Odd Lots Podcast Summary

Episode: Why Paul Kedrosky Says AI Is Like Every Bubble All Rolled Into One
Hosts: Joe Weisenthal & Tracy Alloway
Guest: Paul Kedrosky, MIT Institute for the Digital Economy & SK Ventures
Date: November 14, 2025

Brief Overview

This episode explores the enormous and increasingly complex economic, financial, and technological ramifications of the current AI boom—especially the massive wave of investment in data centers and infrastructure. Guest Paul Kedrosky characterizes the AI bubble as a confluence of all previous major bubbles—combining tech, real estate, loose credit, new financing vehicles, and a government backstop. The discussion examines capital expenditure risks, financing structures, unit economics, international competition, and the sustainability of current approaches.

Key Discussion Points & Insights

1. The Scope and Structure of the AI Boom

• AI Capital Expenditure (Capex):

  • The scale is measured in billions/trillions; “It’s becoming hard to keep up, so I think we’re probably just going to talk in terms of billions and trillions.” — Tracy Alloway [02:23].
  • AI Capex is analogized to Schrödinger’s cat—simultaneously a massive strength and possible weakness for companies, until outcomes are clear.
  • Tech giants and AI startups, like Anthropic and Meta, are heavily investing in or raising capital for data centers.

• Data Centers as GDP Growth Drivers:

  • Kedrosky notes that in recent quarters, data centers drove approximately 50% of US GDP growth:
    “I hadn’t realized what a large fraction of GDP growth in the first quarter data centers were. It was on the order of 50%...” — Paul Kedrosky [06:42].
  • Data center buildouts are an extraordinary private sector stimulus.

2. The “Meta Bubble”: How AI Combines Elements of All Past Bubbles

• Bubble Ingredients:

  • Tech/innovation, real estate speculation, loose credit, government involvement/backstop all present.
  • “For the first time, we combine all the major ingredients of every historical bubble in a single bubble... It’s like we said, you know what would be great? Let’s create a bubble that takes everything that ever worked and put it all in one. And this is what we’ve done.” — Paul Kedrosky [08:37].
  • The discussion positions this as historically unprecedented and unlikely to end with a “soft landing.”

• Loose Credit and Private Credit:

  • The rise of private credit (replacing “shadow banking”) is highlighted as a massive element:
    “It’s now like one point, whatever it is, $1.7 trillion is the size... which is larger than many components of the orthodox lending market combined...” — Paul Kedrosky [10:37].

3. Financing Structures and Market Risks

• Why Companies Use External Financing:

  • Even cash-rich tech companies are limiting on-balance-sheet Capex, using SPVs (Special Purpose Vehicles) and private credit to avoid diluting earnings and to preserve financial flexibility.
  • “We bring in other participants, create new financing vehicles, and then we play this entertaining game of ‘it’s not really our debt, it’s in an SPV, I don’t have to roll it back onto my own balance sheet’...” — Paul Kedrosky [11:25].

• SPVs & Securitization:

  • “There’s incredible complexity, but at the core it’s a mechanism via which I can raise more capital and keep it off my balance sheet...” — Paul Kedrosky [13:11].
  • Not inherently problematic, but widespread use at this scale is reminiscent of past crises.

• Asset-Liability Mismatch:

  • Data centers (and their assets, like GPUs) may only be viable a few years, while loans are much longer (sometimes 30 years).
  • “...Probably unprecedented temporal mismatch with 30 year loans and 2 year depreciation on the underlying collateral...” — Paul Kedrosky [28:48].

4. Data Center Economics, Technology Cycles, & Yield

• Depreciation Schedules and Tech Obsolescence:

  • Chips for high-intensity model training have short lifespans (“like 18 months to 2 years”), unlike those just for storage [16:56].
  • Short-lived hardware complicates financing and profitability.

• Tenant Risk and Securitization:

  • Data centers likened to apartment buildings: hyperscalers (like Google) are low-yield, stable tenants; speculative operators may blend in flightier tenants for higher yields, echoing CMBS/subprime logic.
  • Leads to tranching and complex asset-backed securities (ABS).

• Negative Unit Economics:

  • For most current AI models, costs scale with usage—unlike traditional software.
  • “...We lose money on every scale and try to make it up on volume. That’s the problem here.” — Paul Kedrosky [24:25].
  • Projections for returns involve heroic assumptions about user numbers or market capture.

5. Power, Energy, and Collateral

• Energy Constraints:

  • Data centers run into power availability, compounding risk—e.g., Amazon data centers without sufficient grid power in Oregon [28:48].
  • Some companies are building private power plants, which are long-term assets mismatched with short-lived IT hardware.

• Compute Hoarding and Game Theory:

  • Companies hoard data center capacity and GPUs, sometimes buying more than needed just to deny rivals access.
  • “Once you start thinking of compute as a hoardable commodity and what people are doing is trying to hoard it, control it before someone else can do it...” — Paul Kedrosky [36:19].

6. Global Competition: U.S. vs. China

• Divergent Strategies:
  • U.S. focuses on massive investments in closed models and infrastructure.
  • China emphasizes rapid adoption and efficient open-source models.
  • “The approach of distillation is about efficiency. The Chinese are showing the huge efficiency gains to be had. And...if there are all these efficiency gains ahead...aren’t we completely misforecasting the likely future arc of demand for compute? And the answer is yes.” — Paul Kedrosky [37:55].

7. What Are AI Companies Really Building?

• Purpose Drift:
  • Internally, companies say they are building productivity tools; to rationalize the massive spend, they evoke the idea of creating AGI—a “faith-based call option”.
  • “It’s the first thing [business tools] until you challenge them, and then it’s the second [faith-based arg.].” — Paul Kedrosky [41:28].

8. Structural Economic Risks and Spillovers

• Systemic Risks Beyond "Private Markets":

  • If the bubble bursts, repercussions affect more than just VC/private credit:
    “...It’s firefighters and teachers money. And it’s in retail. Look at the larger holdings and REITs now, increasingly our data centers...” — Joe Weisenthal and Paul Kedrosky [46:01].
  • Wealth effects linked to the S&P 500's AI exposure; popping of this bubble could have broad economic impacts.

• Is AI Capex Just Another Stimulus?

  • The mainstream economic view (stimulus is stimulus) is misleading in this context; if ROI fails, the wealth effect unwinds, with potential fallout for pensions, funds, and the broader market.

Notable Quotes & Memorable Moments

• On the uniqueness of the AI bubble:

  • “This is the first bubble that has all of that. It’s like we said, you know what would be great? Let’s create a bubble that takes everything that ever worked and put it all in one.” — Paul Kedrosky [08:37]

• On collateral and risk:

  • “We’ve got this probably unprecedented temporal mismatch with 30 year loans and 2 year depreciation on the underlying collateral, which is essentially the GPUs…” — Paul Kedrosky [28:48]

• On negative unit economics:

  • “The term of art obviously is these things have negative unit economics, which is a fancy way of saying that we lose money on every scale and try to make it up on volume.” — Paul Kedrosky [24:25]

• On China’s approach:

  • “The Chinese are showing the huge efficiency gains to be had...this [distillation approach] is about efficiency gains…” — Paul Kedrosky [37:55]

• On the AGI promise:

  • “…when you walk through some of the math in terms of justifying the ROI on the spend, all of a sudden then it turns into what I call faith based argumentation about AGI. And they say it’s like the greatest call option ever.” — Paul Kedrosky [41:28]

• On the wealth effect:

  • “These are a massive negative wealth effect when you unwind it, not just in terms of the direct spending, but in terms of the wealth effect with respect to what people’s holdings are.” — Paul Kedrosky [44:34]

Important Timestamps

| Time | Segment/Topic | |----------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------| | 02:00 | Hosts introduce AI Capex explosion; compare to past booms. | | 06:42 | Kedrosky explains just how crucial data center buildout is to current GDP growth. | | 08:37 | Key thesis: “Meta Bubble”—AI as a combination of all previous bubble ingredients. | | 10:37 | Explains the rise of private credit and its systemwide scale. | | 11:25 | Why even cash-rich tech companies use external financing and SPVs for data center Capex. | | 16:56 | Economic and technical nuances of data center lifespans and depreciation. | | 21:50 | Securitization, tranching, and the parallels to pre-2008 subprime risk. | | 24:25 | Unit economics, negative returns at scale, and investor expectations. | | 28:48 | Risks from asset-liability mismatches—short-lived tech, long loans; energy supply issues. | | 36:19 | Why companies contract for third-party compute (“hoarding”); game theory around capacity. | | 37:55 | U.S.-China: contrasting AI strategies; distillation and efficiency gains in China. | | 41:28 | What are AI companies really building—business tools or faith-based bets on AGI? | | 44:34 | Capex as economic stimulus—potential positive/negative spillovers if the bubble bursts, and why it's not like “digging and refilling holes.” | | 46:01 | Real-world implications for pensions, REITs, and more; systemic risk is not contained. |

Tone and Style

• Language: Analytical, candid, sometimes irreverent and humorous. Heavy on analogies (barking dog, Schrödinger’s cat, Le Mans cars, digging holes).
• Expertise: Conversational, but deeply informed and accessible.
• Mood: Curious, skeptical, and at times concerned.

Conclusion

Paul Kedrosky positions the current AI/data center investment boom as both unprecedented in scale and familiar in risk—combining all past bubble dynamics into a “meta bubble.” The episode is rich with insight on risks (financing, asset mismatch, negative unit economics), competitive strategies (especially vs China), and the possibility that, structurally, the AI buildout could drive not just private capital but future financial instability. Both hosts conclude that, with so much complexity and so many moving pieces, revisiting this topic will be essential as the story unfolds.

For more from Paul Kedrosky: [paulkadroski.com]
For Odd Lots discussions: bloomberg.com/oddlots or join the Discord channel.

“Life finds a way.” — Tracy Alloway, (in a Jurassic Park callback) [48:17]