: And the thing that would keep the optimist going is that what I hear are companies
- 11:22SPEAKER_00: who are ready to persist with learning how to do that and feel that they are,
- 11:27SPEAKER_00: even if it's gone slower in 26 than they thought it might have done two years ago.
- 11:32SPEAKER_01: Is what we're seeing companies I think these are two different things,
- 11:36SPEAKER_01: being more efficient at what they do versus them doing like new things
- 11:42SPEAKER_01: to create new sources of revenue.
- 11:44SPEAKER_01: I assume those are two very different things
- 11:46SPEAKER_01: and assuming those are two very different things, are we seeing much of either?
- 11:52SPEAKER_00: Well, I mean, it's definitely the focus is on the cost efficiencies in the first instance.
- 11:58SPEAKER_00: And there's simple, both kind of managerial and financial reasons for it.
- 12:03SPEAKER_00: Financial reason is that you effectively do better with a cost efficiency
- 12:09SPEAKER_00: because you don't have to worry about the margin that your new product
- 12:12SPEAKER_00: or service has to achieve.
- 12:15SPEAKER_00: And managerially,
- 12:16SPEAKER_00: it's easy because you already know exactly how that process works
- 12:20SPEAKER_00: and what good looks like and what the interfaces are between each process.
- 12:25SPEAKER_00: And that's where companies started.
- 12:26SPEAKER_00: And some went too fast.
- 12:28SPEAKER_00: I think we know the story of some of the early fintechs or a couple
- 12:30SPEAKER_00: of car companies who've been trying to hire higher execs back.
- 12:36SPEAKER_00: In terms of genuinely new and novel products.
- 12:41SPEAKER_00: Which I would assume,
- 12:42SPEAKER_00: are those gonna come from the existing companies or we're talking like
- 12:45SPEAKER_00: a new entrepreneurial wave?
- 12:47SPEAKER_00: Well, I mean, you just have to look at Anthropic.
- 12:49SPEAKER_00: So, you know, Anthropic has new products for AI.
- 12:54SPEAKER_00: Google, for example, makes about $2 million per employee.
- 12:58SPEAKER_00: McKinsey makes about $450,000 per employee.
- 13:01SPEAKER_00: Anthropic makes about $10 to $12 million of revenue per employee.
- 13:05SPEAKER_00: That's at the level of the most elite in Wall Street,
- 13:08SPEAKER_00: like Jane Street, well above the Goldman Sachs' and Morgan Stanley's of this world.
- 13:13SPEAKER_00: And that, I think, speaks to what happens when you have an AI-native company.
- 13:17SPEAKER_00: So typically,
- 13:19SPEAKER_00: The first significant breakthroughs will come from new entrants
- 13:23SPEAKER_00: because incumbents face the innovators dilemma.
- 13:26SPEAKER_00: You know,
- 13:26SPEAKER_00: they have to struggle with their existing products,
- 13:29SPEAKER_00: their existing channels and customers.
- 13:31SPEAKER_00: But, you know,
- 13:32SPEAKER_00: I think we will start to see the returns from incumbents
- 13:37SPEAKER_00: because they have scale immediately.
- 13:38SPEAKER_00: You know, the exception of Anthropic, right?
- 13:42SPEAKER_00: Every startup takes 10 years to get to scale.
- 13:44SPEAKER_00: Whereas if one of these banks like Bank of America,
- 13:47SPEAKER_00: Brian Moynihan,
- 13:48SPEAKER_00: the CEO,
- 13:48SPEAKER_00: talked about hundreds of thousands of AI workflows that they were running,
- 13:55SPEAKER_00: their assistant Erica saving something like 10,000 FTE equivalents.
- 14:00SPEAKER_00: I think a company like a business like Bank of America,
- 14:03SPEAKER_00: looking at it from the outside,
- 14:05SPEAKER_00: would be well positioned to adjust the product offering That could only have existed
- 14:11SPEAKER_00: because of AI.
- 14:13SPEAKER_00: But, you know, would you expect a CVS or a slow moving construction business to do that?
- 14:19SPEAKER_00: Probably not as quickly.
- 14:21SPEAKER_01: Just to sort of focus again on that revenue number.
- 14:26SPEAKER_01: I'm looking for sort of green shoots for like that this will result in
- 14:30SPEAKER_01: a big productivity impact.
- 14:32SPEAKER_01: Is that one way?
- 14:33SPEAKER_01: I don't want to overstate what those numbers mean.
- 14:35SPEAKER_01: Is that one way is that Since these companies,
- 14:39SPEAKER_01: there's a lot of companies, smart people, they're finding some value.
- 14:43SPEAKER_01: So can I look at those revenue numbers and say,
- 14:47SPEAKER_01: this is a good sign that this is an important,
- 14:51SPEAKER_01: this is going to be an important general purpose technology,
- 14:55SPEAKER_01: setting aside super intelligence,
- 14:58SPEAKER_01: that just as an important new technology that will have a big impact,
- 15:02SPEAKER_01: I can look at those revenue numbers and say, well, that's a sign.
- 15:05SPEAKER_01: That's a sign that it will.
- 15:08SPEAKER_00: I think, in short, all the things that you'd need to see are starting to be visible.
- 15:16SPEAKER_00: So companies are spending a little bit more.
- 15:19SPEAKER_00: Companies are putting in better budgeting and financial controls on how they're
- 15:26SPEAKER_00: That is a really,
- 15:27SPEAKER_00: really important sign of maturation because you now have an internal financial model
- 15:33SPEAKER_00: that says, well,
- 15:33SPEAKER_00: if we give these software developers $1,500 worth of AI every month,
- 15:38SPEAKER_00: which is a limit to Uber.
- 15:40SPEAKER_00: We have to get something back.
- 15:42SPEAKER_00: And that suggests that they are getting their sort of teeth into it really realistically.
- 15:49SPEAKER_00: So a lot of people were worried about that the idea that attribution
- 15:54SPEAKER_00: and cost management would come into
- 15:59SPEAKER_00: is a bad sign.
- 16:00SPEAKER_00: It's actually a sign of what mature businesses look like.
- 16:05SPEAKER_00: I've launched products out of incumbent companies, sometimes over 100 years old.
- 16:10SPEAKER_00: It's the first six months or year where the finance director says to you,
- 16:15SPEAKER_00: Just don't really worry about the cost control, just like get it working.
- 16:18SPEAKER_00: And it's year two when he comes to you and says, okay,
- 16:20SPEAKER_00: Azeem, now I'm going to hold you to the fire and you've got to hit your numbers.
- 16:26SPEAKER_00: And we're starting to see that in AI.
- 16:29SPEAKER_00: Now, what I don't know, Jim,
- 16:30SPEAKER_00: and this is,
- 16:31SPEAKER_00: I can still see there being a path in all of this where certainly for large parts
- 16:37SPEAKER_00: of sectors like the software sector,
- 16:40SPEAKER_00: This might end up just being an additional cost that they have to pay, right?
- 16:45SPEAKER_00: You're a software vendor,
- 16:46SPEAKER_00: you have to have some AI,
- 16:48SPEAKER_00: so you're having to pay OpenAI or Anthropic something just to stay in the game.
- 16:52SPEAKER_00: It almost becomes like a hygiene tax that you end up paying.
- 16:57SPEAKER_00: I mean, maybe that will happen for some sectors,
- 17:00SPEAKER_00: but for other sectors,
- 17:02SPEAKER_00: I think that the ability to To process information in the way and undertake tasks
- 17:11SPEAKER_00: That you might not have been able to previously at low cost has
- 17:15SPEAKER_00: to open opportunities.
- 17:16SPEAKER_00: But I think the thing that we have to understand is that businesses are not about doing tasks.
- 17:20SPEAKER_00: They are about managers coming together to allocate resources
- 17:25SPEAKER_00: against particular objectives with a view to what the risk is.
- 17:29SPEAKER_00: No business was ever about, can we summarize an email quicker than before?
- 17:34SPEAKER_00: And the question is,
- 17:34SPEAKER_00: at what point do AI tools meaningfully come in
- 17:38SPEAKER_00: to Make those decisions happen more effectively,
- 17:42SPEAKER_00: either at higher quality, at lower cost or at higher speed.
- 17:45SPEAKER_00: And that,
- 17:46SPEAKER_00: I think,
- 17:46SPEAKER_00: is the piece that is the difficult part of the transformation
- 17:49SPEAKER_00: that companies will be dealing with over the next year or two.
- 17:55SPEAKER_01: Where I don't have to ask this kind of like that series of questions when it's sort
- 17:59SPEAKER_01: of obvious,
- 18:01SPEAKER_01: it's obvious that this is that this is a technology being used in all sorts of ways
- 18:07SPEAKER_01: by all sorts of businesses providing very tangible results like.
- 18:14SPEAKER_01: What does the year leading up to that obvious point look like?
- 18:19SPEAKER_01: Again, is this showing up in annual reports, in earnings calls?
- 18:25SPEAKER_01: What does that world look like where it starts to become very,
- 18:28SPEAKER_01: very obvious and I don't ask these kinds of questions anymore?
- 18:32SPEAKER_00: I think one of the earliest points publicly that we could rely on will be
- 18:36SPEAKER_00: in earnings calls,
- 18:37SPEAKER_00: where CEOs and CFOs will start to be much more specific,
- 18:41SPEAKER_00: as Brian Moynihan and his CFO at Bank of America were a few weeks ago,
- 18:45SPEAKER_00: about what they have done,
- 18:46SPEAKER_00: what they And I think the reason that will happen,
- 18:55SPEAKER_00: if that does happen,
- 18:56SPEAKER_00: that will happen on earnings calls is because about three quarters
- 18:58SPEAKER_00: of American CEOs think their job depends on getting AI right.
- 19:03SPEAKER_00: So if they're getting AI right,
- 19:05SPEAKER_00: they are really going to want to tell the world in ways
- 19:08SPEAKER_00: that feel somewhat unimpeachable.
- 19:11SPEAKER_00: We do know that about 30% of earnings transcripts now mention AI,
- 19:15SPEAKER_00: that about a half of those have a quantifiable number attached.
- 19:20SPEAKER_00: Most of that quantifiable value tends to be in a cost saving.
- 19:25SPEAKER_00: And as that mix shifts towards productivity or operating leverage or something else,
- 19:33SPEAKER_00: and we hear it across more and more sectors,
- 19:35SPEAKER_00: that's the moment where you will feel, okay, we're getting towards escape velocity.
- 19:42SPEAKER_01: The impact of AI on job loss or job creation,
- 19:46SPEAKER_01: has that changed at all over the past three years?
- 19:49SPEAKER_01: And sort of where are you on that?
- 19:51SPEAKER_01: Because certainly, you know, the headlines remain screams, screams about job loss.
- 19:58SPEAKER_01: Is that, again,
- 20:00SPEAKER_01: sort of where are you at on that issue, especially as you talk to actual companies?
- 20:06SPEAKER_00: You know,
- 20:06SPEAKER_00: I think I may have been bullied by the amount of noise there was out there
- 20:12SPEAKER_00: because I'm a bit of a simple guy.
- 20:14SPEAKER_00: I think of the lump of labour fallacy.
- 20:16SPEAKER_00: I think of complementary roles.
- 20:18SPEAKER_00: I think of the fact that we constantly invent things for people to do.
- 20:25SPEAKER_00: And the data is really,
- 20:27SPEAKER_00: really mixed,
- 20:28SPEAKER_00: as you know,
- 20:29SPEAKER_00: and Anthropics chief economist Peter McCrory just summarized a lot of this,
- 20:34SPEAKER_00: saying it's extremely difficult to show job losses.
- 20:40SPEAKER_00: Even in the labour markets with the US so at full employment coming off
- 20:45SPEAKER_00: the COVID shock, this market is a really confused market for anyone to benchmark.
- 20:51SPEAKER_00: But that said,
- 20:52SPEAKER_00: as you rightly point out,
- 20:53SPEAKER_00: when I do talk to senior execs,
- 20:56SPEAKER_00: There are some who say we're not going to engage in headcount cuts.
- 21:01SPEAKER_00: And there are others, I mean,
- 21:02SPEAKER_00: particularly in the financial services,
- 21:04SPEAKER_00: where it's reasonably clear that that is the case.
- 21:08SPEAKER_00: Now, that's a job loss.
- 21:10SPEAKER_00: Is it necessarily a net job loss?
- 21:13SPEAKER_00: Well, that's going to depend on what else is going on in the labour market.
- 21:17SPEAKER_00: I don't think that we see a cliff edge shock to to employment
- 21:27SPEAKER_00: because there are all sorts of factors that slow us down.
- 21:32SPEAKER_00: One of those factors is just verification and validation
- 21:36SPEAKER_00: and getting these AI systems to work end to end in companies is really difficult.
- 21:41SPEAKER_00: We also know that politics will play a really significant role.
- 21:44SPEAKER_00: So in China,
- 21:45SPEAKER_00: there was a lower court case a few months ago saying companies can't Use AI as
- 21:52SPEAKER_00: a reason to lay people off.
- 21:54SPEAKER_00: Now, of course, things in China don't happen randomly.
- 21:57SPEAKER_00: So there is a little bit of an alignment that is emerging there.
- 22:00SPEAKER_00: And I think what's happened in data centers in the US,
- 22:03SPEAKER_00: which is remarkably the first thing to bring the left and the right together,
- 22:09SPEAKER_00: the first political issue I can remember in a long time,
- 22:13SPEAKER_00: shows that Americans can organize themselves politically to resist a change
- 22:19SPEAKER_00: that they're uncomfortable with.
- 22:21SPEAKER_00: So we've seen it happen with data centers.
- 22:22SPEAKER_00: I'm not going to forecast that this will happen if there is a spectra
- 22:26SPEAKER_00: of widespread job losses,
- 22:28SPEAKER_00: but I would just say that left and right have got together on an issue.
- 22:33SPEAKER_00: And so there are these things that will slow down the The way in
- 22:39SPEAKER_00: which companies talk about this and maybe give managers pause to think.
- 22:43SPEAKER_00: And you've seen this in the change of the narrative from Dario Amadai in
- 22:46SPEAKER_00: the last few months where he doesn't talk about job losses in quite the way he did
- 22:50SPEAKER_00: a year ago.
- 22:51SPEAKER_00: And I also think that as companies realize that there was so much tacit knowledge
- 23:00SPEAKER_00: In their workforce, that they just let walk out the door.
- 23:04SPEAKER_00: I think back to when Boeing opened their new factory, I think it's in South Carolina, right?
- 23:11SPEAKER_00: So it was,
- 23:12SPEAKER_00: you know,
- 23:12SPEAKER_00: it's across the continent from where all the embedded tacit knowledge of the
- 23:16SPEAKER_00: The Boeing engineers who made the 707 and the jumbo jets and so on was.
- 23:20SPEAKER_00: And they started to use many more subcontractors for key roles.
- 23:24SPEAKER_00: And you had really difficult quality issues because all of that works on
- 23:29SPEAKER_00: a spreadsheet, but it doesn't work in real life.
- 23:31SPEAKER_00: And I think CEOs will start to,
- 23:34SPEAKER_00: unless they feel under a lot of pressure to show savings,
- 23:37SPEAKER_00: want to understand what are we losing when We cut an entire function for an LLM.
- 23:45SPEAKER_00: And all of that, to me, slows down the pace a little bit.
- 23:48SPEAKER_01: You're talking about the left and the right coming together.
- 23:51SPEAKER_01: One place they've come together is viewing the current,
- 23:56SPEAKER_01: the companies that look like the current winners,
- 23:59SPEAKER_01: particularly,
- 24:00SPEAKER_01: you know,
- 24:01SPEAKER_01: OpenAI and throughout the companies,
- 24:02SPEAKER_01: you know, going to go public trillion dollar valuations.
- 24:05SPEAKER_01: We'll see that they view them as the winners.
- 24:08SPEAKER_01: And in order to spread Spread the winnings.
- 24:12SPEAKER_01: They think we should tax them or seize some shares and distribute them.
- 24:19SPEAKER_01: Whether or not you think that redistribution idea is a good idea,
- 24:22SPEAKER_01: the assumption that we have identified the winners and it's those companies that
- 24:31SPEAKER_01: Have we really identified the winners?
- 24:33SPEAKER_01: When we look back at the AI revolution a decade,
- 24:36SPEAKER_01: it will be obvious that those are the companies that won and took the spoils,
- 24:41SPEAKER_01: not Not the businesses using the services,
- 24:46SPEAKER_01: not somebody,
- 24:47SPEAKER_01: I don't know, the healthcare industry, maybe that's where the winners are.
- 24:52SPEAKER_01: I mean, do we know who the winners are?
- 24:54SPEAKER_00: Well, if you'd asked me two years ago,
- 24:55SPEAKER_00: the winner was OpenAI
- 24:56SPEAKER_00: and Anthropic was really just playing around doing God knows what.
- 25:01SPEAKER_00: So I don't think we necessarily know who the biggest winners are.
- 25:07SPEAKER_00: Yeah.
- 25:08SPEAKER_00: Yeah.
- 25:29SPEAKER_00: Anthropic or OpenAI have a significant role to play in several years' time.
- 25:34SPEAKER_00: I'm absolutely sure that will be the case.
- 25:38SPEAKER_00: I mean, frankly,
- 25:38SPEAKER_00: if they stopped growing today, they're practically $100 billion revenue businesses.
- 25:43SPEAKER_00: By the end of this year, certainly Anthropic will be approaching that.
- 25:46SPEAKER_00: So that doesn't preclude there being other bigger winners.
- 25:49SPEAKER_00: And the thing that we can start to see is that value really Accrues to
- 25:57SPEAKER_00: the infrastructure layer.
- 25:59SPEAKER_00: So where the processing actually happens.
- 26:02SPEAKER_00: And already today of that $110 billion,
- 26:05SPEAKER_00: I think in June,
- 26:07SPEAKER_00: that 12 months to June, 2026,
- 26:09SPEAKER_00: 78% of it was accruing to the companies who run the servers like Amazon
- 26:14SPEAKER_00: and Microsoft and Google.
- 26:17SPEAKER_00: And that is why you see OpenAI and Anthropic wanting to vertically integrate
- 26:22SPEAKER_00: because they want to be able to control that margin stack.
- 26:26SPEAKER_00: Now, with all of this open source and open weights innovation that's happened,
- 26:31SPEAKER_00: and you've got this Chinese model called Kimi K3, what does that do?
- 26:35SPEAKER_00: Yeah, what does that do?
- 26:37SPEAKER_01: And will we ever be able to talk about powerful American, or non-Chinese at least, open models?
- 26:45SPEAKER_00: You'll be able to talk about American open models.
- 26:50SPEAKER_00: It's too important strategically for The venture capitalists are backing them.
- 26:56SPEAKER_00: There's one, I think, called Reflection AI, which is doing quite well.
- 27:00SPEAKER_00: You also have Mira Murati has released a small open source model.
- 27:06SPEAKER_00: And it's so economically valuable for the infrastructure providers,
- 27:11SPEAKER_00: the companies who are outlaying hundreds of billions of dollars a year for chips
- 27:15SPEAKER_00: and cooling and memory and data centers.
- 27:18SPEAKER_00: For them to be able to offer top tier open source models because they get much,
- 27:24SPEAKER_00: much better economics.
- 27:25SPEAKER_00: So there are lots of reasons why the market would want there to be good open source models.
- 27:31SPEAKER_00: And if politics does not allow for Chinese models to take that role,
- 27:36SPEAKER_00: then entrepreneurs will step in and they will develop open source models for
- 27:40SPEAKER_00: the American market.
- 27:42SPEAKER_01: Do we need to do anything differently to enable The rise of very big,
- 27:50SPEAKER_01: powerful, open models in the United States.
- 27:56SPEAKER_00: My sense is that there's a little bit of understandable attempts
- 28:01SPEAKER_00: for regulatory capture from the big private companies.
- 28:03SPEAKER_00: They're increasing their lobbying.
- 28:05SPEAKER_00: The messaging for years has been about titanic risks that speak to the purpose of a nation.
- 28:14SPEAKER_00: And I think that that's a little bit of a problem.
- 28:19SPEAKER_00: If Demis Hassabis is right,
- 28:21SPEAKER_00: and this is 10 times the industrial revolution,
- 28:24SPEAKER_00: which also means it's bigger than steel and the railways coming out in the US
- 28:28SPEAKER_00: and electrification, one of the...
- 28:32SPEAKER_00: One of the things that normally happens in history,
- 28:34SPEAKER_00: as you know well,
- 28:35SPEAKER_00: is that all of those social agreements that the state makes with us
- 28:41SPEAKER_00: as individual citizens and economic agents come up for negotiation,
- 28:45SPEAKER_00: whether it is where is the frontier boundary of intellectual property or not.
- 28:52SPEAKER_00: During the build out of the US railroads,
- 28:56SPEAKER_00: another great capital investment splurge,
- 28:59SPEAKER_00: which didn't end as happily,
- 29:01SPEAKER_00: I suspect, the steel companies got together and shared all their patterns.
- 29:07SPEAKER_00: It was called the period of Bessemer collective invention.
- 29:10SPEAKER_00: And why did they do that?
- 29:12SPEAKER_00: They did that because So there is historical precedent for new regimes to emerge.
- 29:24SPEAKER_00: I think it's hard for this administration,
- 29:27SPEAKER_00: given, you know, obviously some of the advisers are split about What looks good?
- 29:33SPEAKER_00: And the situation is changing very, very quickly.
- 29:36SPEAKER_00: But if we are right that this is 10 times the industrial revolution,
- 29:40SPEAKER_00: as Demis Hassabis says,
- 29:42SPEAKER_00: and we can get left and right together around data centres,
- 29:45SPEAKER_00: it does feel like there is space.
- 29:47SPEAKER_00: My concern would be that it's going to be much easier for Congress to think in terms
- 29:52SPEAKER_00: of enclosure than it is in terms of Open source and public goods.
- 29:56SPEAKER_00: And so that if there is legislation that comes out,
- 30:00SPEAKER_00: and I haven't kept an eye on it,
- 30:01SPEAKER_00: but I'm just sort of thinking in rough terms about what would be easiest for them
- 30:06SPEAKER_00: to explain,
- 30:08SPEAKER_00: I would suspect it would be in ways
- 30:10SPEAKER_00: that would support closed source proprietary approaches rather
- 30:14SPEAKER_00: than open source approaches.
- 30:16SPEAKER_00: But, you know, it may go the other way.
- 30:20SPEAKER_01: We barely really scratched the surface of that report.
- 30:23SPEAKER_01: We're going to link to it and all that.
- 30:24SPEAKER_01: It's great.
- 30:24SPEAKER_01: But I did want to ask you something about Europe,
- 30:29SPEAKER_01: about the United Kingdom,
- 30:31SPEAKER_01: about Europe and the role of places that are China and America in the age of AI.
- 30:40SPEAKER_01: What is the economic role?
- 30:43SPEAKER_01: Of that region.
- 30:44SPEAKER_01: There seems to be white papers about innovation and we create our own models.
- 30:51SPEAKER_01: Where does that stand?
- 30:53SPEAKER_00: I think there's an incredible opportunity for Europe at the moment,
- 31:00SPEAKER_00: which you're starting to see hints at.
- 31:02SPEAKER_00: So in France, you have Mistral, which makes open source models.
- 31:06SPEAKER_00: It is growing reasonably fast.
- 31:08SPEAKER_00: It's tiny compared to the American firms for $500 million in revenue.
- 31:12SPEAKER_00: But it is a favored choice,
- 31:15SPEAKER_00: particularly of banks who want to use open source because you've got
- 31:19SPEAKER_00: the French legal system, the EU sitting behind it.
- 31:21SPEAKER_00: It's not coming from China.
- 31:24SPEAKER_00: There are also things that states can do,
- 31:30SPEAKER_00: even if they're smaller states like the UK,
- 31:32SPEAKER_00: because they can clear paths,
- 31:34SPEAKER_00: they can help vertical integration, they can get access to best specific talent.
- 31:42SPEAKER_00: And this is not all about large language models.
- 31:43SPEAKER_00: It's also about...
- 31:45SPEAKER_00: Models that can simulate physics or models that can discover new materials.
- 31:49SPEAKER_00: And so that allows you to pick off things that could be quite significant and support them.
- 31:54SPEAKER_00: One thing that happened in the UK is that we got a new prime minister.
- 31:57SPEAKER_00: It's not really news because we have them every few weeks.
- 32:01SPEAKER_00: But he did elevate a minister called Kanishka Narayan to the cabinet, who's the AI minister.
- 32:09SPEAKER_00: And Kanishka has worked in banking.
- 32:11SPEAKER_00: He's worked in venture capital.
- 32:12SPEAKER_00: He is an incredibly...
- 32:15SPEAKER_00: A clever,
- 32:16SPEAKER_00: capable,
- 32:16SPEAKER_00: driven man who understands the issues as well as anyone, and he sits on the cabinet.
- 32:22SPEAKER_00: So what does that then mean?
- 32:24SPEAKER_00: Well, it's not going to compete with Anthropic or OpenAI,
- 32:27SPEAKER_00: and it certainly won't compete with the Chinese Lao I think one
- 32:31SPEAKER_00: of our biggest problems is that within Europe
- 32:34SPEAKER_00: the labour markets are incredibly inflexible and so
- 32:55SPEAKER_00: That raises the cost of hiring for any founder and entrepreneur really, really significantly.
- 33:01SPEAKER_00: And the correlation between the cost to fire somebody and the number
- 33:08SPEAKER_00: of unicorns per capita you Generate is a very strong negative correlation.
- 33:13SPEAKER_00: It's roughly, I'm going to say minus one, right?
- 33:15SPEAKER_00: So the harder,
- 33:16SPEAKER_00: the worst your labor market is,
- 33:17SPEAKER_00: the less flexibility it has, the fewer great breakout companies you have.
- 33:21SPEAKER_00: But these are all things that can change.
- 33:24SPEAKER_00: You know, they're not laws of physics.
- 33:25SPEAKER_00: It's not like we have to discover the Permian Basin,
- 33:28SPEAKER_00: you know, somewhere between Belgium and Alsace-Lorraine.
- 33:31SPEAKER_00: It is things that people could get together and change.
- 33:35SPEAKER_00: I'm not sure they will,
- 33:36SPEAKER_00: frankly,
- 33:37SPEAKER_00: to be honest,
- 33:37SPEAKER_00: in Europe or in the UK, but the data is increasingly making itself clear.
- 33:45SPEAKER_01: You mentioned energy.
- 33:46SPEAKER_01: If we had 30 more minutes, we would do 30 minutes of energy.
- 33:50SPEAKER_01: I know that is a specialty of yours, solar.
- 33:53SPEAKER_01: We'll have to wait with that.
- 33:54SPEAKER_01: Azeem, thanks so much for coming on.
- 33:56SPEAKER_01: Just marvelous.
- 33:58SPEAKER_00: Really enjoy it, Jim.
- 33:59SPEAKER_00: Thanks very much.