Apr 4, 2024 · 1h 8m · bg2-pod
Ep6. AI Demand / Supply - Models, Agents, the $2T Compute Build Out, Need for More Nuclear & More · Bg2 Pod
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 Episode 6 of the BG² Podcast, Brad Gerstner and Bill Gurley evaluate the AI market boom, comparing high private startup valuations to dot-com era bubbles while analyzing the massive global compute buildout. They highlight power grid constraints and advocate for US nuclear energy regulatory reform to meet exponential AI energy demands.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Brad and Bill hold 100% of the talking time here. How this is scored →
Brad and Bill as informed peer 7.4 Guest teaching 4.1 Guest disagreement 2.5 Brad and Bill pushing back 3.0
speaking balance: gold is Brad and Bill, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 26:39 Bill rejects Sam Altman's demand framing
Bill bluntly dismisses Altman's price elasticity argument as banal and obvious, saying that lowering prices on anything increases demand and calling it unprovocative.
Hardest push from Brad and Bill ▶ 27:06 Brad defends compute elasticity with aviation analogyBrad rejects Bill's dismissal by arguing that lowering commercial airfare tenfold radically transformed global transit, proving that massive compute drops will unlock new intelligence workloads.
Biggest teaching moment ▶ 14:51 Bill breaks down structural limitations of LLMsBill provides a clear breakdown of why LLMs succeed in text and programming due to structured language context windows, but cannot simply be applied to physical manufacturing environments.
Brad and Bill hold their own ▶ 57:47 Bill exposes GenAI pricing fragilityBill details the 60x runtime price collapse between high-end and standard frontier models and the rapid capitulation on $20 monthly consumer subscriptions to illustrate lack of pricing power.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Brad and Bill as informed peer | Guest teaching | Guest disagreement | Brad and Bill pushing back | Why |
|---|---|---|---|---|---|---|
| Friendly Banter & March Madness | 6 | 2 | 1 | 2 | Brad and Bill engage in lighthearted banter about March Madness before Brad introduces historical market analogies like 1998 internet valuations to frame AI frothiness. | |
| Enterprise AI Demand, Co-Pilots, and Autonomous Agents | 7 | 3 | 1 | 2 | Brad details the progression from developer co-pilots to LangChain-style autonomous agents and asks Bill whether current enterprise demand is as structural as 1998. | |
| Practical AI Use Cases vs. LLMs (Tesla FSD Case Study) | 7 | 6 | 4 | 4 | Bill pushes back on generic LLM hype by contrasting Tesla's FSD V12 finite input/output AI with LLM-based wrapper applications, expressing caution over enterprise data integration. | |
| Private Valuations, High Burn Rates, and Startup Failures | 7 | 5 | 2 | 3 | Bill explains the structural trap of high private valuations and high burn rates for pre-revenue startups using Inflection and Stability AI as examples. | |
| Capital Intensity of Frontier Models & Nvidia's Demand | 8 | 3 | 2 | 3 | Brad argues that training demand and inference demand remain massively underestimated, recalling his Nvidia thesis when critics called for taking profits earlier. | |
| Sovereign AI & Redefining Compute Demand Across Industries | 7 | 6 | 3 | 3 | Bill clarifies the distinction between generalized AI and LLMs, explaining that LLMs excel specifically at structured language like code and customer support rather than manufacturing floors. | |
| Next-Gen Models (GPT-5), Real-World AI, and Value Capture | 7 | 4 | 3 | 4 | Brad anticipates multi-modal frontier models like GPT-5 while noting commoditization risks for mid-tier LLMs, while Bill remains skeptical about whether early agent wrappers demonstrate exponential value. | |
| Sam Altman's Narrative Framing & Jevons Paradox in Compute | 8 | 6 | 6 | 6 | Bill calls out Sam Altman's promotional tactics regarding compute and energy limits, while Brad counters by invoking Jevons Paradox and price elasticity of intelligence. | |
| Global Infrastructure Buildout & The $100B Stargate Project | 7 | 3 | 2 | 3 | Brad outlines massive global capex commitments in sovereign compute across the Middle East, Europe, and Microsoft's rumored $100B Stargate supercomputer. | |
| Power Constraints and the Nuclear Fission Imperative | 8 | 5 | 2 | 2 | Bill and Brad discuss surging power grid constraints and advocate for nuclear fission, citing figures like Josh Wolfe and Steven Pinker while pointing out China's massive reactor buildout. | |
| US vs. China Nuclear Race & Natural Gas as a Bridge | 8 | 4 | 2 | 3 | Brad points out China's lead in reactor construction and 10-year US lead times, highlighting natural gas as an indispensable near-term bridge for gigawatt data centers. | |
| Over-Regulation, Infrastructure Speed, and Innovation Barriers | 8 | 3 | 2 | 2 | Bill uses the rapid repair of the I-95 bridge in Pennsylvania and renewable buildouts in Texas to argue that excessive regulation and bureaucracy paralyze American infrastructure. | |
| Public Perceptions of Energy & Zero-Based Regulatory Reform | 7 | 3 | 2 | 2 | Both hosts contrast French nuclear deployment cost curves with US escalating costs, demanding zero-based regulatory reform to unblock fission energy. | |
| Tech Policy Entanglement & Stock Check: Constellation Energy | 7 | 4 | 2 | 2 | Brad discusses the growing geopolitical entanglement of tech and energy policy, while Bill discloses his long position in Constellation Energy (CEG) based on shifting nuclear sentiment. | |
| Federal Reserve Outlook, Rates, and Software Multiples | 8 | 3 | 2 | 3 | Brad reviews Fed economic projections, 10-year yields, hedge fund risk reductions, and the multiple divergence between legacy enterprise software and the Mag 6. | |
| Generative AI Valuations, Fast Failures, and Price Collapse | 8 | 5 | 4 | 4 | Bill and Brad debate high revenue multiples on GenAI startups, the collapse of $20/month subscription defensibility, and the lack of developer switching costs. |
Statements from this episode (34)
Gurley: Market Enthusiasm Makes Marginal Tech Investing Very Difficult
“Well, and it gets, I think it gets even tougher if enthusiasm builds, because that impacts the entry price on the marginal investment that one might make, and I think this particular moment in time is very, very Difficult for investors that are, you know, look…”
Gerstner: Enterprise AI Copilot Development Has Become Ubiquitous
“In a pretty short period of time, these have become really ubiquitous development projects for almost every enterprise in just a few short years wanting these co-pilots.”
Gerstner: AI Evolved From Copilots to Autonomous Systems in Under Two Years
“We've gone from co-pilot to some of these, maybe a little bit more autonomous systems, magic and cognition, all of this in less than two years.”
Gurley: AI Productivity Gains Drop Significantly Outside of Coding Tasks
“We've also talked about the fact that when you move away from coding, the efficacy drops a bit in terms of the productivity gain you might get.”
Gurley: Tesla Full Self-Driving Is Traditional AI, Not LLM-Based
“One thing that, that I think is really different between the full self-driving data point and is that it's not built on LLMs. It's a traditional use case of AI, which allows it to I think have much more finite input, output, and that example use case is fairly…”
Gurley: Enterprise Projects Often Relegate LLMs to Mere Database Interpreters
“A lot of the incremental work that's being done on AI projects are Stitching together external databases with the LLM and the LLM gets relegated to being an interpreter of the human or to being the interpreter of the data back to the human, but is not involved…”
Gurley: Inflection AI and Stability AI Suffer Early Falls From Grace
“So one thing we've seen since you and I talked is a couple of, we'll just call them, you know, early falls from grace, you know, with inflection and stability AI.”
Gerstner: Inflection's Microsoft Deal Was Not a Terrible Stakeholder Outcome
“I don't think it was a terrible outcome for the investors, for Mustafa or for Microsoft”
Gerstner: LLM Compute Intensity Fundamentally Alters Startup Risk-Reward Dynamics
“The capital intensity of that undertaking is very different than starting a software company, right? And so the risk reward to both the founders and the risk reward to the investors changes a lot.”
Gerstner: AI Enterprise Ubiquity Will Match the Database and Internet
“It tells me that the use cases are unlike anything else we've seen since perhaps the database or the internet itself, where those became ubiquitous in every single enterprise.”
Gurley: LLMs Will Provide Little Utility on Manufacturing Floors
“You know, LLM's not going to do much for you on a manufacturing floor. The LLM part. AI might but the LLM won't.”
Gerstner: At Least 20 LLM Companies Will Go to Zero
“There are gonna be 20 LLMs that go to zero or many more”
Gerstner: Open Source Will Rapidly Commoditize Non-Frontier AI Models
“Apart from the people who are on the very frontier, if you're on the frontier and you have something totally different, it seems to me that that's a place where that is defensible. But if you're not on the frontier, man, it seems that these are going to be rea…”
Gurley: AI Agents as App Connectors Lack Exponential Scale
“It doesn't appear to me that if agents are connectors to the external world, you know, to me, that's a lot like an enterprise when people would build connectors to other apps and I just don't see it on this kind of exponential scale that we got, you know, as t…”
Gurley: Sam Altman's Energy Compute Framing Is Jobsian-Level Promotion
“I want to, there's two things Sam did this in this recent Interview that I think put him in the promotion hall of fame. And one of them was he juxtapositioned AI versus the smartphone market and then on the smartphone market is being small. He said, well, that…”
Gerstner: Falling Compute Prices Will Massively Surge Aggregate Consumption
“If you drive down the price of the cost of compute, then the reflexivity is people will consume a lot more of it. Now, this is also known as the Jevons paradox, right? As price goes, goes down, we demand more of it. The aggregate amount of Consumed, of the com…”
Gerstner: US Data Center Power Consumption Projected to Reach 18% by 2030
“And here's a forecast I think coming out of semiconductor analysis, it's similar to a lot of the other ones, and it has data centers as a percentage of US power generation going from something like four percent today to something like, you know 18, 19% in 2030…”
Gurley: US Nuclear Deployment Costs Are Limited by Regulation, Not Technology
“The biggest problem, as I understand it, once again, from talking to experts, not from my own direct knowledge, but our cost of deploying new nuclear fission infrastructure is limited by our own regulatory framework, not by the technology.”
Gerstner: China Builds 300 Nuclear Plants While US Builds Zero
“They have over 300 plants currently either being built or in development. We have zero. Zero plants being built. I think 13 plants that are proposed according to this data that we have here.”
Gerstner: Nuclear Power Cannot Satisfy AI Energy Demand Over Next Decade
“Then all the stuff that we just talked about, the demand that we've got to satisfy over the next 10 years is not going to come from nuclear. It's got to be an all of the above strategy.”
Gerstner: Microsoft Opens a New Data Center Globally Every Three Days
“Microsoft alone is opening a new data center globally every three days.”
Gurley: Renewable Energy Projects Succeed More in Texas Than California
“Another article that I want to link to shows that, that renewable projects are way more successful in Texas than California, which is the ultimate irony, right?”
Gerstner: Cheap Calcium CT Scans Are Blocked by Washington Lobbying
“Calcium CT scans cost a hundred bucks and save lives can bend the healthcare curve around the three million sudden cardiac events we have on an annual basis in this country, but the reason it doesn't happen is they lost the lobbying game in Washington.”
Gerstner: France Out-Innovating the US in Nuclear Fission Is a National Embarrassment
“The French are driving more innovation and more efficiency out of nuclear fission than the United States, and we pride ourselves on innovation. That should be a shock and a national embarrassment.”
Gurley: Energy Market Agrees Regulation Cripples US Nuclear Progress
“I don't think there's anyone in the energy market that doesn't think The bureaucracy and regulation is the problem in the American nuclear fission market.”
Gerstner: US Cannot Resolve AI Leadership Without Addressing Energy Needs
“And so, if Washington wants to get serious, it needs to be an integrated, right, national policy. You can't fix AI without also taking on our future energy needs.”
Gurley: Bought Constellation Energy Betting Nuclear Power Perceptions Would Shift
“I, yeah, I bought this stock, and I, and by the way, I bought a lot of stocks that went down, so I think we got to be careful about cherry picking, but I bought this stock solely based on the bet that perceptions would shift on nuclear.”
Gerstner: The Fed Will Cut Interest Rates Before 2024 US Election
“I'm still in the camp that we're going to have rate cuts ahead of the election, but the market is now pricing and I think only three rate cuts between now and the end of the year.”
Gerstner: Public Software Trades at 6.1x Revenue, Below 10-Year Average
“You can see that in the blue line is it's trading at about 6.1 times. The tenure average is 6.9 times. Rates have obviously moved up a lot off the bottom, but you know, this is a long winded way of saying that software is not really participating in this run u…”
Gurley: Relative Valuation Games Are How Financial Bubbles Are Built
“There's this relative valuation game, which is how bubbles are built. And because you adjust up to, so you re-rate to a new norm, right?”
Gerstner: $20 Monthly Fees for Consumer AI Are Not Defensible
“Listen, I don't think a 20 dollar a month fee for my consumer AI is going to be defensible. And the reason I don't think that's defensible is because they're going to be way too many people in the pool. Apple's going to have one. Google's going to have one. Me…”
Gurley: Late-Stage AI Startup Multiples Make Me Nervous
“So when I look at all those things, And the entry prices that a marginal late stage investor would be asked to pay, which are these multiples here. I'd be like nervous. I'd be nervous. I'll leave it at that.”
Gerstner: Altimeter Passed on Many AI Startups Over Unjustifiable Multiples
“We passed on, you know, on a lot of these companies simply because we couldn't get comfortable with the entry multiples, given how opaque it still is at this moment.”
Gurley: Developer Switching Costs Between LLMs Are Extremely Low
“Today the switching costs based on every developer I've talked to are we mentioned this before, but about as low as you could possibly imagine. And if you're just using big context windows and not doing fine tuning, then you're just not, you can bounce from on…”