Aug 12, 2026 · 1h 4m · big-technology
Why The AI Bubble Will Burst: The Most Logical Case — With Paul Kedrosky
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of the Big Technology Podcast, host Alex Kantrowitz and investor Paul Kedrosky deliver an analytical critique of the artificial intelligence boom, arguing that unprecedented capital expenditures, rapid token deflation, flawed real estate financing models, and systemic debt contagion make a severe financial correction inevitable.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 31.9% of the talking time here. How this is scored →
Alex as informed peer 6.1 Guest teaching 7.0 Guest disagreement 5.5 Alex pushing back 4.5
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 19:15 Paul dismisses Jevons paradox as economic enumeracy
Paul aggressively attacks the standard bull argument, calling the invocation of Jevons paradox historically naive and mathematically innumerate given the requirement for hundred-million-fold volume growth.
Hardest push from Alex ▶ 16:05 Alex challenges depreciation thesis with CoreWeave chip pricingAlex constructs a multi-layered counterargument using first-hand reporting from CoreWeave to challenge Paul's thesis on hardware obsolescence and falling token yields.
Biggest teaching moment ▶ 36:25 Paul dismantles the Perez framework rationalePaul breaks down the theoretical flaw in modern tech investors citing Carlota Perez, demonstrating how justifying present overspending based on historical survivorship creates a dangerous reflexive bubble.
Alex holds their own ▶ 51:45 Alex synthesizes market headwinds across open source, pricing, and appsAlex systematically lays out concrete competitive factors driving down model margins, including small SLMs, Meta's open-source strategy, and super-app commoditization.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Historical Precedents and the Scale of AI Capex | 6 | 5 | 4 | 2 | Alex prompts Paul to contextualize AI capital expenditure against historical precedents, adding his own observation about the compressed timeline. Paul lays out detailed macroeconomic metrics comparing current capex to electrification, railroads, and fiber builds. | |
| Evaluating AI Infrastructure Through Real Estate Frameworks | 5 | 7 | 5 | 3 | Alex asks what return is mathematically necessary on projected trillion-dollar spends. Paul reframes the problem entirely, correcting the premise by explaining how project financiers treat data centers as commercial real estate using cap rates rather than top-down TAM calculations. | |
| Structural Flaws in the Data Center Real Estate Analogy | 6 | 8 | 6 | 5 | Alex tests the real estate thesis by asking why investors should worry if Meta can easily cover standard yields. Paul forcefully deconstructs the analogy, detailing continuous capex churn, GPU failure rates, duration mismatch, and token price hyper-deflation. | |
| Debunking Jevons Paradox and Analyzing Model Convergence | 8 | 8 | 7 | 7 | Alex pushes back hard citing Jevons Paradox and CoreWeave data showing high rental rates for older chips. Paul dismisses the Jevons argument as innumerate wishful thinking, explaining that an 80% compounding price decline requires unrealistic volume growth. | |
| Frontier Labs Moving Up-Market and Enterprise Software Realities | 7 | 7 | 6 | 6 | Alex raises Alex Karp's CNBC comments and explores whether labs can capture enterprise software markets like Figma. Paul counters by arguing frontier labs cannot simply swallow the enterprise stack because SaaS exists primarily to absorb liability and customer service overhead. | |
| The Perez Framework and the Fallacy of Overbuilding | 6 | 8 | 7 | 5 | Alex presents the technological inevitability thesis that transformative tools always justify overbuilding. Paul attacks this reliance on Carlota Perez frameworks, pointing out that reflexive overspending justified by past survivorship bias has historically triggered severe economic crashes. | |
| The AGI Call Option and Sovereign Fund Dynamics | 6 | 8 | 7 | 5 | Alex presses on whether massive funding is rationalized as an unpriceable call option on AGI. Paul terms AGI a 'God of the gaps' marketing narrative used in public but explains that institutional allocators and sovereign wealth funds are driven by check-size filters and herd mentality. | |
| Sponsor Spotlight: Gravity AI Agent Security Documentary | 5 | 7 | 5 | 4 | Following an ad break, Alex asks what specific catalyst could halt the funding cycle. Paul explains the system is overdetermined and details how reliance on software harnesses over massive pre-training runs could quickly destroy hyperscaler capex narratives. | |
| Financial Contagion and Personal Investment Positioning | 6 | 6 | 5 | 5 | Alex asks how financial contagion would spread if capex slows and asks Paul how he positions his portfolio. Paul details similarities to 2008 CMBS contagion across insurance balance sheets and notes his indirect short exposure through hedge funds. | |
| Comparing AI Infrastructure Dynamics in China and the US | 6 | 6 | 4 | 4 | Alex asks if state-backed Chinese AI infrastructure is insulated from market crashes. Paul explains that provincial overbuilding in China mimics previous cycles in solar, batteries, and ghost cities, creating massive misallocation without Western financial contagion. | |
| The Future of AI as a Commoditized Utility | 6 | 7 | 5 | 3 | Alex summarizes Paul's thesis and asks about the long-term endgame for frontier labs. Paul concludes that LLMs will not reach AGI due to static weights and will ultimately commoditize into invisible low-margin utility providers like hydroelectric dams. |
Statements from this episode (26)
Kedrosky: AI capex exceeds all historical Western buildouts except WWII rearmament
“As I said, we're now larger than everything except for World War II rearmament, which doesn't play as well in Germany as it does everywhere else. But nevertheless, so the point being that as a percentage of GDP, as a percentage of non-residential fixed investm…”
Kedrosky: Tech leads high-yield and non-financial investment-grade bond issuance
“Tech is now the largest piece of the high yield bond market. It's now the largest piece outside of financial services of the high of the, Investment grade, bond market, so in terms of new issuance”
Kedrosky: Over 50% of data center funding is external financing as of Q2 2026
“As of the second quarter of twenty-twenty-six, this is now more than 50% of the funding for data centers is, is external financing, which is obviously the term of art for off balance sheet and out of your own cash flows”
Kedrosky: Financiers evaluate AI data centers like multi-tenant apartment buildings
“The way that I try to analogize this loosely, and this is very loose, is that data centers from the context of many capital providers are real estate. They're really just multi-tenant apartment buildings. It just so happens there's no humans in the apartment b…”
Kedrosky: AI infrastructure models based on labor TAM are ridiculous and speculative
“It's wrong to say, for better or worse, we're going to be putting in a trillion, therefore I need a hundred trillion out of this. That's not the way investors are looking at this, and it's, and it will lead you down the wrong path if you take that approach, be…”
Kedrosky: AI Data Centers Require Major Hardware Overhauls Every 4-7 Years
“This is a project that would require wholesale replacement of most of the hardware and probably changes in the cooling system and probably changes in other aspects of these data centers continuously and probably, you know, depending on the math, anywhere from …”
Kedrosky: Some Data Center GPUs Fail on an 18-Month Cycle
“So we have some GPUs that are failing inside of modern data centers on an 18 month cycle, some that are failing on a much longer period.”
Kedrosky: AI Token Prices Fall 70-80% YoY on Constant Performance
“And these tokens are among the most rapidly depreciating assets we've ever seen in a modern economy, that they've continually been falling 70 to 80% year over year on a constant performance basis for at least the last four years.”
Kantrowitz: CoreWeave was renting Nvidia H100s at higher rates
“So I was speaking with CoreWeave at the end of the year last year, beginning of the year this year, you know, around New Year time and they said they were actually renting out H-one hundreds for higher prices than they had previously.”
Kedrosky: GPU failure rates are much higher in training than inference
“The failure rates of GPUs used so intensively for training purposes are much higher than inference specific usage.”
Kedrosky: AI models are converging and now primarily compete on price
“While models are still improving, they're improving at a much slower rate. And I often do this kind of Pepsi Coke test where I'll put a couple of different models in front of people using some kind of a harness like open code and ask them to tell the differenc…”
Kedrosky: Token demand is unlikely to achieve 100M-fold growth in 6 years
“You have to see around a hundred million fold growth over the next six years in terms of tokens. Is it possible? Absolutely. It's possible. Is it likely? No, it's not likely, but it could happen.”
Kedrosky: Minimal model differentiation will crush AI investment returns
“The convergence means that the model differences while there are so minimal as I can't tell the difference in the kind of Pepsi Coke phenomenon, which again, to cut to the investment chase suggests that the competition then becomes much more about marketing ex…”
Kedrosky: Enterprises buy SaaS for liability and vendor accountability, not innovation
“There's a deep misunderstanding about why companies buy software. It's not because they think service now or Salesforce or whoever is somehow, you know, bold innovators that could not be replaced. No, it's because they have a problem. They don't want to build …”
Kedrosky: Insurance companies are among largest buyers of data center debt
“Increasingly some of the largest purchasers of data center related debt or insurance companies.”
Kedrosky: Justifying AI spending with historical bubble theory drives dangerous overbuilding
“We've created this reflexivity. We're now, we justify overspending on the basis of prior overspending having worked out. Well, in prior episodes where that happened, people were not justifying the overspending by saying, say, in rural electrification, you know…”
Kedrosky: AI investors pitch AGI to LPs but evaluate it as real estate internally
“Investors that I've talked to are very cynical, so they're perfectly happy to use that in front of their own LPs, but they don't believe that in house. They look at this very cynically and with very cold calculating eyes and compare it to other similar real es…”
Kedrosky: Sovereign funds filter investments by check size rather than economic value
“Once you're managing hundreds of billions of dollars, you start looking at opportunities, not in terms of their economic value, but in terms of check size. And you say, I need to write a check for fill in the blank, a hundred billion dollars, because I do not …”
Kedrosky: AI is the first U.S. bubble combining tech, real estate, credit, and policy
“The largest bubbles in US history usually had either to do with technology loose credit, government policy, right? Some combination of these things. The US in particular is very good at ones that also include real estate, so we can add that to the mix. So tech…”
Kedrosky: Recent AI gains stem from orchestration harnesses, not model breakthroughs
“So much of the improvement we've seen in the last 18 months has really been about the imposition of harnesses, effective nannies, sitting on top of bratty kids, and not about the actual structural improvements in the models themselves, and that's a sort of a h…”
Kedrosky: Investors will pressure AI labs to slash massive pre-training spend
“Once investors look under the hood and see more and more of this, they'll be questioning, why are we spending so much on pre-training? Why are you doing billion dollar training runs anymore? If most of the gains and models are coming from post training and RLH…”
Kedrosky: AI and hyperscalers make up 40-odd percent of S&P 500
“And so people are realizing that they're in it even by holding an S&P 500 index fund because of the concentration of hyperscaler and AI related names, which is something like 40 odd percent now.”
Kedrosky: I have avoided passive index funds for over two years
“I'm very loathe to invent, to commit any, haven't committed new capital to any sort of broad index class passive categories in over two years for that reason.”
Kedrosky: Chinese Premier Is Warning Provincial Governors to Stop Overbuilding Data Centers
“And so the Chinese premier has been recently cautioning the provincial governors, stop building so many data centers because this is now the new thing.”
Kedrosky: China Will Overbuild AI Data Centers Like Past Real Estate Booms
“I expect you'll see the same phenomenon, albeit with a little less social consequences once all of this turns out to be, it'll be much like what happened with their overbuilding in apartment buildings and residential and industrial space over the last decade.”