20VC: The Future of Datacentres: What You Need to Know | Why Everyone Gets GPU Depreciation and AI's Energy Costs Wrong | Who Really Makes Money From AI & Why Most Moats Don't Exist with Chase Lochmiller

Chase Lochmiller shares his journey from theoretical physics and quantitative finance to founding Crusoe Energy, explaining how mountaineering principles of resilience, planning for contingencies, and safety culture inform his business approach. He discusses Crusoe's evolution from Bitcoin mining monetization to becoming a comprehensive AI infrastructure company providing data centers, GPU capacity, and managed inference services.

On data center supply constraints, Lochmiller explains that the real bottleneck is not GPUs themselves but physical locations to deploy them. He details Crusoe's vertical integration strategy, including in-house electrical manufacturing, which enabled them to deliver 200+ megawatts of compute capacity in Abilene, Texas in one year versus competitors' 2.5-year timelines. The company designed custom power distribution centers internally because vendor lead times were 100 weeks, reducing this to 28 weeks through internal manufacturing.

Regarding infrastructure location, Lochmiller argues that AI workloads don't require centralized hubs like Northern Virginia because compute time dominates over network latency. He believes distributed data centers in low-cost, abundant energy regions represent the future. He addresses common misconceptions: data centers use minimal water (their 140MW Abilene facilities use about as much as 10 homes annually), and energy prices actually decrease in communities receiving data center investments as they catalyze new power generation capacity.

Lochmiller discusses Crusoe's three core revenue products: data centers, GPUs, and tokens (inference). The managed inference business currently has the highest margins due to supply scarcity. He uses an oil and gas analogy, positioning Crusoe as an "AI supermajor" vertically integrated across upstream (energy), midstream (infrastructure), and downstream (services), allowing margin flexibility across the value chain as commodity prices fluctuate.

On GPU depreciation, he challenges conventional wisdom that chips become worthless after three years. Hopper GPUs command higher rental rates today than when new three years ago. He attributes this to developer ingenuity in creating valuable applications. The company uses six-year depreciation cycles and plans to extend this through service-based monetization of older chips.

Lochmiller addresses political concerns about data centers, acknowledging legitimate anxiety about job displacement but arguing data has shown data centers create significant employment in skilled trades and blue-collar sectors. He emphasizes that communities see tangible benefits: Crusoe will represent a third of Abilene's tax revenue and has more than doubled school tax receipts.

On competitive moats, he states a significant shift in his thinking: most moats are illusions and ephemeral, especially amid accelerating technological progress. Success depends on speed and adaptability rather than sustainable competitive advantages. He estimates approximately 50% of planned data centers won't be built due to permitting, entitlements, land acquisition, and utility interconnection challenges.

Regarding open versus closed models, Crusoe's data shows users spend more money on closed-source frontier models but generate more total tokens using open-source models. He believes both will persist, with companies increasingly fine-tuning models on private data for domain-specific applications.

On work-life balance, Lochmiller prioritizes parenting through red-eye flights to avoid missing bedtimes, protecting morning hours with family as unreachable time, and being home most weekends for children's activities. He disagrees with concerns about wealth inequality, arguing that the value created by AI's abundant intelligence far exceeds the market cap concentration.