The Acquirers Podcast · · 1h 5m
AI Spending & Market Risk: Capex, Valuations & Returns | Dan Rasmussen
With Dan Rasmussen
ChorusBring the market's conversation to your AI. Learn more
Summary
Dan Rasmussen and Tobias Carlisle debate whether the AI boom is a bubble, focusing on capex-driven overbuild risk, the slow pace of real-world AI adoption, and the cyclical economics of chips. Dan argues AI revenue must reach roughly $600 billion a year to justify spending, sees no productivity gains yet, and warns that Nvidia's circular financing, Oracle's wide-trading bonds, and the depreciation burden of data centres could unwind abruptly. Tobias, while calling AI revolutionary, frames hyperscaler spending as equity- and debt-funded and compares it to the dotcom fibre overbuild, and he flags Microsoft Office and DocuSign as exposed to cheaper AI substitutes. On credit, Dan watches high-yield spreads (about 330bp, with CCC spreads above 1,300bp) and private credit stress, while Tobias sees spreads as coincident rather than leading. Dan closes with a constructive case for Korea and Japan, where family-control holding structures and governance reform create discounts that he expects to close over the next few years, and notes the US cyclically adjusted P/E of about 41.6 as a caution.
Key takeaways
- Tobias Carlisle on AI infrastructure & data centers: Hyperscaler AI spending is debt- and equity-funded, so it must show revenue soon or the boom collapses.
- Dan Rasmussen (bearish on Microsoft): AI agents like Claude and Cursor are displacing Microsoft Office as the daily productivity app.
- Dan Rasmussen (bearish on Oracle): Oracle's bonds trading near 9% on an A-rated issue signals stress in AI-linked funding.
- Dan Rasmussen (bearish on NVIDIA): Nvidia's circular financing deals look like a very high-risk decision that will blow up in a downturn.
- Tobias Carlisle (bearish on DocuSign): A DIY AI clone of DocuSign's signing tool exposes its pricing to substitution.
What was said, by company and theme
25 views from 2 speakers. Every quote is checked against the episode's transcript (how).
Microsoft
Across the shows: 47 speakers in other episodes, 51 bullish and 11 bearish calls, past 30 days
Where everyone stands on Microsoft
Dan Rasmussen · Verdad AdvisersBearish
AI agents like Claude and Cursor are displacing Microsoft Office as the daily productivity app.
Dan notes that his analysts now work with Claude Code or Cursor open rather than Excel, Word or the command line. A year ago that wasn't true. The day-to-day productivity app is shifting to Claude, Cursor or other agents. He says this seems most threatening to Microsoft Office, and that he is nervous about it, as people no longer use Excel or PowerPoint much. He says he does really worry about that kind of productivity software.
“in some sense, like, that seems most threatening to, Microsoft Office.”at 18:42
Oracle
Across the shows: 32 speakers in other episodes, 10 bullish and 30 bearish calls, past 30 days
Where everyone stands on Oracle
Dan Rasmussen · Verdad AdvisersBearish
Oracle's bonds trading near 9% on an A-rated issue signals stress in AI-linked funding.
Dan notes Oracle's bonds are trading very wide. He says an A-rated bond yielding about 9% is a little scary, especially since Oracle is a big funder of OpenAI. His firm is doing a deep dive on Oracle to understand what its bonds indicate about the health of the AI economy, because Oracle is a good case study.
“But a lot of things, Oracle's trading very wide and I think that's an A rated bond that's trading at like 9% on Oracle bonds. That's a little scary. And they're a big funder of OpenAI.”at 56:10
NVIDIA
Across the shows: 171 speakers in other episodes, 185 bullish and 160 bearish calls, past 30 days
Where everyone stands on NVIDIA
Dan Rasmussen · Verdad AdvisersBearish
Nvidia's circular financing deals look like a very high-risk decision that will blow up in a downturn.
Dan says Nvidia's circular financing is good on the way up and terrible on the way down. As long as the market holds, the arrangements look fine. But if the market turns, the structures will blow up in Nvidia's face, so he calls it a very high-risk decision. He adds that the risk sits in the finance rather than the technology.
“I mean, this stuff again, it's just like you described, good on the way up, terrible on the way down, right? Like all well and good, but gee, if the market takes a turn for that stuff, it's gonna blow up in their faces. It seems like a very high risk decision.”at 56:40
DocuSign
Tobias Carlisle · Acquirers Funds / The Acquirers PodcastBearish
A DIY AI clone of DocuSign's signing tool exposes its pricing to substitution.
Tobias describes a widely shared example. DocuSign quoted a 15-person firm about $8,700 a year to collect signatures. Someone then reverse-engineered a clone called Sign New Doc. The clone is not perfect and lacks some of DocuSign's security features. Still, Tobias sees a prisoner's dilemma: customers will defect to a cheaper alternative at some price level, and he thinks that price level is a pretty big number. He is hedged, saying it is not clear, but the pricing exposure looks real.
“DocuSign had quoted a 15 person firm like $8,700 a year so they could collect signatures. And so the guys reverse engineered it. He calls it Sign New Doc. And it's not perfect.”at 1:01:42
AI infrastructure & data centers
Where everyone stands on AI infrastructure & data centers
Dan Rasmussen · Verdad Advisers
Data center capex cycle will damage returns: firms overbuild, and Chinese low-cost rivals will erode economics.
Dan argues that AI infrastructure is a capital-intensive, cyclical business. Big tech is now more capital intensive than industrials. Each firm models its own data-center ROI without accounting for competitors doing the same math, so they overbuild and then prices collapse. High capex has historically led to low future returns because companies always overbuild and overestimate demand. He says the depreciation math on chip investments means delay works against them: AI investors need something good to happen soon, or depreciation curves become a ticking time bomb. Dan also sees a Chinese low-cost competitor, aggressively trying to win, turning compute into a commodity and destroying the economics. He thinks the cyclical capex math will damn these companies once the S-curve flattens.
“And I think that's what's changed too, where there's sort of a law of CapEx where, you know, high amounts of CapEx spend produce low future returns because people always overbuild and overestimate demand.”at 9:29
Tobias Carlisle · Acquirers Funds / The Acquirers Podcast
Hyperscaler AI spending is debt- and equity-funded, so it must show revenue soon or the boom collapses.
The hyperscalers and semiconductor companies are in an unprecedented investment gold rush. The hyperscalers' spending doesn't come out of free cash flow; it has to be funded by equity and debt raising. Tobias thinks this means they must get to the other side and show revenue to keep the party going. Otherwise the AI boom will follow the classic hype curve and get ahead of itself, as with earlier technology installation phases that Carlota Perez describes as often attended by bubbles. He draws a parallel with the dotcom fibre overbuild, where the infrastructure builders went bankrupt and the later users and builders (Google, Meta) captured the profits. He wonders whether the semis will be the layer that gets destroyed by overbuilding.
“the fact that it doesn't come out of free cash flow and instead it has to come out of equity and debt raising makes me think that they really do need to get to the other side and sort of start demonstrating that revenue in order to keep the party going.”at 11:12
Tobias Carlisle · Acquirers Funds / The Acquirers Podcast
AI data center buildout looks like a gold-rush overbuild that overestimates near-term demand.
Tobias says history shows classic gold-rush overbuilds, where capacity is built on overestimated short-term demand. The economy has a lot of inertia, so it takes a long time for everyone to adjust and fully use the new capacity. He notes a counterpoint: older chips are being rented out at very high rates because of scarcity. He does not know how quickly the supply can be turned on. He still sees scarcity as a bottleneck that will eventually go away, which suggests the overbuild risk is real.
“Think if you look at history, it's a classic gold rush overbuild in the short term. Just overestimate the demand in the short term. There's so much inertia in the economy.”at 57:17
Dan Rasmussen · Verdad Advisers
Chip scarcity will go away, and the buildout will overbuild capacity that stops all at once.
Dan agrees that the scarcity bottleneck has to go away, and that it has to go along with capacity overbuild. He says that when demand stops, it will stop all at once, because everyone is doing the same thing. The buildout is a crowded, synchronized bet, so its unwinding will be abrupt.
“It's got to, and it's gotta go with capacity overbuild. Right? And, you know, when the demand stops, it'll stop all at once because everyone's doing the thing.”at 57:58
Dan Rasmussen · Verdad Advisers
Local opposition to data centers is a real political risk that is motivating midterm voters.
Dan notes that data centers and pickleball courts are the two topics dominating city council hearings across America. He finds it odd, since a data center seems like just another big building. He admits he does not fully understand the opposition, and says every factory produces noise, so it should be resolvable by siting them in an industrial area. Still, he thinks the political issue is real and that it has motivated people in the midterm elections.
“I don't get it, but it seems to be a big issue for people... I do think the political issue is real and it seems to have motivated people in the midterm elections.”at 1:00:38
International equities
Where everyone stands on International equities
Dan Rasmussen · Verdad Advisers
Korean stocks are very cheap, and governance reform should re-rate the market and its holdco and preferred discounts.
Korean stocks trade around 8x earnings and 0.4x book, far cheaper than Japan at about 10x and 0.6x. The discount comes from family control: families use holding companies that own about 30% of an operating company yet control its board, and holdco stocks trade roughly 50% below the value of their stake. Families also issue preferred shares, which have no voting rights, and these now trade about 50% below the common. Holdcos and prefs together mean buying an already cheap opco at a 40-50% discount, letting Dan build a small-cap Korea portfolio at about 0.3x book and 6x earnings. A populist movement is now pushing the legislature to make boards owe fiduciary duties to all shareholders, increase dividends and close the discount. Dan expects these changes to drive a revaluation over the next few years, including of these structures. He adds that Korea's exclusion from MSCI developed-market status, due to offshore won trading and licensing limits, is being addressed, bringing more inflows.
“First of all, you go to Korea and the stocks are so cheap. I mean, I thought Japan was cheap and I'm buying stuff there at like 10 times earning and point six times book.”at 22:57
Dan Rasmussen · Verdad Advisers
Japan's governance reform is real but early; buybacks and dividends rising, several more years to run.
Japanese companies have historically been extremely conservative. For about 20 years they hoarded earnings in cash, cross-shareholdings and real estate, rather than paying dividends or buying back stock, and managers often don't understand cost of capital. Dan is buying Japanese stocks at about 10x earnings and 0.6x book, and argues it would be a better economic decision for companies to buy back their own cheap stock than build factories at a low return. Change has been happening since 2023: buybacks, dividends and reductions in cross-shareholdings are rising sharply, and there is much more openness to cost of capital and balance-sheet size. The process is slow and conservative, though, and companies are only dipping their toes in. Dan says it will take a few more years, so the Japan governance trade is still early.
“the sort of Japanese way is that this thing is going very slowly. It's very conservative, but it is happening.”at 35:03
Dan Rasmussen · Verdad Advisers
Japan and Korea offer less competitive, better-priced opportunities than crowded US equities.
Dan prefers less competitive markets. He compares the US market to a thunderdome of the most vicious, aggressive capitalists competing. Korea and Japan have competitors too, but relative to the US they are less crowded. That is why he has been reluctant to put all his chips in the US basket. His firm spends a huge percentage of its time on Japan and recently Korea, against 5-10% on US value.
“relative to The US markets, like The US markets are thunderdome of the most like vicious aggressive capitalists in all mankind competing. And so I've just not been as eager to put all my chips in that basket.”at 43:39
Dan Rasmussen · Verdad Advisers
Samsung and SK Hynix are half of Korea's benchmark, so an AI downturn could drag the whole market.
Samsung and SK Hynix are half the Korea benchmark, so the Korean market moves with those two stocks. Their rise has been a big contributor to Korea's meteoric run. Dan sees two main scenarios. In the best case, they act as an advertisement that draws investors into Korean equities more broadly. In the worst case, if AI has a rough patch, those two stocks drag the whole market down. He does not think the worst case is likely, because other Korean stocks are not as correlated with them. He also notes that both companies have preferred and holding-company structures that let investors buy exposure to them at a discount.
“Those two stocks are half the Korea benchmark. So the minute you start trading Korea, you become very cognizant of how those two stocks are doing because as they go, it goes the benchmark.”at 44:58
AI & jobs
Where everyone stands on AI & jobs
Dan Rasmussen · Verdad Advisers
No evidence yet that AI is producing enough productivity or cost savings to justify AI spending.
For AI to pay off, ordinary companies have to become paying customers, to the tune of roughly $600 billion or more of revenue a year, about a quarter of SG&A spend at most large companies. Dan asks whether AI is actually delivering productivity gains and replacing workers, and says he sees no evidence yet of profit improvement, cost savings or productivity gains big enough to make the numbers work. He thinks AI people underestimate corporate inertia: even with perfect technology, companies take a long time to find uses, implement them, trust them in production and fire workers to replace them. He expects the timeline to be much slower. He also notes that call-centre replacement hasn't worked well yet and self-driving penetration is slow. He says he is hugely bullish on the technology itself but doubts it will roll out into everyday businesses at scale quickly.
“But I'm not seeing any evidence in the real world of significant enough profit improvement, significant enough cost savings, significant enough productivity gains to make the numbers work.”at 3:06
Dan Rasmussen · Verdad Advisers
AI is a massive productivity boost at Dan's firm, which is hiring more people rather than firing.
Dan says AI is doing a lot of work at his firm and is very impressive. He calls it a massive productivity enhancer for the firm. Despite that, the firm has not fired anyone because of AI. It is hiring more people because it is getting so much more done. He admits he does not know where these developments will land overall.
“here in our firm, it's doing a lot of work. It's very impressive... It's a massive productivity enhancer for us. But we haven't fired anyone because they've been replaced by AI, right? In fact, we're hiring more people because we're getting so much more done.”at 1:02:55
US stock market outlook
Where everyone stands on US stock market outlook
Tobias Carlisle · Acquirers Funds / The Acquirers Podcast
S&P 500 CAPE near 41.6 sits close to the 1999 peak of 44, signalling stretched valuations.
Tobias cites the cyclically adjusted P/E of the S&P 500, about 41.6 today. That is close to the 1999 peak of 44. He argues that the investors who lost in past crashes were the ones who paid too much, while the technology kept spreading. He notes that the Dow fell 89% after 1929 and the Nasdaq fell 78% after 2000. The valuation level, close to the peak, is his basis for caution on paying too much.
“Today, the cyclically adjusted [P/E] on the S and P 500 is about 41.6. In '99 at the peak, it was 44.”at 40:06
Dan Rasmussen · Verdad Advisers
US value is thin, but cheap healthcare and US energy stocks are the most fertile areas.
Dan says he does not do much in US value. His firm has spent only about 5-10% of its time there, compared with a huge share on Japan and recently Korea. When his firm does invest in the US, it has had decent exposure. He has found interesting opportunities in healthcare, where there are many cheap stocks. He has also found energy fertile, with a lot of cheap US energy names. Small-cap energy stocks are hard to find outside the US, so the US is the place to get cheap energy and resources exposure.
“I've found some pretty interesting opportunities in healthcare. There's a lot of cheap healthcare stocks these days. I'd say another area that I found pretty fertile is energy. A lot of cheap US energy.”at 44:12
Credit markets
Where everyone stands on Credit markets
Dan Rasmussen · Verdad Advisers
High-yield spreads are back near 330bp, still no material risk, but a blowout would signal crisis.
Dan's firm watches the high-yield spread, the gap between junk bond yields and Treasuries, as an economic indicator. The spread was unusually low this summer, below 300bp, meaning the market has priced almost no default risk for a long time. It has since widened to about 330bp. The long-term median is around 400bp, and about 600bp is crisis level. The spread has not reached 600 since March 2020, when it got into the high 500s around Silicon Valley Bank. Dan compares it to a forest with no fires: kindling is building, but the indicator does not say when the spark comes or what it will be. Once spreads blow out, the widening becomes self-reinforcing and accelerates the crisis. Weak hands supported by tight spreads will be hurt when they widen. He says that at 330bp the market is still not seeing material risk, but he is very nervous whenever spreads blow out.
“Now they're back up to like three thirty or so. The long term median is probably around 400. And we would say like 600 is crisis level.”at 50:45
Tobias Carlisle · Acquirers Funds / The Acquirers Podcast
High-yield spreads are a coincident rather than early-warning signal, but they show fundamental drawdown causes.
Tobias says the high-yield spread spikes only right in the teeth of a drawdown and is not really an early-warning signal. It moves in concert with the stock market rather than ahead of it. He still sees it as a good indicator in one sense. A widening spread tells you the market drawdown has a fundamental credit reason behind it rather than just being a sentiment move.
“But it's not really forward looking particularly... it'll move with the market. So I think good indicator in the sense that it does tell you that the drawdown in the market is there's some fundamental reason for the drawdown”at 54:35
AI bubble & tech valuations
Where everyone stands on AI bubble & tech valuations
Dan Rasmussen · Verdad Advisers
AI equity valuations are too high; a major market correction could follow if AI growth disappoints.
Dan's concern is about equity prices, not about AI itself taking jobs. Growth stocks need a lot of growth, and they need it fast. He says he doesn't see that growth materialising as quickly as people expect because real-world AI use cases are not yet working well. He notes AI revenue is still accelerating, which is what sustains the boom, so a meaningful deceleration would be needed to unwind it. He thinks that if the more dire forecasts prove right, the AI trade could trigger a major market correction.
“I'm quite worried about it and I think that it could cause a major market correction if some of these more dire forecasts are correct.”at 4:22
AI chips & semiconductors
Where everyone stands on AI chips & semiconductors
Tobias Carlisle · Acquirers Funds / The Acquirers Podcast
Cyclical chip stocks may be better bets when optically expensive on earnings, not when cheap.
Tobias notes a chat question about Micron trading at about 3x earnings, and says the same could be said of Nvidia and SK Hynix. These are historically cyclical businesses, so you have to take care when they are over-earning. When they trade on low multiples, the market is treating them as closer to the top of their cycle. When they trade on high multiples, it is treating them as closer to the bottom. Unlike a traditional value stock, they may be a better bet when they look more optically expensive. He contrasts this with the hyperscalers, where the bet is more nuanced and less clear.
“When they're trading on low multiples, that's the market sort of treating them as if they're closer to the top of their business cycle and higher multiples when they're closer to the bottom of their business cycle.”at 10:23
AI: overall outlook
Where everyone stands on AI: overall outlook
Tobias Carlisle · Acquirers Funds / The Acquirers Podcast
AI is real and revolutionary, likely as economically impactful as the Internet.
Tobias has been using AI tools like Work and Codex and building things with them, and finds it clear that the technology is doing something special and extraordinary, almost verging on magic. He thinks the technology is revolutionary and its economic effects will be profound. He argues it's different in kind from earlier technologies, and expects it will be at least as impactful as the Internet, which has transformed the economy over the past 25 years. He is writing a whitepaper on the singularity and the last invention.
“And I do think that the effects on the economy will be profound, probably in the same... I think it will be at least as impactful as the Internet itself.”at 15:00
Power demand & the grid
Where everyone stands on Power demand & the grid
Tobias Carlisle · Acquirers Funds / The Acquirers Podcast
When chips stop being the AI bottleneck, profits will shift to power and land.
Tobias argues that profits gather wherever an industry is still held back. IBM built the defining 1980s personal computer, but the profits went to Intel and Microsoft, the bottleneck owners. Today the bottleneck is computing power, and Nvidia earned a 75% gross margin in the quarter ending July. When chips stop being the constraint, those profits will move to power and land, or to whoever owns the computer.
“Today, the bottleneck is computing power. NVIDIA earned a 75% gross margin in the quarter ending July. When chips stop being the constraint, those profits will move again to power and land or to whoever owns the computer.”at 39:25
Private credit
Where everyone stands on Private credit
Dan Rasmussen · Verdad Advisers
Weak borrowers have migrated to private credit, so public high-yield spreads understate underlying stress.
Dan argues the public high-yield index has become higher quality because the bad debt has migrated to private credit. The stress is visible in the CCC-rated spread, which was around 900bp this summer and is now over 1,300bp, a large blowout. A credit hedge fund called Arini blew up recently in Europe when European CCC spreads widened. That took down a number of distressed debt funds. Dan says the lower end of the credit market is finally giving out. He is unsure whether this is a leading indicator of wider trouble or is contained among the weakest issuers.
“One of the issues is that the high yield spread has become higher quality because all the bad debt has migrated to private credit. So if you look at like the triple C rated spread, it has blown out quite a bit.”at 52:42
Software & SaaS
Where everyone stands on Software & SaaS
Dan Rasmussen · Verdad Advisers
Software will feel AI's disruption first, since AI is most advanced at coding.
Dan argues that software is the sector that feels the pain first from AI. That is why San Francisco is so concerned. Coding is among the tip-of-the-spear uses where AI is strongest, along with self-driving. He points to people in San Francisco riding Waymos and using agents to do their coding, which he sees as a preview of the future.
“if it's software that sort of feels the pain first. So that's probably why San Francisco is so concerned. Right. And it's what AI is best at right now.”at 1:02:17
Across the shows
What other episodes Chorus covered said about the same companies in the 30 days up to this one.
Topics
- AI displacing productivity software
- AI disruption of software
- AI economic impact
- AI productivity and hiring
- AI productivity and job replacement
- AI stock valuations
- AI substitutes for SaaS pricing
- Japan and Korea value
- Japan governance reform
- Korea equities and governance discount
- Korean index concentration in memory chip stocks
- US equity valuations
- US value sector opportunities
- bad debt migration to private credit
- circular financing of AI customers
- data center capex
- high yield spread as crisis indicator
- hyperscaler capex funding
- hyperscaler debt funding
- local opposition and siting
- power for AI
- semiconductor cyclicality and multiples
More from The Acquirers Podcast
- Is AI Killing Adobe? We Discuss · November 14, 2026
- The Quality Growth Playbook for Biotech | David Ferreiro · November 12, 2026
- Why Bill Ackman Loves Uber at 16x Earnings + Our 2 Favorite Cheap Stocks · November 7, 2026
- The Best Business You've Never Heard Of | Todd Wenning · November 5, 2026
- Who Pays for the AI Boom? Big Tech’s Financing Problem · October 31, 2026
Ask your AI about this episode
Chorus gives Claude, ChatGPT and other assistants every view from this episode and every other show Chorus follows. Ask who agrees, who changed their mind, and what else was said about the same companies.
Chorus reports what speakers said. It isn't investment advice. How Chorus works · About Chorus