The Angle from T. Rowe Price

Benedict Evans argues AI adoption is wide but shallow, citing high costs and fundamental uncertainty in scaling laws.

37 min Listen to the original Episode page
Artificial IntelligenceVenture CapitalCloud InfrastructureEnterprise SoftwareSemiconductorsMarket Strategy

Summary

Benedict Evans compares the current AI boom to the dot-com era, noting that while adoption appears fast due to existing cloud infrastructure, usage remains shallow with limited automation. He highlights fundamental uncertainty in AI science, contrasting it with previous tech waves where physical limits were known, and points to supply chain bottlenecks and high marginal costs as barriers to mass consumer adoption. Evans suggests that companies are adopting multi-sourcing strategies to avoid dependency on single providers, leading to market fragmentation. In the second part, he discusses the shift from traditional CRM systems to AI-native CRMs, emphasizing a move towards mission-based tactics and the psychological challenge of letting go of familiar metrics.

Episode sections

  1. 0:07 – 2:25

    Introduction and Motivation

    Justin Thomson introduces Benedict Evans, highlighting his 25-year career in tech strategy and venture capital. Evans describes his motivation as boredom and curiosity, aiming to find new puzzles in technology analysis.

  2. 2:25 – 5:26

    AI vs. Dot-Com Analogy

    Evans compares the current AI boom to the late 1990s, noting similar euphoria and uncertainty about strategic building blocks. He argues AI adoption looks faster superficially due to existing cloud infrastructure but is actually shallow, with wide but limited usage.

  3. 5:26 – 11:20

    Fundamental Uncertainty and Scaling Laws

    Evans discusses the economic challenge of frontier LLMs being high-cost, fast-depreciating assets. He highlights a fundamental uncertainty in AI science, contrasting it with previous tech waves where physical limits were known, and notes that scaling laws may have practical limits or be bypassed by agentic swarms.

  4. 11:20 – 13:15

    Geopolitics and Market Fragmentation

    Evans predicts a fragmented market with multiple sourcing strategies to avoid dependency on single providers. He notes a decoupling dynamic between the US and China, with Chinese labs optimizing for cost and being slightly behind the frontier rather than chasing AGI at all costs.

  5. 13:15 – 19:29

    Capital Deployment Challenges

    The discussion covers the difficulty of deploying trillions in capital, with hyperscaler CapEx guidance exceeding $700 billion. Evans cites supply chain bottlenecks in semiconductors and power equipment, noting that capital markets are becoming essential as cash flows are outpaced by spending.

  6. 19:29 – 21:45

    Consumer Adoption and Marginal Costs

    Evans argues that high marginal costs prevent the 'free viral' adoption model seen in earlier internet waves. He states only about 5% of ChatGPT users are paying, limiting mass consumer demand and token consumption compared to the industry's capacity.

  7. 21:45 – 26:12

    Future Consumer Applications

    Evans identifies potential demand drivers including automation of boring tasks, overcoming inertia in decisions like insurance, and friction reduction in shopping. He suggests AI agents could change product distribution curves and discovery methods.

  8. 26:12 – 31:27

    Venture Capital and Imagination Asymmetry

    Evans introduces the concept of 'imagination asymmetry,' where entrepreneurs have an advantage over model labs in knowing what to ask. He advises investors to look for fundamental leverage points in industries and to test models personally to understand their capabilities.

  9. 31:27 – 33:07

    AI-Native CRM vs Traditional CRM

    Speaker_0 contrasts traditional CRM systems with complex data schemas against AI-native CRMs that drop all client data onto a simple timeline. This shift allows users to ask new, high-level questions about client churn rather than relying on rigid SQL queries.

  10. 33:07 – 34:16

    Mission-Based Tactics in Business Strategy

    The speaker draws an analogy to military mission-based tactics, where junior personnel are empowered to achieve an objective using available resources rather than following strict procedural steps. This suggests a shift from managing specific metrics to focusing on overarching business missions.

  11. 34:16 – 37:12

    Letting Go of Comfort Metrics

    Speaker_1 and Speaker_0 discuss the difficulty of abandoning familiar metrics and 'knobs and switches' that users have mastered. They argue that businesses must identify what truly matters beyond the comfort of known data fields and manual oversight.

Claims worth verifying

AI today looks like the late nineties in terms of excitement and uncertainty, but adoption is wider yet shallower due to existing cloud infrastructure.

Unidentified speaker (speaker_0) ·2:26

How to verify: Compare user adoption metrics for AI tools (e.g., Copilot) with historical data on PC/mobile adoption; review analyst reports on AI workflow integration depth. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Frontier LLMs are the most expensive and fastest depreciating assets ever created.

Unidentified speaker (speaker_1) ·5:26

How to verify: Review financial filings of major AI model labs (OpenAI, Anthropic, Google DeepMind) for R&D and compute depreciation schedules; compare with historical asset depreciation rates. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

AI infrastructure is being built out way behind demand, unlike the fiber overdeployment of the dot-com bubble.

Unidentified speaker (UNKNOWN) ·6:08

How to verify: Analyze semiconductor and data center capacity utilization rates; compare supply growth forecasts with demand forecasts from industry bodies like SEMI. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Scaling laws have practical limits, and capability can be improved via agentic swarms without further model scaling.

Unidentified speaker (speaker_1) ·8:44

How to verify: Review academic papers on LLM scaling laws; analyze technical reports from AI labs on agentic frameworks and compute efficiency. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Hyperscaler CapEx guidance is over $700 billion for the year, with Meta, Microsoft, and Alphabet at or above 50% CapEx to sales.

Unidentified speaker (UNKNOWN) ·14:06

How to verify: Check quarterly earnings reports and investor presentations for Meta, Microsoft, and Alphabet for CapEx guidance and sales figures. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

Only about 5% of ChatGPT users are paying.

Unidentified speaker (speaker_0) ·20:29

How to verify: Review public statements from OpenAI executives or third-party analytics firms tracking ChatGPT subscription metrics. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Chinese labs are optimizing for being a month behind the frontier at a tenth of the cost.

Unidentified speaker (speaker_0) ·12:47

How to verify: Analyze public statements from Chinese AI labs (e.g., Baidu, Alibaba, Tencent) regarding their product roadmaps and cost structures. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

The shift from information asymmetry to imagination asymmetry favors entrepreneurs over model labs.

Unidentified speaker (speaker_0) ·26:12

How to verify: This is a conceptual claim; verify by reviewing interviews or articles by the CEO mentioned (unnamed) or Evans' own writings on this topic. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Large companies use 400-500 SaaS apps.

Unidentified speaker (speaker_0) ·28:58

How to verify: Review industry reports from SaaS management platforms (e.g., Gartner, Forrester, or SaaS management vendors) on average SaaS sprawl in enterprise environments. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

AI marginal costs are high, preventing the 'free viral' adoption model of the early internet.

Unidentified speaker (speaker_0) ·19:49

How to verify: Compare inference costs per token for modern LLMs with historical web hosting costs; analyze business models of successful AI startups. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Traditional CRMs use complicated data schemas with hundreds of field types, while AI-native CRMs might drop all data onto a timeline with just a date and client name.

Unidentified speaker (speaker_0) ·31:27

How to verify: Compare documentation for legacy CRM platforms (e.g., Salesforce) with emerging AI-native CRM products (e.g., Gong, Chorus) to assess data ingestion methods and schema complexity. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Primary source

AI-native CRMs enable users to ask questions like 'what are our most important clients who are likely to churn soon?' which are hard to answer with traditional SQL queries.

Unidentified speaker (speaker_0) ·32:45

How to verify: Review case studies or whitepapers from AI-native CRM vendors demonstrating predictive churn analysis capabilities versus traditional CRM reporting features. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Mission-based tactics involve defining an objective and empowering junior personnel to determine how to achieve it, rather than prescribing specific steps.

Unidentified speaker (speaker_0) ·33:35

How to verify: Search for academic or military literature on 'mission command' or 'mission-based tactics' to verify the definition and application of this leadership style. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

The speaker argues that users must let go of familiar metrics and 'knobs and switches' to focus on what truly matters in an AI-driven workflow.

Unidentified speaker (speaker_1) ·34:16

How to verify: This is a subjective opinion/claim about user behavior. It can be verified by surveying enterprise software adoption rates and user feedback regarding the learning curve and resistance to AI tools. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

The podcast episode was recorded in October of 2026.

Unidentified speaker (speaker_3) ·35:49

How to verify: Check the podcast's official website or press release for 'The Angle' from T. Rowe Price confirming the recording date. For recent events, verify 2026-08-18 through 2026-11-18. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify hyperscaler CapEx guidance and cash flow statements to assess reliance on capital markets for debt and equity financing.
  • Check semiconductor supply chain reports and power equipment production forecasts for bottlenecks mentioned.
  • Analyze ChatGPT user data to confirm the percentage of paying users versus free users and token consumption rates.
  • Review enterprise CRM software architectures to compare traditional SQL-based schemas with AI-native timeline-based approaches.
  • Investigate scaling law research papers and engineering reports on 'agentic swarms' versus larger model scaling.
  • Assess geopolitical tech decoupling policies between the US and China for impacts on AI lab strategies and sourcing.
  • Examine venture capital portfolios for 'imagination asymmetry' opportunities where entrepreneurs leverage AI for fundamental leverage points.
Full transcriptexpand
0:07
Unidentified speaker (UNKNOWN): And
0:07
Unidentified speaker (speaker_0): he said that he thought there was a shift from information asymmetry to imagination asymmetry. And what he meant by that was you have to have the imagination to know what to ask.
0:19
Unidentified speaker (UNKNOWN): Welcome
0:19
Unidentified speaker (speaker_1): to the angle from T-Row Price, a podcast for curious investors. Just a reminder that outside of the US and Australia, this podcast. is for investment professionals only. Thank you for joining me today. I'm Justin Thompson, head of the T Row Price Investment Institute. My guest today is technology analyst Benedict Evans. Benedict has seen
0:39
Unidentified speaker (speaker_1): various technology waves at close hand in a 25-year career on the sale side as a strategist in the telco industry and as a partner in venture capital.
0:50
Unidentified speaker (UNKNOWN): As we
0:50
Unidentified speaker (speaker_1): talk about today's AI investment boom, we are going to explore why this matters for investors and why this matters now. Benedict, welcome.
0:59
Unidentified speaker (speaker_0): Thank you.
1:04
Unidentified speaker (speaker_1): What motivates Benedict Evans?
1:08
Unidentified speaker (UNKNOWN): Uh
1:08
Unidentified speaker (speaker_0): Boredom and curiosity, I suppose. Um I think I I'm might as a general answer, I did my degree in history, which is read a lot and try and work out what you can say that's interesting and different. And that's what analysis is. In technology, the point that you understand
1:28
Unidentified speaker (speaker_0): something is generally the point you should be looking for a different question. And so you kind of have to keep moving and asking, well, what's the next thing? Um, what's the next puzzle? And so it's kind of nice to have some new puzzles to talk about.
1:43
Unidentified speaker (speaker_1): Interesting because part of our branding is all about curiosity and ask better questions, get better answers. Um, but I've also noticed in that that you have a low boredom threshold, so I hope I don't bore
1:55
Unidentified speaker (UNKNOWN): you.
1:55
Unidentified speaker (speaker_1): Um
1:57
Unidentified speaker (UNKNOWN): there
1:58
Unidentified speaker (speaker_1): is a saying, this is totally different, just like the last time.
2:03
Unidentified speaker (speaker_1): So
2:05
Unidentified speaker (speaker_1): I know every technology wave has its differences,
2:10
Unidentified speaker (speaker_1): but what period in history um you can allude to recent waves of technology, or you can go back to railways canals?
2:20
Unidentified speaker (speaker_0): Yeah.
2:21
Unidentified speaker (speaker_1): Is the closest anal analogue to where we are today?
2:25
Unidentified speaker (UNKNOWN): Um
2:26
Unidentified speaker (speaker_0): I think. from a technology rather than a financial point of view, AI today looks a lot like the late nineties in that this is very exciting and it's clear it's going to change everything. And you have people getting into a sort of state of euphoria um that
2:46
Unidentified speaker (speaker_0): the entire world is looked kinda going to look completely different in like three weeks and you have twenty-five-year-olds who've never really done anything except work in computer science and work for a tech company who suddenly imagine that they understand the whole of the rest of the world and can know exactly how everything's going to work in places that they've never been and things they've never seen. You
3:06
Unidentified speaker (speaker_0): within the technology, like it work it's it's amazing and exciting and it's clearly it's the next big thing, but a lot of it doesn't really work. It just kind of sort of works, yeah. yet. And it's not clear what the actual strategic building blocks are going to be. So, you know, go back to the mid to late 90s. Mark Zuckerberg is in
3:25
Unidentified speaker (speaker_0): middle school, Larry and Sergei are students. Mark um Jeff Bezos is a bookseller with a bad haircut and a goofy grin and no muscles. And it wasn't clear that. AOL and Yahoo were going to disappear off the face of the earth. And that there was going to be this thing called social. It
3:45
Unidentified speaker (speaker_0): wasn't clear how any of it would work. And I think that's kind of where we are now.
3:51
Unidentified speaker (speaker_0): It's very exciting. It's not clear how any of this is going to work.
3:56
Unidentified speaker (speaker_0): The obvious way that this is different, I mean, there's may probably maybe two ways that this is different. One of them is that it looks like this happens much faster, which I'm not sure I think is slightly deceptive.
4:09
Unidentified speaker (speaker_0): And the cloud is already there, the data is already there. Chat GBT, from a consumer perspective, is just a website. So of course you can get hundreds of millions of people using it straight away. I mean, it's an unfair comparison, but it's kind of a relevant comparison. Um However, it actually is taking a lot longer than that to get to real usage. So you've got very wide adoption but very shallow adoption. Every company
4:29
Unidentified speaker (speaker_0): has deployed Copilot, but that doesn't result in like wide-scale automation or changes in how anybody does things. So it looks much faster superficially than it really is. I think the other fundamental way that this is different from the dot-com bubble or indeed other things like the growth of mobile. And remember, mobile didn't come with a bubble. Or the mobile
4:49
Unidentified speaker (speaker_0): bubble was in 2000 and then mobile. actually happened in 2010.
4:53
Unidentified speaker (speaker_0): But the other way that this is different from like PCs or the internet or mobile is that with those, we didn't know exactly how they were going to work, but we knew what the physical limits were. Whereas with this, we don't actually understand the science of AI, really. We don't know why these models produce such good results. So there's a sort of un
5:12
Unidentified speaker (speaker_0): unfundamental fundamental uncertainty
5:15
Unidentified speaker (speaker_0): in how this will evolve that wasn't the case with previous platform shifts.
5:21
Unidentified speaker (speaker_1): I think that moves us on nicely to talk about the technology.
5:26
Unidentified speaker (speaker_1): And I don't know whether you're re-quoting someone, but you quoted that a Frontier LLM is the most expensive and fastest depreciating asset ever created, which would call into question the economics.
5:48
Unidentified speaker (speaker_1): Trevor
5:48
Unidentified speaker (UNKNOWN): Burrus, Jr.:
5:48
Unidentified speaker (speaker_0): Yeah, I
5:48
Unidentified speaker (UNKNOWN): mean
5:48
Unidentified speaker (speaker_0): this is another of these. Is it the same or is it different? And there's a certain kind of person that loves to talk about fiber here.
5:54
Unidentified speaker (UNKNOWN): Um
5:55
Unidentified speaker (speaker_0): where we had obviously we had this huge overdeployment Deployment of fiber in the dot dot-com com bubble. Um it's not self-evident, it's not it's not like that in the sense of that that was built out way ahead of demand, whereas this is being built out way behind demand, which is really kind of important
6:08
Unidentified speaker (UNKNOWN): to understand. Trevor Burrus, Jr.:
6:09
Unidentified speaker (speaker_1): Did we know that ex ante? Trevor
6:11
Unidentified speaker (UNKNOWN): Burrus, Jr.:
6:11
Unidentified speaker (speaker_0): Yes.
6:14
Unidentified speaker (speaker_0): There were these claims about how fast the demand was growing that turned out not to be really true.
6:20
Unidentified speaker (speaker_0): But I think the but clearly today we have massively more demand than we have supply. I think the
6:27
Unidentified speaker (speaker_0): the challenge here is um
6:31
Unidentified speaker (speaker_0): like it's kind of I mean i th the advantage of the fiber comparison is you've got so many unknowns in trying to forecast this. So you can kind of try and model out, okay, this is how much data center is going to get built in the next couple of years. Um how much is going to get lit up with power, what chips are going to go into the them,
6:51
Unidentified speaker (speaker_0): what the token per watt is for doing that. And therefore you can pick a number of how much kind of underlying fundamental capacity is going to expand. And you know, you can pick a number 50x, 100x, 200x, whatever the number is.
7:05
Unidentified speaker (speaker_0): Then you know that we've only really got product market fit for one thing, which is software development. And then there's got a then you have a few other things around the edges of that, but a much are much smaller. We do not have broad consumer product market fit. We do in the sense we don't have like a thing like Instagram that hundreds of millions of people are using a lot
7:25
Unidentified speaker (speaker_0): every day to do with LLMs. Um so we've got this explosive demand, which looks like it's at you know $50 or $100 billion run rate revenue just for the two model labs. So it's not unreasonable to say, well, yeah, you could have way, way, way more demand quite quickly if you had
7:45
Unidentified speaker (speaker_0): other things start working.
7:47
Unidentified speaker (speaker_0): And so you don't really know how much more capacity we're going to get. You don't really know how much more the demand is going to grow. It's plausible that you could have demand ahead of supply for quite a long time in investment terms, for like multiple years. going out
8:04
Unidentified speaker (speaker_0): outside of the framework of capital markets or the business dynamics.
8:11
Unidentified speaker (speaker_1): Forgive me, but I'm going to pin you against a wall.
8:16
Unidentified speaker (UNKNOWN): That
8:16
Unidentified speaker (speaker_0): wasn't the metaphor I first thought
8:18
Unidentified speaker (UNKNOWN): of.
8:19
Unidentified speaker (speaker_1): So we talk a lot about scaling laws. By that in frontier models, that the the the capability and performance of the model rises in proportion to the investment in the compute.
8:32
Unidentified speaker (speaker_1): Is there any physical or indeed practical limitation to those scaling laws?
8:39
Unidentified speaker (speaker_0): I can give you a really, really short answer. We don't know.
8:44
Unidentified speaker (speaker_1): But I'm pinning you against the wall.
8:46
Unidentified speaker (UNKNOWN): We
8:46
Unidentified speaker (speaker_0): don't know. I mean, this is the thing. So you can hypothesize some practical limits. So you can say, well, at a certain point we run out of training data. And that's kind of a jagged statement in that there are some fields where training data, there's much more training data, it's much easier to get where synthetic data works. There are other fields where it maybe doesn't.
9:06
Unidentified speaker (speaker_0): So like it's much easier to do synthetic data in coding than in writing, for example. Or in video, say, than in writing. Um Some of the approach in the last couple of years has been shifting from just more training data to having the model think longer, which is reasoning, and then having lots of models, which is a sort of agent-swarm thing that's happened in the last six months.
9:27
Unidentified speaker (speaker_0): So there are things that you can do that can expand the capability
9:32
Unidentified speaker (speaker_0): without more um without scaling law. all. So, I mean, for the sake of argument, if you were to say, well, presuming the model stopped had stopped in January, you could have still taken the model progress that had stopped in January and say, yes, but if we throw 10,000 of these things at that problem, we still get a much better result. So there's a bunch of different
9:52
Unidentified speaker (speaker_0): engineering and deployment ways that you can get better results and use more compute, even if the underlying scaling doesn't c keep keep carrying on. And there's a bunch of ways in ch which you can say, well, it looks like in that field maybe that isn't going to continue. But the underlying issue is that because you don't know how this works at a
10:12
Unidentified speaker (speaker_0): theoretical level, all of these conversations become hunts for metaphors, which is why the doomers start talking about, you know, imagine if there was a meteorite or imagine if this was a nuclear weapon, because you don't know. And the metaphor I've always kind of liked is that it's sort of like you're building the Apollo program when you don't have a theory of gravity and you don't actually have a theory of the chemistry of rockets either. Like, well, but you made
10:32
Unidentified speaker (speaker_0): it twice as big and it went three times higher. Okay, so what if we make it twice as big again?
10:40
Unidentified speaker (speaker_1): One of the great things about T-Row Price is we have a genuinely global
10:45
Unidentified speaker (speaker_1): research platform. So
10:49
Unidentified speaker (speaker_1): we're able to draw comparisons. between what's going on in China and what's going on
10:56
Unidentified speaker (speaker_1): in the US? I mean, there's an awful lot to talk about in Europe, unfortunately. But I'm interested to know what your interpretation of
11:05
Unidentified speaker (speaker_1): how this is all going to work. Do you think all these models are going to be fungible, or will the Chinese and the US ecosystem
11:16
Unidentified speaker (speaker_1): evolve separately?
11:20
Unidentified speaker (speaker_0): I think everything is still wide open and everything is to play for. You don't want to depend on one company. You don't want to depend on Dario or Sam deciding what you're going to be able to do. So every company I talk to is doing multiple sourcing. I think every product and use case is making a decision around, but do I need a frontier model? And if you are doing
11:40
Unidentified speaker (speaker_0): a particular point solution for fraud detection inside an apparel, an e-commerce apparel company, you're going to choose the right model for that, and you're going to look at the cost and the reward, and that might be open, it might be closed, it might be last year's anthropic model running on a NeoCloud, it might be a Chinese model, it might be both. I think you have a lot of
12:00
Unidentified speaker (speaker_0): fragmentation along along that that price performance curve. And then a conscious decision that we're not going to depend on one.
12:08
Unidentified speaker (speaker_0): Second answer is a pure geopolitics question, which is on the one hand, America is determined on decoupling, and on the other hand, it looks like China is determined on decoupling. And China doesn't want to have that dependency on American technology either. They don't want to risk being cut off.
12:24
Unidentified speaker (speaker_0): The lack of access to chips.
12:27
Unidentified speaker (speaker_0): at the same scale that the US labs have, and a much more kind of pragmatic, less winner-takes or less um driven to make AGI kind of culture means that the Chinese labs are not pushing for absolutely the best possible lab m result every day. Instead they're pushing for, you could
12:47
Unidentified speaker (speaker_0): argue what they're actually optimizing for is let's be among a month behind the frontier and a tenth of the cost. But clearly we have a kind of a decoupling dynamic there that points towards multiple models, multiple choices for quite a long time.
13:09
Unidentified speaker (speaker_1): Positioning and asset allocation, this is
13:12
Unidentified speaker (speaker_1): the topic that we cannot ignore.
13:15
Unidentified speaker (speaker_1): Um
13:17
Unidentified speaker (speaker_1): what fascinates me about this capital cycle is how commonplace trillions has become, whether it's trillions of capital employed, whether it's pre-mark pre-IPO valuations, whether it's the number of trillion-dollar market cap companies we have. I mean, trillions has become common. So you've got a a a very
13:37
Unidentified speaker (speaker_1): visual metaphor that I
13:39
Unidentified speaker (speaker_1): really like, which you've described the current cycles like a pig in a python, i.e., we we are s in the process of swallowing this enormous amount of capital, but actually digesting it and deploying it is going to be a little bit more difficult. Is it possible to deploy
13:59
Unidentified speaker (speaker_1): capital at this scale?
14:02
Unidentified speaker (UNKNOWN): Um
14:03
Unidentified speaker (speaker_0): this quickly, clearly it's difficult.
14:06
Unidentified speaker (UNKNOWN): Um
14:06
Unidentified speaker (speaker_0): the tech industry, I mean, from memory, I think the big four platform companies were spending like $100 billion a year on CapEx as we came through the pandemic. This year, their guidance is something over $700 billion this year, and that probably numbers kind of crept up in the course of the year. And if you add Oracle and NeoClouds in China,
14:26
Unidentified speaker (speaker_0): you get close to a trillion. And then most people think that's the number's going to go up next
14:30
Unidentified speaker (UNKNOWN): year.
14:31
Unidentified speaker (speaker_0): So you have this enormous surge in the semis industry and a pricing crunch in the semis industry. So the revenue's gone up a lot more than the volume in semis. The semis industry has always been cyclical, so they were always super nervous about increasing their capacity. And like it's really only this year, I think, that they've really started increasing capacity. CapEx
14:51
Unidentified speaker (speaker_0): to fill this.
14:54
Unidentified speaker (speaker_0): And then you have challenges of every bit of the supply chain around data centers, down to like obscure power electrical equipment that none of us had ever heard of. And there are all sorts of companies suddenly spiking because they're getting,
15:08
Unidentified speaker (UNKNOWN): you know,
15:08
Unidentified speaker (speaker_0): the business has suddenly doubled to build all of this stuff down to things like gas turbines, where, you know, from memory, the whole industry does like 75 to 100 units a year of this of these kinds of things. And suddenly they're being asked to double production and they're like, yeah, it doesn't really work like that.
15:25
Unidentified speaker (UNKNOWN): Um
15:26
Unidentified speaker (speaker_0): and then of course you've got the whole political issue, particularly in the US, which some of which is actually about data centers, some of which I think is not actually about data centers, it's about other things. It's broader border political issues. Um So short answer is yeah, yeah, yeah, it's hard. to deploy this stuff. And of course, power, more than anything else. It's hard to actually get enough power into these things this quickly.
15:45
Unidentified speaker (speaker_0): That gets me to the the the conversation we're having earlier, that like, you know,
15:50
Unidentified speaker (UNKNOWN): uh
15:50
Unidentified speaker (speaker_0): it's easy to say shortages are followed by glutes. Um it's important to remember how narrow the adoption is and how much more it could go up quite quickly.
16:01
Unidentified speaker (speaker_1): To my mind, there are three ways.
16:03
Unidentified speaker (speaker_1): this capital cycle could come to an end.
16:07
Unidentified speaker (speaker_1): One is Nash Equilibrium,
16:11
Unidentified speaker (UNKNOWN): the
16:11
Unidentified speaker (speaker_1): ending of the Nash Equilibrium, where everybody to date, their optimal strategy is to invest as capac as they can. And it's in nobody's interest to to break that.
16:24
Unidentified speaker (speaker_0): Prison's dilemma.
16:25
Unidentified speaker (speaker_1): Prison's dilemma what w of which is all part of game theory. Two it could be regulation, whether it's on the built- physical build of of data centers or on the regulation of
16:37
Unidentified speaker (speaker_1): or it could be capital markets. Just capital markets go on strike. What w You mean
16:44
Unidentified speaker (speaker_1): in investment in markets
16:46
Unidentified speaker (UNKNOWN): we're we would dealing
16:47
Unidentified speaker (speaker_1): in probabilities working?
16:53
Unidentified speaker (speaker_0): I'm not a politics analyst by any stretch, but it's it's hard for me to see um consensus shifting. Like the the US political system is is unlikely to do that quickly.
17:06
Unidentified speaker (speaker_0): So the first is like does something happen that we phrase this a different way. Does something happen internal to the tech industry that results in the investment slowing down? And secondly, does something happen external? to the tech industry? What's happened in the last, certainly this year, arguably, like course of last year, is that the capital outgrew the cash flows of the hyperscalers. So certainly
17:26
Unidentified speaker (speaker_0): in 2024, this was coming out of cash flow because these were very cash-generative companies. It's now reached Reach a the point that I think Meta and Microsoft and Alphabet. are all at or above 50% cap ace to sales this year and probably higher next year. Amazon is a slightly different story because
17:46
Unidentified speaker (speaker_0): it's a much low margin business. So they've outgrown it. And so now they are dependent on you. Um now they are dependent on the capacity of the cap and willingness of the capital markets to continue providing that in all the ways that that means, and that's debt and private equity and public equities and all sorts. of other stuff. So that's one piece.
18:06
Unidentified speaker (speaker_0): And I'm not going to have an opinion on like what's the capacity of the capital markets to absorb this. That's for other people to talk about. I think the other side of this is
18:18
Unidentified speaker (speaker_0): do we get a paper tomorrow that says, hey, guess what? I get five A's X are results for a tenth of the compute. Seems unlikely. But given we don't have a good theory. theoretical model of how this works, it's possible. What happened, one of the things that's happened this year is a sort of agentic is is Agentic
18:37
Unidentified speaker (speaker_0): and B a sort of agentic swarm, if I can make that distinction, which we we talked about earlier. In that what happened with like maths mass and what's happened with cyber is you just take 10,000 of these things and throw them at the problem.
18:54
Unidentified speaker (speaker_0): And so you have an ability to use way, way, way more compute and get better results. It's kind of back to our scaling point. Maybe the models don't get better, but now you throw 500 models or 5,000 models or 50,000 of the same model at that problem, and that gets you better results. So there's also all
19:14
Unidentified speaker (speaker_0): sorts of questions around this: of like, well, what happens to the demand? And the reason I say this is it's perfectly possible that we get to a point that actually we can absorb way more demand without needing as much capacity as we thought.
19:29
Unidentified speaker (UNKNOWN): I
19:29
Unidentified speaker (speaker_0): mean, one of the things we haven't talked about here is that the marginal cost problem shapes consumer adoption. In that what we're sort of used to from the internet is that you could make it free, go viral, get a bunch, get tens of millions of users, and then work
19:49
Unidentified speaker (speaker_0): out the revenue because the cost was kind of could be absorbed within the funding structure of traditional venture capital. That whole playbook of make it free because marginal cost is so low, and then work out the revenue. later doesn't work with AI in the same way.
20:06
Unidentified speaker (UNKNOWN): Because marginal costs
20:08
Unidentified speaker (speaker_1): are very high.
20:09
Unidentified speaker (speaker_0): Because marginal cost is high. And so this is what you see at um OpenAI is and this is part of the thing that shapes public perception of this stuff, which is that if you're not paying, you're not seeing today's technology, you're seeing the technology from a while ago because they can't afford to give you unlimited access to today's debt. technology for free. And so
20:29
Unidentified speaker (speaker_0): what is the number like? Something like 5% of ChatGPT users are paying. So most people are using a relatively limited model, not very much.
20:40
Unidentified speaker (speaker_0): But this is kind of the the the the the the thing I keep coming back to is we don't have mass consumer use to drive demand at like to really, really drive demand for tokens. Because almost is and it is almost kind of circular. Like we don't we couldn't afford the industry couldn't afford to provide that for free.
20:58
Unidentified speaker (speaker_0): But whatever the the whatever you want to say about it, clear clearly right now we don't have something where user consum where hundreds of millions of consumers are using this a lot.
21:07
Unidentified speaker (speaker_1): Okay, well let's le uh in in which case let's move this on from straight economics to I I'd I like would to get a win window into Benedict Evans' mind as a dreamer.
21:20
Unidentified speaker (speaker_1): What are you most excited about here? A. B
21:25
Unidentified speaker (speaker_1): you know, we focus a lot of attention on the supply side and less on the demand side. And you know, what are the s what are the consumer applications or the services
21:37
Unidentified speaker (speaker_1): that you think
21:39
Unidentified speaker (UNKNOWN): are
21:39
Unidentified speaker (speaker_1): going to be the ultimate drivers of of of demand here?
21:45
Unidentified speaker (speaker_0): So
21:46
Unidentified speaker (speaker_0): and we talked about a little bit about this before. I think maybe sort of two answers to this question. One of them is: I was uh working at a mobile operator in the mid-2000s, and there was kind of a joke that you would go to a conference. and people would say, Oh, you'll be walking past a Starbucks and your phone
22:06
Unidentified speaker (speaker_0): will know, and you'll get sent a coupon by location services, and you'll get sent a coupon for a free coffee, and this became a joke. And I've actually got some research reports from that period that had like forecasts of how much money mobile operators would make, charging you every time you did a location lookup. And meanwhile, nobody noticed taxis.
22:26
Unidentified speaker (speaker_0): And we had everybody, everybody had a phone with a camera. And it didn't really occur to anyone that that was a video camera as well. And it didn't occur to anyone that people would post those pictures publicly because social media kind of wasn't a thing. And so, like, you don't know what the use cases are going to be. I mean, this is
22:46
Unidentified speaker (speaker_0): a kind of tech version of the lump love of labor fallacy. Like, you don't know what the new things will be. Human needs are infinite. So people will create a whole bunch of new stuff doing things that you couldn't do before LLMs, just as we had with mobile, just as we had with the internet, with smartphones. So that's one answer. Unhelpful answer, but like it's a deterministic observation. There'll be a whole bunch of new stuff that looks dumb that we'll all
23:06
Unidentified speaker (speaker_0): love that will be created with this technology. Second answer is the things that I try and think about are where does an LLM, what are problems that we can see now that we think an LLM can solve for a consumer? And I have a sort of a like maybe four bullet points from firstly,
23:26
Unidentified speaker (speaker_0): like automation of boring stuff, difficult stuff, stuff you can't be bothered with, which is our conversation about bank accounts. Secondly, stuff that is inertia, bank accounts, I'm on the wrong hand. health insurance plan, I'm on the wrong, I should change this, I should change that, but I can't be bothered. Is that
23:46
Unidentified speaker (speaker_0): done by the LL agent? Is that done by a dedicated product? Is it somewhere else? How does that work? Thirdly, like
23:54
Unidentified speaker (speaker_0): friction, where it's like friction and complexity, like get me onto the right credit card. And I think that's kind of different from inertia. I think there's automation and you could argue about this, but there's like in automation and inertia and friction. Purchases, and I suppose what I'm getting at is like the things where you should do that but don't, and the things where it's not clear what you
24:14
Unidentified speaker (speaker_0): should do. I think that's a useful distinction. And so maybe everyone will be you will be on a different credit card if you had an agent that could help you work out the right one to be on. And then the other extreme is my Instagram coach coat example.
24:27
Unidentified speaker (speaker_0): It's there is so much product. There are so many things you might like, how would you know? And there we now have systems that might be a much better way of showing you the things that you might like. You know, you didn't know that you would be interested in that book. You didn't know that you would really like that chair, that lamp, that cushion, that coat, that scarves, that pair of
24:47
Unidentified speaker (speaker_0): shoes. Um Now that systems will show this to you. And the conversation in Can Lion every year for the last three years has been, you know, I ask ChatGPT what X, Y, Z should I buy? What will it tell me? And the guys at Google like well, sorry, the people at Google put their head in their hands and say it's just SEO.
25:05
Unidentified speaker (speaker_0): Same answer. Like what you know, that's an SEO conversation. Much more interesting to me is the my previous point is it might suggest different kinds of things, like the shape of the distribution curve, the distribution of what gets bought might look different. The curve might get steeper or it might get shallower, uh, might get flatter in different product
25:25
Unidentified speaker (speaker_0): categories. Um and so um just as we had like the wave of D to C um on the internet, consumer internet, we might have like different kinds of things that are now possible to sell, possible to go to market, possible to get people to buy. because agents can walk you through it, or AI can AI systems can can surface those things and suggest those things to you.
25:45
Unidentified speaker (UNKNOWN): I
25:45
Unidentified speaker (speaker_1): think we've got a fantastic window into your imagination of possibilities here. But here's me gonna push you up against the wall again. So you're back, you're you're back as a partner at a venture capital firm
26:00
Unidentified speaker (speaker_1): and
26:03
Unidentified speaker (UNKNOWN): I
26:03
Unidentified speaker (speaker_1): want to know what's what's going to be in your picnic. Where are you going to lay the bets? Where where do you think the value is going to accrue in all this?
26:12
Unidentified speaker (speaker_0): So before I go to that, I just I made a note to make this point, which is used word imagination. I had a conversation with a CEO of a software company on Monday, and he said that he thought there was a s a shift from information asymmetry to imagination imagination imagination asymmetry. asymmetry. And what he meant by that was you have to have the imagination to know what to ask.
26:32
Unidentified speaker (speaker_0): And that's a problem for a model lab because you're asking the users to have the imagination to think of what to ask. And it's an advantage for an entrepreneur because the entrepreneur is having the imagination to think: oh, this is a thing you could solve with an LL. LLM and I'm going to build a
26:52
Unidentified speaker (speaker_0): product around that.
26:53
Unidentified speaker (UNKNOWN): And
26:53
Unidentified speaker (speaker_0): I'm going to show people and I'm going to build the tooling around that. This is a thing that you can do. It's also a, you know, it's a problem for you. Like, okay, the LLM can now analyze all this stuff in a different way. Great. What do I ask for? Other than like, what's giving me what what stocks are going to gonna go up? Which seems a little bit too easy.
27:12
Unidentified speaker (speaker_1): Well, it's interesting because Dan Gardner, who wrote a very popular book called Super Forecasting, the point he made on one of our previous episodes of The Angle is that in a world of abundant answers, which is the world of AI, having the right questions is gonna be key. And I think you've made the just
27:32
Unidentified speaker (speaker_1): made the same point in a in a in another way. But let's I wanna want to get back to the point. What's gonna going to be in your V VC C picnic?
27:38
Unidentified speaker (speaker_0): So
27:39
Unidentified speaker (speaker_0): maybe a couple of
27:41
Unidentified speaker (speaker_0): levers here. One of them is
27:46
Unidentified speaker (speaker_0): where is AI getting at some fundamental point of leverage in that industry? Is it unlocking something? And you can kind of back test this against the internet. Like there were some places where the internet didn't change very much. And if you'd said, you know, we're going to be an internet. internet-enabled law firm, or we're going to
28:06
Unidentified speaker (speaker_0): be a PC-enabled accounting firm, that wasn't I mean, that was maybe an opportunity to gain market share. And if you were the first of your peers to deploy spreadsheets, or your execution on that was better, then you could take share. But that didn't fundamentally change the nature of law firms or the nature of accounting firms. It's like saying if you were the first airline to do
28:26
Unidentified speaker (speaker_0): online booking, you probably gained a bit of share that year. Twenty-five years later, who knows who was the first person to join to do airline online airline booking online flight booking, who cares? Um
28:37
Unidentified speaker (UNKNOWN): So
28:38
Unidentified speaker (speaker_0): the question is, does the internet get to some fundamental point of leverage? Does AI get to some fundamental point of leverage that changes how you can do this whole thing? And you can pull the whole thing inside out and do it in a completely different way? This is the thing, you know, that and in if you go and look at, you know, I mentioned earlier, the typical big company today uses four to five hundred SAS apps.
28:58
Unidentified speaker (speaker_0): And each of those are some entrepreneur realizing that there was this problem inside the compliance department at uh at long-only funds. There is this thing that HR departments at universities do.
29:15
Unidentified speaker (speaker_0): Somebody has to think of that and work out this is a problem and this is how we're going to do it and this is how we're going to solve it and this is how we're going to get everybody to use it. And that's 500 SaaS apps inside a big company today. And so it's one at a time. You know, is that and and this is like the kind of the sort of fundamental venture problem, which is very different from like being a long-only
29:35
Unidentified speaker (speaker_0): fund, is it's not what are all the ways that this could go wrong. It's you suspend disbelief and you say,
29:43
Unidentified speaker (speaker_0): could this work? If it did work, what would that be? Is this a team that can make that work?
29:51
Unidentified speaker (speaker_0): And every VC now isn't looking at anything that isn't doing AI, but then you have to ask all these questions.
29:59
Unidentified speaker (speaker_0): And that's industry by industry, team by team, product by product.
30:03
Unidentified speaker (speaker_0): Is this have they identified some problem that is really big and important? And could it work? Have they worked out the right approach? Are they going to be able to work out the right approach? Because that's really what you're betting on.
30:16
Unidentified speaker (speaker_1): This is intended to be an instructional series
30:22
Unidentified speaker (speaker_1): with some element of practicality. So we're going to close this.
30:28
Unidentified speaker (speaker_1): And I would like your view on one skill that investors should learn in the age of AI
30:37
Unidentified speaker (speaker_1): and one habit they should unlearn.
30:40
Unidentified speaker (speaker_0): So there's a there's a sort of super practical point, which is try all the models all the time, throw questions at them.
30:49
Unidentified speaker (speaker_0): But I think a more
30:51
Unidentified speaker (speaker_0): useful thing is that imagination asymmetry. What are the completely different questions? What happens if you let go? And this is maybe something we should have spent more time on earlier. But like what happens with each wave of technology is that you shift the level of abstraction. And the
31:11
Unidentified speaker (speaker_0): questions that used to be important stop mattering because the new technology just does those things for you. And instead, there are new questions, new layers of abstraction that become possible that weren't possible before. And so
31:27
Unidentified speaker (speaker_0): imagine you're building a using a CRM. And classic CRM, you've got this very complicated data schema, you've got hundreds of different field types. And you are a great power user when you've constructed this elaborate query that lets you surface all these different things.
31:47
Unidentified speaker (speaker_0): And
31:48
Unidentified speaker (speaker_0): now you build an AI-native CRM and you build it as a system of record. And maybe what you do is you just drop every gob of data down on a timeline with a date and a client and no other field type.
32:02
Unidentified speaker (speaker_0): And like client call,
32:05
Unidentified speaker (speaker_0): app analytics, meeting notes, any piece of data that you have around that client is just dropped onto a timeline with the client name and a date and nothing else. And on one level, this is insane because now you can't run your 11 field SQL query that says, show me all clients where these kind of salespeople sales people
32:25
Unidentified speaker (speaker_0): were who are senior a salesperson. spoke to them and they're about to this, and they have this client value, and they have this usage data, and their contract is up at this period. And then that tells me who we need to make sure we call when we sit down in our Monday morning meeting. But you can, but it means you can also say,
32:45
Unidentified speaker (speaker_0): what are our most important clients who are likely to churn soon?
32:52
Unidentified speaker (speaker_0): Okay. And that's a real, like, I'd say a hard thing to let go of all the knobs and switches that you kind of can't ask that other kind of question. But now you can ask this new kind of question.
33:07
Unidentified speaker (UNKNOWN): And
33:08
Unidentified speaker (speaker_0): that was implicit It in a lot of the stuff I was saying about friction and Instagram and so on. And it's the imagination.
33:15
Unidentified speaker (speaker_0): asymmetry of thinking, wait, I could just ask that question. There's a concept in military tactics of like mission-based tactics, which is instead of saying to that lieutenant, you need to go to this grid square and you need to march to that bridge and secure that bridge. Instead, you say,
33:35
Unidentified speaker (speaker_0): okay, the objective of this entire division wide operation is to achieve this. And you have these resources, and this is the other people, go do it. And you empower your most junior people to go and you tell them what the mission is. And it's their job
33:55
Unidentified speaker (speaker_0): to be empowered to go and do that. And so here the point of MySQL analogy is is your objective of head of sales to know the right SQL query? Or is your objective, okay,
34:08
Unidentified speaker (speaker_0): which bits of our mission aren't working and which bits are, and how do we achieve
34:12
Unidentified speaker (UNKNOWN): that?
34:14
Unidentified speaker (speaker_0): And I think that's really hard.
34:16
Unidentified speaker (speaker_1): So I'm gonna I'm gonna read that by saying,
34:20
Unidentified speaker (UNKNOWN): you
34:21
Unidentified speaker (speaker_1): know, what you've got to give up is some of the metrics, some of the comfort metrics that you've been used to working with and really figure out what really matters in all this.
34:33
Unidentified speaker (speaker_0): Yeah. It's you know, what is the stuff that you do every day and you know how to do, and you know how to turn the handle and you know the knobs and switches, and then you know some obscure knobs and switches no one else knows around knows about, and you're comfortable that you've seen every single field and that you know
34:51
Unidentified speaker (speaker_0): you know that you have eyeballed every single meeting that meets the following criteria. And maybe it hasn't occurred to you that maybe there are other things that are important that where the answer to your yesno question is no, so it's not on the search result, but maybe you should have seen that one too.
35:08
Unidentified speaker (speaker_1): Okay.
35:09
Unidentified speaker (UNKNOWN): I'm
35:09
Unidentified speaker (speaker_1): gonna I'm I'm gonna say this conversation could be infinite. Our time is finite. Um at which point I'm gonna say Benedict Evans, thank you so much.
35:21
Unidentified speaker (speaker_2): Thank you.
35:22
Unidentified speaker (speaker_1): And thank you for listening to The Angle. We look forward to your company on future episodes. You can find more information about this and other topics on our website. Please rate and subscribe wherever you get your podcasts. The Angle. Better questions. better insights. Only from T. Row Price.
35:49
Unidentified speaker (speaker_3): This podcast episode was recorded in October of 2026 and is for general information and educational purposes only. Outside of the United States and Australia, it is for investment professional use only. It is not intended to be used by persons and in jurisdictions which prohibit or restrict distribution. of the material herein. This podcast does not give any advice or recommendation of any nature, or constitute an offer or solicitation to
36:09
Unidentified speaker (speaker_3): buy or sell any security in any jurisdiction. Prospective investors should seek independent legal, financial, and tax advice before making any investment decision. Past performance is not a guarantee or a reliable indicator of future performance. All investments are subject to risk, including the possible loss of principal. Discussions relating to specific securities are for informational and illustrative purposes only. are not recommendations, and the securities may or may not
36:29
Unidentified speaker (speaker_3): have been held in any T Row Price portfolio. There should be no assumption that the securities were or will be profitable. T Row Price is not affiliated with or a subsidiary of any company discussed. However, T Row Price portfolios may be passively invested in the company. The views contained herein are of the speakers as of the date of the recording and are subject to change without notice. These views may differ from those of other T Row Price associates and/or and or affiliates.
36:49
Unidentified speaker (speaker_3): Information provided by Thurston. third party speakers has not been verified by T Row Price. Information is from sources believed to be reliable, but is not guaranteed to be accurate, complete, or current. T Row Price Associates Incorporated, TERPA's research platform is global. T Row Price Investment Management's TERPIM is not. TERPA and TERPIM are separate investment advisor entities and do not collaborate on research.
37:09
Unidentified speaker (speaker_3): This podcast is copyright by T Row Price, twenty twenty six.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights