Overview of 20VC with Chase Lochmiller
In this 20VC episode, Harry Stebbings speaks with Chase Lochmiller, cofounder and CEO of Crusoe, about the future of AI infrastructure, the economics of data centers, and why energy, labor, and speed are becoming the real bottlenecks in the AI race. Lochmiller explains Crusoe’s vertically integrated model across data centers, GPUs, and inference, argues that many common fears around data centers are overstated, and shares his view that most “moats” are overrated in fast-moving technology markets. The conversation also touches on his personal philosophy, including how mountaineering shaped his approach to leadership, resilience, and safety.
Key Themes and Takeaways
AI infrastructure is becoming the new strategic battleground
- The central thesis is that AI growth is constrained less by model ideas and more by physical infrastructure:
- power,
- land,
- labor,
- data center capacity,
- and GPU availability.
- Lochmiller argues that AI infrastructure will be distributed, not centralized, and will gravitate toward places with abundant, low-cost energy.
Crusoe’s model: data centers, GPUs, and tokens
- Crusoe sees its business as spanning three main products:
- Data centers
- GPUs / compute
- Tokens / inference
- This vertically integrated approach lets the company capture value across the stack, similar to an oil-and-gas “supermajor” spanning upstream, midstream, and downstream.
Speed and vertical integration matter more than traditional outsourcing
- A major example: Crusoe needed a power distribution center with a 100-week vendor lead time, but built it internally in 28 weeks.
- Vertical integration helps Crusoe:
- unblock supply chain bottlenecks,
- reduce dependence on scarce vendors,
- improve cost visibility,
- and move faster than competitors.
Data centers are often misunderstood
- Lochmiller strongly disputes several common criticisms:
- Water usage is often exaggerated; modern AI data centers can be effectively water-neutral or use very little water relative to the size of the project.
- Energy prices do not necessarily rise; in some communities, large data center investments can help bring more generation online and lower prices.
- He says many communities benefit through:
- jobs,
- tax revenue,
- local business activity,
- and long-term infrastructure investment.
The real bottlenecks: energy and labor
- Power is the biggest constraint on scaling compute.
- Labor is another major issue, especially skilled trades like electricians, welders, plumbers, and construction workers.
- Regulatory friction matters, but he frames policy as something to navigate rather than a fatal blocker.
GPU economics and depreciation are misunderstood
- Crusoe depreciates hardware over six years, which he says is standard.
- He argues people underestimate:
- how long GPUs remain valuable,
- how applications evolve to extract more value from older hardware,
- and how managed services can extend monetization beyond the raw hardware lifecycle.
- His point: even older chips can still generate economic value if paired with the right services and workload management.
Business and Market Views
Managed inference is a high-value layer
- Crusoe’s managed inference offering is positioned as a way to abstract away infrastructure complexity for customers.
- The most important performance metrics include:
- time to first token
- latency
- throughput
- GPU utilization
- He emphasizes that the GPU is often the most valuable asset in the data center, so maximizing usage is critical.
Open source vs. closed source
- Lochmiller believes:
- more dollars are currently flowing to closed-source frontier models,
- but more tokens are likely being generated on open-source models.
- He sees a future where companies use a mix of:
- frontier models,
- open-source models,
- and proprietary/domain-specific models trained on private data.
- He does not think one approach will dominate completely.
“Most moats are an illusion”
- One of the strongest statements in the interview:
- most moats don’t last,
- many are just temporary wrappers around fast-changing technology.
- He believes the real advantage is:
- speed,
- adaptation,
- and execution in a shifting market.
Leadership, Personal Philosophy, and Mountaineering
Mountaineering as a company culture
- Lochmiller says Crusoe explicitly embeds mountaineering into its culture.
- Lessons from mountaineering include:
- planning for change,
- having backup plans A, B, C, and D,
- prioritizing safety,
- enduring difficult conditions,
- and understanding that “getting to the top is optional, getting down is mandatory.”
His path from science to wealth creation
- In university, he expected to become a theoretical physicist.
- Over time, he realized academic discovery was slower-moving than he wanted.
- He then became motivated by the idea of building wealth and doing something faster-paced and more impactful.
- Starting Crusoe after some financial success gave him the freedom to take bigger risks.
Parenting and work-life integration
- He describes being highly intentional about family:
- red-eye flights to be present for bedtime,
- weekends dedicated to kids’ activities,
- and protected morning time when he is completely unavailable.
- His message: prioritization matters more than perfect balance.
What He Thinks People Get Wrong
About data centers
- People overstate:
- water consumption,
- energy price inflation,
- and negative community impact.
- He argues the facts show many communities actually benefit materially.
About AI demand
- He does not believe consumer demand is “cracking” in a meaningful way.
- Instead, he sees AI as still early and becoming more capable every year.
- In his view, the technology is already proving itself across:
- coding,
- scientific discovery,
- materials research,
- and knowledge work.
About policy and politics
- He’s uncomfortable with data centers becoming a political flashpoint.
- But he thinks the debate is emotional rather than fact-based.
- He believes the industry should focus on being a good local citizen and communicating the real economic benefits.
Notable Quotes and Ideas
- “Think like a mountaineer.”
- “Getting up to the top is optional, getting down is mandatory.”
- “Most moats are an illusion.”
- “We can sell data centers, we can sell GPUs, and we can sell tokens.”
- “The GPU is actually the most valuable thing in the entire data center.”
- “Time to token” is a critical metric for AI infrastructure.
- The future of AI infrastructure will be shaped by where energy is abundant and cheap, not just where legacy data centers already exist.
Bottom Line
This interview presents Crusoe as a vertically integrated AI infrastructure company built around a strong thesis: energy is the bottleneck to intelligence. Lochmiller argues that the winners in AI will be the companies that can move fastest, secure power, build infrastructure efficiently, and monetize compute across multiple layers of the stack. The conversation is both a defense of data centers and a broader argument that AI’s next phase will be defined less by model hype and more by hard industrial execution.