James van Geelen on the Next Phase of the AI Buildout

Episode Summary

Executive Summary: The episode examines whether the AI buildout is a bubble or a durable infrastructure cycle, using on-the-ground reporting from Stargate’s Abilene data center and related projects. The guests argue that even if valuations look frothy, the physical spending on power, cooling, GPUs, storage, and industrial equipment is real and still accelerating, benefiting “picks and shovels” companies while creating financing, power, and crowding-out risks.

Main Topics: Physical AI infrastructure buildout (Priority: 5/5): James Van Gielen describes touring Stargate’s Abilene site via drone and satellite imagery, emphasizing the sheer scale of the data-center construction and its resemblance to a major industrial buildout rather than a speculative software story. Power as the key bottleneck (Priority: 5/5): The conversation centers on electricity, turbines, and behind-the-meter generation as the binding constraint for AI data centers. The guests note that firms are increasingly building their own natural gas plants because grid dependence is too risky. Bubble debate vs. real demand (Priority: 4/5): The hosts and guest debate whether AI resembles prior bubbles. The view presented is that AI is likely in a bubble phase, but unlike dot-com, the spending is tied to real capital expenditures and full utilization of new facilities. Circular financing and private credit risk (Priority: 5/5): They discuss opaque financing structures—off-balance-sheet deals, GPU-backed loans, and circular investing among NVIDIA, OpenAI, CoreWeave, and others—as a hallmark of bubble-like behavior that could become fragile if capital markets tighten. Beneficiaries among industrial and equipment suppliers (Priority: 4/5): A major theme is the hunt for winners beyond chipmakers: Caterpillar, GE Vernova, Siemens Energy, Eaton, Mitsubishi Heavy, storage firms like Seagate and SanDisk, and other industrial suppliers are seen as direct beneficiaries of the AI capex surge. AI’s next form: video and robotics (Priority: 3/5): The guest argues that video models and robotics are the next frontier because they generate training data and interact with the physical world. This helps explain why firms are pushing into video generation despite skepticism about “slop.” Macro and labor spillovers (Priority: 3/5): The hosts raise broader concerns that AI construction may crowd out workers and resources in local economies, potentially contributing to inflationary pressure and labor shortages in places like Abilene.

Key Arguments: The AI story is not just software hype; it is a massive, real-world infrastructure cycle with concrete spending on land, power plants, cooling, and transmission. Data centers are already being used at or near full capacity, which the guest presents as evidence that this is unlike the dot-com era’s “dark fiber” overbuild. Power availability, not water, is becoming the central constraint because liquid cooling is reducing water use while electricity demand keeps rising. Hyperscalers have a prisoner’s-dilemma incentive to keep spending so they do not fall behind in the race to build “machine God” or AGI. The financing structure looks bubble-like because it involves off-balance-sheet vehicles, private credit, and customer financing, but the underlying demand is genuine. Industrial suppliers may offer a cleaner way to express the AI buildout trade than owning speculative AI software names, because their upside is tied to actual construction. Video models and robotics matter because they are more closely linked to the physical world and to future training data generation than chatbots alone. If the capex cycle stalls, the downside could spread beyond tech into the broader economy through lost spending, fewer jobs, and weaker industrial activity.

Data Points: Stargate planned investment: $500 billion - Total planned investment across the Stargate data-center buildout discussed in the interview. Stargate site count: 5 data centers - Number of data centers planned in the Stargate program. Stargate Abilene turbines: 10 natural gas turbines - On-site power generation seen in drone footage over the Abilene facility. Turbine capacity: 35 megawatts each - Output per turbine at the Abilene site; described as lower-end/simple-cycle equipment. Turbine suppliers: GE Vernova and Caterpillar/Solar Turbines - Half of the turbines are from GE Vernova and half from Caterpillar’s Solar Turbines unit. Construction workforce: 7,000 people - Estimated number of workers building the Abilene site. Abilene expansion: 600 megawatts - Planned additional capacity expansion mentioned for Abilene. Meta Hyperion debt financing: $26 billion debt / $3 billion equity - Structure of the Hyperion data-center financing discussed as an off-balance-sheet-like arrangement. Hyperion total project size: $29 billion - Total financing package for Meta’s Hyperion project. CoreWeave SPV: $7.5 billion - Blackstone-linked SPV financing referenced for CoreWeave. GPU-backed loan market: $11 billion - Size of the GPU-collateralized loan market cited in the discussion. Oracle debt leverage: Highest among hyperscalers - Guest notes Oracle has the highest debt-to-equity ratio among hyperscalers, though still with room to lever up. Caterpillar stock move: $334 to $486 - Share price rose from April 2 to an all-time high during the AI infrastructure rally. GE Vernova stock move: $330 to $609 - Share price cited as having more than doubled during the period discussed. Siemens Energy stock move: $56 to $108 - Share price increase cited as another winner from the buildout. Seagate stock move: $84 to $254 - Storage demand tied to video and AI data creation was highlighted. SanDisk stock move: $48 to $126 - Another storage-related beneficiary of the AI boom. Unitree robot dog starting price: $3,000 - Lower-end price for the robot dog discussed as part of robotics hardware progress. Unitree advanced robot dog price: $8,000 - Price of the more advanced robot dog with an attachable robotic arm. Unitree humanoid price: $16,000 - Price for Unitree’s humanoid robot, presented as surprisingly low.

Pivotal Quotes: "“If it’s not a bubble, it will be. And it probably is already.”" — James Van Gielen: On whether AI resembles historical speculative booms; he distinguishes the existence of a bubble from where the cycle is located. "“We’re building machine God.”" — James Van Gielen: Describing the internal logic and almost religious urgency driving hyperscaler spending and risk-taking. "“The constraint is still power... they’re just going to build their own power plants, they’re not going to rely on the grid.”" — James Van Gielen: On why behind-the-meter generation is becoming the decisive bottleneck and design choice for AI data centers.

Implications: The AI boom now hinges on real-world bottlenecks—especially power, financing, and industrial supply chains. Investors may find more durable exposure in infrastructure and equipment suppliers, but the cycle could still turn sharply if capex, credit, or electricity supply falters.