Odd Lots Podcast Summary

Episode: James Van Geelen on the Next Phase of the AI Buildout

Date: October 6, 2025
Hosts: Joe Weisenthal & Tracy Alloway
Guest: James Van Geelen, author of the Citrini newsletter

Overview:

This episode dives deep into the physical and financial realities of the ongoing artificial intelligence (AI) buildout in the U.S., with a particular focus on the rise of massive data centers and the immense investments funneling into the sector. The conversation explores the "picks and shovels" phase, unpacking which companies and industries are quietly powering the AI gold rush, how capex and financing deals are evolving, and whether we're experiencing a true technological revolution or an inflated bubble.

Key Discussion Points & Insights

1. Scaling Up: Touring Hyperscale AI Data Centers

• Impressions from the Field

  • James recounts using drones and satellite images to observe the sheer magnitude of new data centers, such as Stargate in Abilene, Texas.
  • Quote: "A year ago this was dirt, and now there’s this thing and it’s the size of lower Manhattan." – James Van Geelen [06:32]
  • These builds are likened in scale and effort to post-WWII infrastructure projects, far surpassing former industrial benchmarks.

• The Role of Natural Gas and Power Needs

  • Operators are building their own natural gas power plants on-site to ensure reliability and energy capacity.
  • The turbines are simple cycle (rather than higher-efficiency combined cycle) because advanced models have 7-year wait times.
  • Quote: "Without power, these are just hunks of metal." – James Van Geelen [07:57]
  • Suppliers like GE Vernova and Caterpillar are key beneficiaries due to the scale and hardware demand.

• On-Site Labor & Construction Scale

  • Each hyperscale build is attracting waves of skilled trades: HVAC, fiber, electrical, and plumbing workers.
  • Quote: "There are 7,000 people working on building this. And this is just one site." – James Van Geelen [09:39]

2. Investment Implications: The Picks and Shovels of AI

• Industrial Winners

  • "Old line" industrials like Caterpillar, Mitsubishi, Eaton, and Siemens Energy are suddenly red-hot stocks, often overlooked as AI plays.
  • Tracy jokes about a "Joe Hyperscale ETF," buying the companies that literally supply every nut and bolt.

• Unprecedented Demand & Company Confessions

  • Some vendors can only supply a portion of the parts needed—demand is outstripping even the world’s biggest industrial players.
  • Quote: "The pie is so big...the biggest companies in the world...can only do 60% of it." – James Van Geelen [13:03]

• New Metrics: Measuring by Megawatts

  • The size of data centers is discussed in megawatts/gigawatts, not square footage.
  • Liquidity cooling advances have reduced water needs, shifting the constraint to power, not water.

3. The Bubble Question: Bubble or Paradigm Shift?

• Echoes of the Past:

  • The hosts compare today’s buildout to previous bubbles, especially the dot-com era: infrastructure leads, followed by speculative excess.

  • Quote:
    "If it’s not a bubble, it will be, and it probably is already. But where are we in the bubble? That’s the interesting thing for investors." – James Van Geelen [21:09]

• Contrasts to the Dot-Com Era:

  • Unlike "dark fiber" left dormant in 2000, every data center coming online is immediately used to capacity—no slack exists.
  • Demand keeps surging (e.g., the rise of compute-intensive video models like OpenAI’s Sora 2).

• Circular Financing and Bubble Dynamics

  • Intricate financing deals: off-balance-sheet infrastructure, ABS-backed by GPUs, debt/equity structures.
  • Nvidia is itself financing customers to buy its GPUs, creating a potentially risky circular system.
  • Quote: "That’s a bubble dynamic. That’s visible circular financing. But the demand is real." – James Van Geelen [24:17]

4. Risks, Warning Signs, and Strategies

• Possible Early Warning Indicators:

  • Used GPU prices dropping

  • Financing waivers or renegotiations in private credit

  • Delays or issues with power delivery and supply

  • Slowdown in revenue growth for AI companies

  • Quote: "Monitoring used GPU prices...is a great way to see what’s going on." – James Van Geelen [29:35]

• Protecting Against the Bubble:

  • Suggests "long old-line physical hardware, short private credit/financiers" as a risk-balanced investment approach.
  • Quote: "The companies actually building it have unlimited upside, and...old line industrials...still have a core business to go back to." – James Van Geelen [31:13]

5. AI Use Cases: Video, Robotics, and Storage

• “Slop Factories” or Strategic Business?

  • Much of AI’s current output (e.g., endless video generation) is seen as commercial revenue drivers and vital stepping stones toward AGI, not just frivolous content.

  • Video models are essential for robotics, which in turn generate the training data to further AI.

  • Quote: "Video models are going to become the most important part of this… becomes extremely important for robotics." – James Van Geelen [32:21]

• The Forgotten Boom: Data Storage

  • Demand for storage is soaring as video models produce vast volumes of data.

  • Stocks like Seagate and SanDisk are surging, driven by underestimated AI-linked demand.

  • Quote: "Basically one of the only areas of semiconductors that has gotten ignored has been storage… and if we’re going to start throwing out the slop on video [storage demand will jump]." – James Van Geelen [34:28]

6. Robotics: Field Trips & Real-World Training

• China’s AI Robotics Push

  • James describes visiting Unitree in Hangzhou, China, where robot dogs and humanoids are mass-produced for practical deployment and data gathering.
  • Quote: "That’s the competition we have stacked up against us in the US vs. China." – James Van Geelen [37:33]
  • These units are advanced, cheap, and being deployed widely to generate the real-world movement and interaction data needed for future robotics.

• Consumer Pricing

  • Unitree robot dog: starts at $3,000
  • Advanced models: $8,000
  • Humanoid robot: $16,000 (cheaper than a used car)

Notable Quotes & Memorable Moments

• On Hyperbole and Belief:

  • "The party line is we’re building machine God—and if we’re building machine God, then that’s kind of a religious zealotry." – James Van Geelen [19:52]
  • "Three years ago, most of these hyperscalers were making net-zero by 2030 pledges. That is not happening… now, you’re using natural gas and then say, once we create machine God, machine God will fix it." – James Van Geelen [20:06]

• Bubble Humor:

  • "Are we in a bubble? Yes. Is the bubble over? There are good arguments for that. I don’t think…they’re leaning into this slop… it’s a company—they’re going to do whatever they can to make money. That’s capitalism." – James Van Geelen [33:44]

• Market Manias:

  • Joe live-checks tickers throughout, illustrating the speed and pervasiveness of the AI-driven stock surge:
    • "Caterpillar was a $334 stock… now 486 at an all-time high." [11:17]

• The Macroeconomic Question:

  • "I wonder about the crowding out possibility…what if you had another business in Abilene and you’re trying to hire an HVAC worker right now? There’s no hope—all the HVAC workers are at that big Stargate operation. This is a very interesting macro question…" – Joe Weisenthal [42:09]

• On Market Timing:

  • "It’s not my job to time the market. So I’m glad I don’t have to have a view on all of this stuff." – Joe Weisenthal [39:53]

Timestamps for Important Segments

• Opening banter about data center field trips: [02:01–03:10]
• Discussion of Abilene & Hyperion data centers: [06:04–07:52]
• Details on power generation (gas turbines, suppliers): [07:54–08:53]
• Labor & construction scale: [09:35–10:13]
• Market dynamics: "picks and shovels" stocks: [10:23–13:35]
• Scarcity, location, & utility constraints (power, water): [13:35–16:22]
• Physical security & building standards: [19:23–20:44]
• Bubble theory, history & dynamics: [21:00–25:12]
• Financing details & circular deals: [23:45–26:18]
• Red flags for AI buildout slowdown: [29:24–30:38]
• Hedging AI exposure in markets: [31:02–31:30]
• The strategic importance of video models & storage: [32:21–35:09]
• Unitree robotics segment: [36:24–38:49]
• Macro ‘crowding out’ effects of AI spending: [41:50–42:40]

Conclusion & Takeaways

• Physical AI buildout is massive, real, and still expanding.
• Energy and infrastructure providers are huge, underappreciated winners.
• Financial innovation in the sector is both enabling rapid growth and adding systemic risk—watch for cracks.
• Signs of froth are everywhere, but end demand for compute and video data is as real as ever.
• Even as companies pursue AGI with religious fervor, physical limits—to power, capital, labor—are shaping the new tech economic landscape.

This episode is a must-listen for anyone tracking the intersection of technology, industrial stocks, infrastructure, and financial innovations driving the next phase of the AI revolution.