Advanced Search
Deutsche Bank Research Institute
Bringing the world to Europe and Europe to the world
  1. null
  2. Home
Macro | October 22, 2026
Chart of The Day: A $1.5tn Muse?
Chart of The Day
US Equities: Tech, Earnings and Midterms
October 22, 2026

Executive Summary

In this episode of Rate Check, host Shreyas Gopal (FX Strategist) and co-host Henry Allen (Macro Strategist) are joined by Parag Thatte from the US Equity Strategy team. The discussion covers the persistent market pressures in Europe, the latest dynamics in the energy sector, and a deep dive into US equity resilience, Q3 earnings expectations, and the relationship between interest rates and equity valuations. Additionally, the team previews a packed global calendar featuring crucial central bank decisions, the UK budget, and the US midterm elections.

Cleaned & Structured Transcript

Section 1: Introduction and Macro Market Drivers

[00:00] [Voiceover]: Rate Check, where macro meets markets, a podcast from Deutsche Bank Research.

[Shreyas Gopal]: Hello and welcome to Rate Check, a podcast from Deutsche Bank Research. I'm Shreyas Gopal, FX strategist, and I'm joined as always by my co-host and macro strategist, Henry Allen. Hey, Henry.

[Henry Allen]: Hello, Shreyas.

[Shreyas Gopal]: So Henry, we'll dive right in this week. When we recorded a fortnight ago, we discussed in detail the drivers of markets being energy and France in particular, the OAT boom spread. Are those two still the main themes in markets or what else has sort of come to the forefront over the past couple of weeks?

[Henry Allen]: Well, those two theme trends have very much still been in the spotlight. We've still got market concerns around France. And in recent days, we've seen the OAT Bund 10-year spread move above 150 basis points, which has added to the concern on the fiscal side. And we've actually seen other European spreads like Italy's and even Spain's to some extent widen as well. And then as you say on the energy side, we've also had ongoing concerns because in the last few sessions since we did this podcast last, Brent crude oil has actually gone back above $110 a barrel again to its highest level since May. So those inflation concerns are already swirling. That's got some more momentum. And then on top of those two things, we've had kind of a third concern creep in the last week, which is a fresh round of concerns about AI with some of the most exposed names, particularly among chip stocks coming under pressure last week. So that has had an influence on bond yields as well as concerns around some kind of AI slowdown have gathered pace again. There's been question marks as to how aggressively some of the big central banks can keep hiking rates. So kind of three big risks that have very much been on the horizon lately in terms of the agenda. Yet, despite all those growing risks, one thing that's actually held up remarkably well and ties in well with our guest today is US equities. At the time of recording, the S&P yesterday closed just over 1% beneath its record high. So it gives me great pleasure to welcome Parag Thatte to our podcast, who is in the US equity strategy team based in New York. Hello, Parag.

[Parag Thatte]: Hi Henry. Hi Shreyas. Thanks for having me on the podcast.

Section 2: US Q3 Earnings Outlook and Tactical Tech Rotation

[03:00]

[Henry Allen]: So Parag, we're recording this in the midst of earnings season. What are you guys expecting and what should we be looking out for?

[Parag Thatte]: So just to back up the tape a little bit, in the second quarter, we had blowout earnings with growth of 34% year over year. And that kind of growth is almost unprecedented, except when we are coming out of recessions. and big downturns. And so we haven't been anywhere close to a recession in recent years. and so the question that everyone's been having is if that kind of blockbuster growth continued into Q3, or are we going to see a deceleration? And before we can look at Q3, it sort of helps to look at what drove the very strong earnings growth in Q2 in the 1st place. And we see three drivers of that very, very strong growth. First, of course, is that we had an acceleration in AI demand-led growth. And all indicators suggest that there has been no slowing on the AI demand front, at least in Q3 so far. We also had a pickup in manufacturing and the other cyclical parts of the economy and everything again there points to a continued pickup. And lastly, we had a jump in oil and commodity prices. which helps earnings for energy and materials. And there hasn't been any let up on that front either. So when we consider all the three drivers that we think drove the off the charts earnings growth in Q2, they all continued into Q3. And so we are expecting another quarter of extremely strong growth. So our forecast is that we will get another quarter of 34% growth year over year. And the consensus is at around 27% coming into this earnings season. So we would expect beats of about 5 to 6 percentage points this earnings season, which is just a little bit about the historical norm for earnings seasons. It's still early in the season. but with the results we have had so far, we are comfortable with the forecast.

[Shreyas Gopal]: Great. So record earnings or record earnings growth continuing at a slightly above historical average pace, which is quite remarkable, I guess, as you mentioned, given the pace that we saw in Q2. One of the drivers you mentioned there was, of course, AI demand. And I noted that alongside this expectation of strong earnings growth, You're also quite presently a couple of weeks ago calling for a rotation away from US tech. And we're macro guys, we're more on the fixed income side, but at a very simple high level, one of the simple charts one might look at is just how tech has performed or the NASDAQ versus the main S&P index. And you have had a bigger drawdown over the past couple of weeks in tech names. Is this just about a rotation which you'd been sort of penciling in or are what you're reading in the press around, say, neo clouds and AI CapEx, is that giving a bit more pause for thought into your end?

[Parag Thatte]: Our call to basically downgrade tech a couple of weeks ago was purely tactical. And since we don't think the uptrend in performance of tech versus everybody else which has been in place for the last decade or so. We don't think that itself is broken. We think that is still intact. But tactically, we just saw there were sort of three asymmetries that made us move our call from being overweight to neutral. The first sort of asymmetry was just on performance after tech had already outperformed over the previous two months. The second was in terms of investor positioning. Tech was the only sector where investors seem clearly overweight. Everywhere else, they were either neutral or underweight relative to history. And so just that asymmetry in performance as well as positioning. And the last point I would say was in the asymmetry in terms of the perception around growth. And I say perception because we expect Q3 earnings for tech to be spectacular again. We expect earnings growth of about 55%. But the broad view and persistent question that we keep getting from investors is that there is no growth outside of AI. Pretty good. If you completely excluded the AI beneficiary companies, growth for the rest of the S&P 500 is still around 20, 21%. The median company in Q3, we would expect to grow at in the mid-teens, somewhere around 14, 15%. So there is very broad base growth outside of AI demand. And the perception being no growth outside of AI, we think created this asymmetry where if the other companies do deliver as we expect, we think there would be rotation out of tech into other places. And of course, when you are getting these rotations into tech, out of tech, all sorts of narratives take hold because the price action has just gone too far and it's reversing. The price performance is what is driving the narratives. And of course, there are concerns around how the AI, the CapEx is going to be financed and who's going to do it and what the returns are and so on. But those narratives sort of take life of their own when tech is underperforming.

Section 3: Midterm Elections, S&P 500 Targets, and Rate Resilience

[08:00]

[Henry Allen]: Thanks Parag. So earnings season is obviously one event we've got going on right now, but the other event coming up is the US midterm elections. What should we be looking out for there?

[Parag Thatte]: On the midterm elections, there are two regularities that we've seen historically. On the political side, the president's party has usually lost seats in the House as well as the Senate. This fact is now very well known. For instance, in the House, the president's party tends to lose about 28 seats on average. And current forecasters and projections put the Democrats on track to gain about 20 seats or so in the vicinity of the historical average. On the market side, 21 of the last 23 midterm elections have seen the markets rally in the Q4 of the year on average by about 7%. So the historical risk reward is pretty solid for equities going into a midterm. This time around, the market may just want to wait for getting more clarity on a few policies that might change once the midterms are out of the way. And so the usual midterm rally might just get pushed out. In the end, the earnings trajectory is the anchor or the magnet for equities, and eventually equities do catch up with what is happening in earnings. So that is where we focus on, but the midterms might just create noise around it.

[Shreyas Gopal]: That's super interesting. Thanks. So to round off maybe the sort of year end discussion, how we're viewing equities into year end, if I sort of summarize what we have so far, we're in the middle of what we think is another very strong earnings season where perhaps the non-AI story is a little underappreciated by the market and where there's room for some solid earnings growth there and the fundamental AI story that's still very much on track. Combine that with sort of the fundamentals returning to the market's focus probably after the midterms? Where, if I put you on the spot, where is the target for the S&P? I guess the question that most of our listeners have on the top of their mind.

[Parag Thatte]: So we've had the target for the S&P 500 at 8,000 since last November. So we have not changed it since then. We are still very comfortable with it. But if anything, we think the risk is to the upside simply because the earnings trajectory is pointing to equity should be a lot higher than where we are right now. And if you look at the S&P 500 in the last four years, they have been rising in this very, very strong print channel of, it's almost 2 percentage points a month in terms of the return that we've been getting. And what we've seen over the last, let's say now, 15 months or so is we get shocks and the S&P 500 sells off below the channel. But then once the shock recedes or some catalyst recedes, we have bought back to back into that channel. And so if that does happen, we are talking about something like 8,500 by year. And that is not our target because of all the catalysts that are in place, which we might just not let the S&P rise that quickly. But the risk for that, therefore, is to the upside. And we have seen the sort of catch up rallies, S&P find it going from range to range to range with step ups in the middle, often happen around earning season. But as we just discussed, this earning season is a little bit different. We have the FOMC during the earning season, we have the midterm elections. And so there may be a little bit more noise, but earning season historically have been pretty good for equities to do these catch up rallies back into that trend channel.

[Henry Allen]: That's really interesting. Actually, just to pick you up on that, I think one question a lot of people have been asking us, and I know you'll be the expert on is Why have equities been so resilient these last four years? Because the latest question that people are asking is, even as bond yields have gone up to multi-year highs, equities have seemed to remain really resilient. What's your take on that? And not just the latest resilience in the face of bond yields, but that wider resilience over the last four years or so. What's been driving that?

[Parag Thatte]: So specifically on rates, I mean, we could discuss this a day and not be done with it. But I would say there are sort of three key points to consider here. The first is simply that just the premise underlying the question that high rates mean lower equities or at least lower equity multiples is the question that we often get. And in theory, it seems to make sense because the theory says that all else equal, if rates go up, the discount rate should be going up. And if the discount rate is going up, that is bad for equities and equity multiple. And on paper, this is completely uncontroversial. But the phrase all else equal does a lot of heavy lifting in this view of the world. In that view, if in the middle of the night for no reason, just in a vacuum, if rates did go up, then yes, equities should decline, or at least equity multiple should decline. But in reality, rates do not go up in a vacuum for no reason. And for example, if rates were moving up because growth expectations were moving up, then sure, the risk-free rate could go up, but at the same time, the risk premium on equities would be coming down. And so where that leaves equities is not very clear. And empirically, what we have seen historically is that movements in rates turn out to be a wash for equities as the risk premium seems to move nearly equally in the opposite direction. And so the first point I would make is that even on a theoretical basis, the premise that high rates should mean lower equity multiples does not necessarily hold up. The second point here is then that it really matters why rates are moving up. And in the present sort of the last few weeks, the commentary has all been on rates are moved up because of higher inflation and the Fed wanting to fight inflation. But most of the sort of empirical evidence points to growth being as much, if not a bigger factor driving rates higher. And if rates are moved up because growth expectations have moved up, then it's not a surprise that it's a positive for equities. And of course, the growth that matters in the end for equities is earnings growth. And that, as we just discussed, has been spectacular. So We think that fact helps understand what has explained why equities have done well with rates moving up. And lastly, I want to be clear that equities do react to movements in rates, just that they react to the volatility of rates and not so much to the level of rates. And so when rates volatility moves up, equities do sell off. But then as volatility dies down, equities rally back. And even if rates stay high, and as a reminder, for wallet to stay high, you need more shocks, more catalyst to keep coming. If you don't get that, rates would stop moving and wallet comes down and equities have usually reacted positively with that. We have seen that play out numerous times in the last five years. And even in the last few weeks, I would say that it's not correct that equities have not reacted. The index has been kind of in this very, very tight range. But if you look beneath the surface, a lot of stocks have reacted to the rates volatility. The average stock in the S&P 500 is down about, was down, it's come up a bit, but it was down about 6.5% from peak to the bottom just last week. And some pockets, some more cyclical pockets were down even more than that. So equities, as they have usually done, have reacted to the rates volatility, even though the index itself had been sort of holding up because of the very strong growth.

[Henry Allen]: Thanks a lot, Parag. That was really interesting. Certainly found out a lot about the US equity market I didn't know beforehand. So thank you.

[Parag Thatte]: Thanks, Henry. Thanks, Shreyas. Thanks for having me on the show.

Section 4: Upcoming Global Event Calendar

[15:00]

[Henry Allen]: So Shreyas, we are obviously in an eventful period in markets right now, but we've got plenty more events happening over the next couple of weeks. So what should we expect?

[Shreyas Gopal]: Yeah, I mean, we've already discussed a few of them. We're in the middle of earnings season. We have an FOMC meeting as well, which I'll come on to in a little bit. And of course, these midterm elections where geopolitical experts Helen who we've had previously on the podcast has noted how the foreign policy implications or the trajectory for foreign policy will be particularly in focus after this set of elections. Almost all G10 central banks also meet over the next couple of weeks, every single one with the exception of the Swiss National Bank. Amongst the major central banks, we're actually not expecting any rate increases at this meeting, at this set of meetings. So despite a pretty hot US inflation print, our economists expect the Fed to remain on hold until hiking in December instead. Similarly, amongst the ECB, our expectation is for the next hike to come in December. I imagine the focus in the press conference for Lagarde will be again around the situation with French shields and some of the mechanisms that the ECB have in place, as we discussed with you seen on the previous episode. And the Bank of Japan, likewise, having hiked at their last meeting in September, we're expecting the hike not to come this October meeting, albeit an important one with some fresh forecasts. Where we are expecting rate hikes is here in Europe, a couple of central banks, we do expect it to lift rates. smaller ones in Sweden to raise rates to 2%, their first of the cycle and the first of the cycle for the Bank of England as well, which may prove to be the bigger market event. We have two large UK events over the next couple of weeks, the budget next week where our preview is out and in general calling for a quieter affair than we've perhaps seen in the past couple of years. And then the Bank of England meeting in November as I discussed where markets are very much pricing in and we're also expecting the first hike of the cycle. And then very finally, last but not least, a technical event that bond markets in particular will be focused on is the Treasury's quarterly refunding announcement. We've talked a little bit about Treasury buybacks and the impact on US yields in previous episodes. And one of the feedbacks from our conference in New York that our colleagues in rates have written about, one of their feedbacks and takeaways is there is some reporting around potentially more forceful debt management actions, including issuance cuts on the long end, even potential cancellations of bonds. And so all these themes and questions around that will probably increase in frequency and something that markets will be focused on over the next couple of weeks.

[Henry Allen]: Thanks very much, Shreyas, for that rundown. It sounds like there's plenty to happen before our next episode. And thank you for listening to another episode of Rate Check. Look forward to joining you next time.

[Shreyas Gopal]: Thank you very much.

[Voiceover]: This podcast has been produced by Deutsche Bank and may contain research as defined in MiFID II. The information discussed is believed to be reliable and has been obtained from public sources believed to be reliable, although Deutsche Bank makes no representation as to its accuracy or completeness. Opinions, estimates and projections discussed constitute the current judgment of the speaker at the time of recording. They do not necessarily reflect the opinions of Deutsche Bank and are subject to change without notice. For further important information, please visit research.db.com. This is Rate Check, where macro meets markets, a podcast from Deutsche Bank Research. To ensure you never miss an episode, subscribe now.

Disclaimer

This transcript has been generated using artificial intelligence and may contain minor inaccuracies or omissions. For complete accuracy and context, please refer to the original podcast recording here.

Audio: US Equities: Tech, Earnings and Midterms
Publication Type:
Rate Check
AGI doom or boom?
October 21, 2026

Executive Summary

This document presents a structured and cleaned transcript of a panel discussion with Azeem Azhar, founder of Exponential View, and Adrian Cox, Thematic Strategist, Deutsche Bank Research, analyzing the timeline, economic impacts, and risks associated with Artificial General Intelligence (AGI).

They were discussing a report written jointly with Azhar as an expert collaborator for the Deutsche Bank Research Institute: “AI boom or doom: what will superintelligence feel like and when will we get there?”.

The discussion, moderated by Jim Reid, Global Head of Macro Research at Deutsche Bank and Head of the Deutsche Bank Research Institute, centers on the following key themes:

  • Recursive Self-Improvement (RSI) as an Investment Engine: RSI is identified as a critical driver for AGI, requiring a continuous feedback loop of massive scaling, capital investment, and widespread corporate adoption.
  • Diffusion and Domain-Specific Superintelligence: While superintelligence has already been achieved in specialized domains (e.g., Go and complex mathematics), broad economic diffusion faces significant friction due to real-world infrastructure constraints.
  • Hyperscaler Capital Expenditure (CapEx) and Return on Investment (ROI): The transition of GPU capacity from model training to enterprise inference provides a viable pathway to monetize massive CapEx, independent of achieving immediate AGI.
  • Labor Market Dynamics and Historical Precedents: The panel explores the risk of a near-term transition period marked by underemployment and a shift in income from labor to capital, contrasting it with historical technological revolutions (e.g., the automotive sector) where automation ultimately generated net job growth.
  • Regulatory Capture versus Genuine Safety Concerns: The call by industry leaders to implement safety guardrails is analyzed as a mix of genuine concern over model alignment ("Sorcerer's Apprentice" risk) and strategic regulatory capture.

Key Takeaways

  1. RSI is Central to the Investment Thesis: Recursive self-improvement (where AI accelerates its own development) is the foundation of the massive capital investments currently being made by major technology firms.
  2. Enterprise Inference Drives Near-Term Payback: Hyperscalers are shifting GPU capacity away from frontier model training and toward enterprise inference. This structural shift means investments can be amortized through commercial demand even if RSI does not yield immediate AGI.
  3. Real-World Friction Limits AI Velocity: While developers in AI laboratories experience rapid, verifiable progress, broader macroeconomic diffusion is bottlenecked by physical constraints such as power grid capacity and hardware supply.
  4. The Enterprise Learning Curve: Research indicates a strong learning effect in AI adoption. Companies experienced with AI show an 80% confidence level in predicting project success (compared to 50% for laggards), allowing CFOs to allocate capital more predictably.
  5. Transition Stress in the Job Market: Near-term indicators of AGI progression include a decline in entry-level and junior hiring. Unlike traditional industries, infrastructure like data centers represents a high concentration of capital that generates very few direct jobs.
  6. Safety as a Competitive Strategy: AI labs face a conflict between slowing down to resolve alignment risks and accelerating development to ensure they achieve AGI first, leading to strategic public calls for regulatory guardrails.

Cleaned & Structured Transcript

Section 1: The Path to AGI and Recursive Self-Improvement (RSI)

[00:00]

Adrian: Google DeepMind's Strategy Officer recently said that RSI—recursive self-improvement—is a key part of the investment thesis in what he called the biggest scientific bet civilization has ever made. Demis Hassabis, his boss, has always said that step one is to solve intelligence, and step two is to use it to solve everything else. How do you do that? To have Artificial General Intelligence (AGI), you need to have recursive self-improvement. To have recursive self-improvement, you need to have massive scaling. To have massive scaling, you need to have massive investment. And to get that investment, you need to have huge adoption. That is where we are reaching now, where adoption is absolutely key. It is something we will talk about in the panel discussion. You need to have companies that are prepared to invest in primitive steam engines as a way to fund the creation of the motor car for the future.

Section 2: Domain-Specific Superintelligence and Technical Diffusion

[01:09]

Jim: Azeem, there seems to be a plausible, but not inevitable, route to AGI by maybe 2030 or not far after that. Do you think we will get something that looks like AGI by around that date? How will we know, and what do we look for?

Azeem: One of the points Adrian made was that these models get much better in particular domains. It is very clear that in some domains we are already at superintelligence, whether it is Go, which I think we have had for a decade, or recently the extraordinary results in certain classes of mathematics. I spoke to the dean of a university mathematics department the day after those results were announced. She said to me, "It's devastating, but we have just got to move on now. This entire field has now been replaced by the capabilities of the machines."

One of the key questions about what we will feel in our everyday lives is the diffusion question that Adrian described. If you work in one of these AI labs in San Francisco, you experience a world where everything you do is verifiable—it is software code, it is testing, and you are running hundreds of AI agents with the latest models. You feel like you are moving at the speed of AGI, but the rest of us are not.

If the San Francisco consensus is right, however, the 2030 period will feel like there is a lot less friction in the day-to-day. Companies will be able to run far more efficiently and effectively because information is not being lost between a decision and its execution. We will have moved an awful lot of economic activity onto these data centers as a consequence. The reason I say "if" is because the global economy is huge, and four years is just not very long for it to move.

Jim: You still need physical infrastructure like power, energy, and clinical trials for drugs. Can you expand on that? What would make you more negative?

Azeem: To achieve RSI, what you need to continue to see is the measure of how much AI is speeding up AI research. There is a research group called METR which tries to measure this. Effectively, this is a second-derivative measure of the impact of AI on research. If that number keeps constant at 1.5 or starts to rise, we are on track for RSI and on track for that outcome. If that number starts to slow down because we are having diminishing marginal returns, we know that we are not on the path of the San Francisco consensus. There are another couple of things you can watch for.

Jim: Can you repeat what that measure is?

Azeem: The measure is how much AI is accelerating the rate of AI research.

Jim: And how do we measure that?

Azeem: METR has an index where they look at the declared amount of AI being used at each generation over time. But again, it is a lab test. It is a lot like Adrian's task horizon, which is also from METR, and is something happening in the laboratory rather than in the real world.

There are a couple of other things you would look for. What is the pattern of hiring within the AI labs themselves? If they are hiring fewer junior developers and more people who understand algorithms and distributed systems, they are preparing for that type of scale.

Furthermore, we must monitor the supply of compute. Currently, Graphics Processing Units (GPUs) remain in sharp under-supply because we have been slow to deal with enterprise demand for AI. If that under-supply continues into 2029 or 2030, that would be an indication that we are on an RSI path, because the AI models do not sleep and are insatiable for compute. The converse would be that if those things are not happening, perhaps we are not on the path for RSI; we would be on a path for accelerated AI improvement, but not the type defined within the San Francisco consensus.

Section 3: Capital Expenditure (CapEx) and Market Returns

[05:38]

Jim: It seems like the San Francisco consensus hyperscalers feel that RSI is necessary to justify the huge CapEx they are spending. Is that true? I am an AI convert. I think it is going to be a fantastic productivity-enhancing general-purpose technology. I just have a suspicion that, while the CapEx being spent now will make a fantastic product, the people spending that money will not actually make much money out of it, and we will get a bust on the way. Can you give me your view on that?

Azeem: That is a five-trillion-dollar question. The CapEx that is being spent right now and brought online is shifting heavily over towards inference. The ratio of those GPUs spent on training frontier models versus running inference for enterprises—like Deutsche Bank and member companies in the FTSE 100 or S&P 500—is shifting toward inference. In other words, you are seeing a traditional corporate structure where companies transition from development to deployment of their products in the market.

Consequently, I do not think the hyperscaler CapEx and the payback they receive is entirely dependent on RSI being achieved. Extraordinary returns might be dependent on it, but current enterprise demand is significant enough that, if it continued, it could pay for the capital being deployed. It might not be a brilliant return, but you certainly might not get fired.

However, roadblocks could happen along the way, such as power shortages. If you cannot get power to data centers, you cannot energize the chips. If you cannot energize the chips, you cannot make a return on them, nor can you train your AIs using RSI or any other method.

If we listen to the discussion around RSI, I am used to environments where corporate executives are driven by their customers, market expectations, and clear price signals. Sometimes when you hear the general discussion about the San Francisco consensus, it feels disconnected from a product we can sell to a Midwestern corporate client for an annual contract. That is a complicating factor for an investor because you are trying to understand the underlying motivation and whether it is a clear profit motive.

Section 4: Productivity Statistics and Enterprise Learning Curves

[09:04]

Jim: When should we start to see this in the macroeconomic statistics? The US economy is doing fairly well at the moment, but a lot of that is driven by CapEx spending rather than a huge leap in productivity. At what point should we start to see this in productivity stats?

Azeem: ChatGPT came out in November 2022, and it was a year later that Microsoft and Google made their corporate generative AI services widely available to enterprises. We are coming up to the three-year anniversary of this technology. It is probably too early to expect to see anything in aggregate statistics, but we are starting to see two or three different things.

First, executives are discussing quantifiable results from AI in public earnings transcripts. Approximately 33% of earnings transcripts in the last quarter reported a quantified AI outcome. The challenge is discerning how significant these outcomes are, and whether executives are forced to report them to protect their jobs due to the pressure to get AI right.

Second, while we hear that CFOs are underwhelmed with the immediate results, when asked if they are going to spend less on AI next year, the answer is resolutely that they are going to spend more.

I recently saw research from a consultancy that conducted an interesting survey of enterprises claiming to perform well with AI, seeing a few points of extra revenue and a couple of points taken off their costs. Those companies said they had an 80% confidence level in predicting whether a new AI initiative would pay off. For lagging companies, which are spending less and getting lower returns, their degree of confidence was only 50%.

This suggests a learning effect: the more an enterprise uses AI, the better it gets at deployment, and predictability improves from 50% to 80%. For a CFO, that is a much better position from which to judge corporate capital allocation on an AI project.

Regarding when we start to see this in macro statistics, if that learning effect is replicated across cohorts, companies will scale from one pilot to 10, and then to 100 over the course of a few years. We might start to see sector-by-sector impacts toward the tail end of next year, though it is still very small in aggregate.

Section 5: Labor Market Implications and Historical Parallels

[12:23]

Jim: What is the implication of this for the employment market? You mentioned in the report that a sign of nearing AGI is a slowdown in junior hiring. Is there a risk that productivity gains will trigger a severe job displacement crisis, or is it more nuanced than that?

Azeem: There is some evidence of this. A Stanford study titled Canaries in the Coal Mine showed that junior hires were down by about 19%. One of the issues with that study, however, is that we do not know how much of that trend was caused by AI versus executives using general economic uncertainty to slow down hiring. It is often easier to attribute headcount reduction to hard technology than to difficult business circumstances.

What you would expect to see—and there is some data in the UK suggesting this may be happening in larger companies—is a lower number of vacancies, particularly at junior levels.

In general, this is a challenging technology because capital expenditure typically scales alongside job creation. In the UK manufacturing sector, every £200,000 of capital equipment generally correlates with one job. In contrast, for a data center running £100 million of CapEx, you might see only four jobs. This represents an investment concentration that does not distribute employment throughout other parts of the economy.

I do not think we will see an abrupt cliff edge, because companies do not move that quickly and diffusion takes time, but we will certainly see pressure.

Jim: Looking out five to ten years, is there a risk that the economy performs well but we see structural unemployment rise to 6%, 8%, or 10%?

Azeem: We could see an environment where a significant portion of national income shifts from labor to capital, leading to elevated underemployment. The US already faces an underemployment problem where many graduates work in roles that do not require degrees; these individuals do not show up in traditional unemployment statistics. Absent policy intervention, this trend could intensify.

Jim: My background as an economic historian influences my view here. For 250 to 300 years since the Industrial Revolution, every major technological leap has triggered warnings of massive structural unemployment and labor destruction. Historically, the productivity gains from these technologies have generated sufficient aggregate wealth to create demand for new, different types of employment.

My favorite historical parallel is the automotive sector. In the early 20th century, as the US automotive sector was becoming a viable employer, Henry Ford decided to automate part of the Model T production line, replacing manual labor with machines. Despite initial public outrage, automation lowered production costs and made cars affordable to the mass market. Demand surged exponentially, requiring more factory workers even with automated processes in place.

Furthermore, the automobile catalyzed massive secondary employment sectors: suburban housing construction, national highway systems, tourism, and automotive mechanics. While we often lack the collective imagination to foresee future industries, productivity gains historically lead to labor market adjustment. However, in my discussions with global clients, I find myself in the minority with this optimistic view.

Azeem: I agree with that long-term economic mechanism. My concern regarding the 5-to-10-year horizon is the mismatch between the rapid pace of technological displacement and the speed at which the economy can construct new roles, which depends on our unlimited wants as described by Lionel Robbins.

We must consider whether these new jobs will align geographically, socially, and psychologically with the workers displaced from legacy roles. This is where active policy intervention becomes critical. I do not think jobs disappear entirely, but we could experience a highly disruptive transition period.

Additionally, looking at the fiscal capacity of many countries, debt levels are high and we face simultaneous demographic headwinds. This raises the question of whether there is sufficient political and fiscal capacity to manage this transition, or if it will be left entirely to market forces over a 10-to-15-year period.

Jim: During the Industrial Revolution, the "Engels' pause" occurred where productivity grew faster than wages, leading to social unrest because workers did not share in early gains. However, democratic representation was limited then, so there was little recourse. Today, with established democracies, rapid job displacement could drive populist political shifts. Could this political friction slow the development of AI, particularly in the US, or lead to a structurally different society?

Azeem: There is notable public resistance in the US. Survey data indicates that institutions like Immigration and Customs Enforcement (ICE) poll more favorably than AI. Industry leaders have struggled to communicate the long-term benefits of the technology, even though 50% of people in the US and UK use it daily.

However, historically, when businesses strongly demand a technology, political resistance rarely halts its rollout entirely—especially in decentralized market economies. Instead, political pushback is more likely to alter the social license to operate for infrastructure build-outs.

For example, in Lanarkshire, Scotland, a developer building a 400-megawatt data center agreed to a community benefit payment of £72,000 per megawatt to fund local training and skills development. Because hardware access is constrained, the payback period on operational data centers is highly compressed, enabling developers to absorb local community costs. Hyperscalers are increasingly offering guarantees regarding local water and energy pricing to maintain their social license.

Additionally, looking through an economic history lens, general-purpose technologies—such as the printing press, the steam engine, electricity, and the internal combustion engine—inevitably reconfigure society by dramatically lowering the cost of a key production input. I am certain we will see similar societal shifts with AI.

Section 6: AI Safety, Regulation, and Geopolitical Pressures

[22:27]

Jim: Recently, prominent AI leaders publicly petitioned to slow the pace of frontier AI development to establish safety guardrails. Was that a genuine humanitarian initiative, or was it a strategic move to protect their respective market positions?

Azeem: It is reminiscent of a scenario where a bully holds your hand and hits you with it, saying, "Stop punching yourself." They are actively pressing the accelerator while pleading for external intervention to slow them down.

There are complex motivations at play. AI leaders understand the emerging technical risks. Specifically, advanced models do not always align with human intent, or they execute instructions to an extreme degree. This is the "Sorcerer's Apprentice" dilemma, where processes scale beyond the operator's control. Developers want to address these alignment challenges, but doing so diverts resources away from the raw capabilities race required to achieve AGI primacy.

Thus, while there is a call for third-party oversight, there is also an element of strategic regulatory capture to establish barriers to entry. We will see a constant tension between the rapid pace of development in AI labs and our limited understanding of how to ensure these technologies behave safely under all conditions.

At the same time, figures like Jensen Huang of Nvidia and Brad Smith of Microsoft have publicly discussed the necessity of pacing controls and operational "kill switches." The political class is increasingly focusing on this issue. Ultimately, the products being built are highly compelling and customers want to buy them, but developers remain somewhat insulated from conventional risk management practices. I do not think it made any difference to what their revenues will look like because we want last year's technologies anyway. This discussion will now play out for the next six to nine months.

Adrian: I think you are absolutely right, and there is indeed a very strong cynical case to be made. At the same time, it is interesting that a lot of these organizations were founded with the idea of working for the public good, and that remains a strong intellectual strain within them. I think they genuinely believe they are acting for the good of humanity—perhaps not all of them, but there is a core group involved since the founding who believe that only they can be trusted to develop and implement this safely. They feel they must get there first because if they do not, a competitor will develop it in a far more dangerous way.

Analyst & Key Figure Registry

The following key figures and historical references were noted in this transcript:

  • Demis Hassabis: Chief Executive Officer, Google DeepMind.
  • Azeem: Azeem Azhar, technology analyst and creator of Exponential View.
  • Adrian: Adrian Cox, AI analyst at Deutsche Bank Research referenced by the panel regarding domain capabilities and task horizons.
  • Adam Smith: Classical economist referenced regarding business collaboration and public interest.
  • Lionel Robbins: Economist referenced regarding the definition of economics and unlimited human wants.
  • Jensen Huang: Chief Executive Officer, Nvidia.
  • Brad Smith: President, Microsoft.
  • Henry Ford: Founder of Ford Motor Company, referenced regarding automation and the Model T.

Disclaimer

This transcript has been generated using artificial intelligence and may contain minor inaccuracies or omissions. For complete accuracy and context, please refer to the original podcast recording here.

The author of the report wishes to acknowledge the expert advisory contributions made by Azeem Azhar of Exponential View.

Publication Type:
dbInSights
47.6.0