Talk Your Book: AI Is Not a Bubble

Episode Summary

Executive Summary: The episode features Dr. Ankor Crawford arguing that AI is not a bubble but an exponential, supply-constrained supercycle affecting semiconductors, power, and industrials. She says demand for compute is outrunning supply across the entire chain, lifting pricing, margins, and earnings potential for select infrastructure names while changing the economy’s growth engine from consumer-led to industrial-led.

Main Topics: AI as an exponential, non-linear innovation cycle (Priority: 5/5): Crawford argues that AI should not be judged by historical tech booms because software can now improve itself and digital adoption is immediate, creating a pace of change unlike prior cycles. Supply constraints across the semiconductor ecosystem (Priority: 5/5): She emphasizes that shortages are not limited to memory; they span foundries, equipment makers, networking, and power, making the whole AI supply chain rate-limited. Reframing valuation through earnings growth (Priority: 4/5): Crawford says investors focus too much on price multiples and not enough on what earnings should be when pricing and demand are still rising due to shortages and adoption. The market confirmation problem (Priority: 4/5): The hosts revisit her earlier concern that the market seemed to disagree with her on Micron, but she says recent stock moves and business fundamentals now confirm the thesis. Open-source vs frontier model economics (Priority: 4/5): She argues that frontier models should capture value at the model layer, while open-source models likely push value to the infrastructure layer because they become highly competitive. Broadening beyond AI into industrials and power (Priority: 4/5): Crawford frames AI as part of a wider shift toward reshoring and capex-led growth, highlighting power, independent power producers, and industrial suppliers as beneficiaries.

Key Arguments: The AI trade has duration because demand is still outrunning supply, as shown by rising GPU-hour pricing and accelerating customer usage. Historical bubble comparisons are misleading because this cycle is digital, exponential, and rapidly adoptable, unlike prior hardware rollouts. Semiconductors have moved from neglected cyclical businesses to constrained bottlenecks across the entire ecosystem, improving pricing power and margins. Valuation debates are incomplete without first determining the correct earnings base, which is still moving higher in a shortage-driven market. The supply chain itself limits how fast capacity can expand, reducing the risk of immediate bubble-like oversupply. Investors should distinguish between trades and long-term investments; Crawford targets roughly three-year holds and looks for names that can triple. The portfolio’s AI exposure is not just semis; it includes power and industrial names that benefit from reshoring and data-center buildout. Open-source AI may commoditize model-level economics, but frontier models and infrastructure providers can still capture meaningful value. Skepticism is useful, but current fundamentals such as pricing increases and profitability rumors suggest the market is still underestimating the cycle.

Data Points: Micron share price: $454 to $1,080 - The hosts cite Micron’s move since the prior conversation as evidence that the thesis was early but correct. Micron price move: +135% - Approximate gain since the prior episode mentioned by the hosts. Micron price move: +136% - A repeated estimate in the conversation describing the stock’s rise. Nebius price: 95 to much higher (not precisely stated) - Referenced as an example of AI-related stock appreciation; host notes it was around 95 in late March. Astera Labs price: 95 to close to 400 - Used to illustrate the speed of market re-rating in AI infrastructure. Anthropic ARR: 30B to 44B-49B - Crawford cites rapidly rising revenue estimates as evidence of demand strength. Hopper pricing: up 30% to 40% - Nebius old-chip pricing reportedly rose instead of falling after new-generation chips launched. Expected Hopper pricing decline: down 15% to 20% - Crawford says this was the expected direction once newer chips arrived, but pricing moved the opposite way. GEV turbine pricing: $1,250 per MW to about $2,500 per MW - Example of pricing power in AI-adjacent power infrastructure. Holding period: about 3 years - Crawford says she prefers investments she can hold for roughly three years. Potential upside target: triple in three years - Her mental framework is whether a position can triple over a three-year horizon. Portfolio size: about 30 names - The concentrated equity strategy is described as holding a limited number of positions. TSM competitive structure: 1.5 to 1.75 players - Crawford describes the foundry market as highly concentrated, with Samsung and Intel trying to regain footing.

Pivotal Quotes: "There is no historical precedent for what is happening today in terms of how fast the innovation is occurring." — Dr. Ankor Crawford: Her core rebuttal to the idea that today’s AI buildout should be judged by prior bubbles. "We are, in fact, rate limited by the supply chain, which is a really important aspect of what is happening." — Dr. Ankor Crawford: Explains why she thinks AI spending can persist without immediate oversupply or bubble dynamics. "The overall economy is going to move from a consumer-led economy into an industrial-led economy." — Dr. Ankor Crawford: Her broader macro thesis linking AI, reshoring, and industrial capex.

Implications: Listeners should watch AI infrastructure, power, and industrial suppliers as the real bottlenecks and beneficiaries. If Crawford is right, the opportunity extends well beyond software and may reshape earnings, valuations, and the economy’s growth mix for years.