The Twenty Minute VCCrusoe CEO: Why Everyone Gets GPU Depreciation & AI Energy Costs Wrong
At a glance
WHAT IT’S REALLY ABOUT
Crusoe CEO on AI factories: energy bottlenecks, GPUs, and misinformation
- Chase Lochmiller explains Crusoe’s thesis that AI infrastructure will decentralize toward regions with abundant low-cost power, making energy the core bottleneck behind compute growth.
- He describes today’s true infrastructure constraint as “places to plug in GPUs,” with shifting supply-chain bottlenecks that Crusoe addresses via deep vertical integration and internal manufacturing to accelerate build timelines.
- Lochmiller argues key public narratives about AI data centers—especially around water use and rising electricity prices—are often misinformation, citing closed-loop cooling and the possibility of electricity prices declining as generation investment increases.
- Crusoe’s business model is to monetize across three products—data centers, GPUs, and tokens—creating an “AI supermajor” hedge as margins rotate across the stack over time.
- He contends investors misunderstand GPU depreciation and AI energy economics, believing older chips can remain valuable longer when packaged into managed services and sold as lower-cost intelligence.
IDEAS WORTH REMEMBERING
5 ideasEnergy—not chips alone—is the binding constraint on AI scale.
Crusoe frames AI compute as an energy problem: as models scale, energy availability and price become the limiting factors. This drives their thesis that the winning infrastructure will be built where power is abundant and cheap, not necessarily in legacy data-center hubs like Northern Virginia.
AI data-center supply chains behave like “Whac‑A‑Mole,” so vertical integration buys speed and certainty.
Lochmiller describes bottlenecks shifting across components (e.g., power distribution centers with 100-week lead times). Crusoe’s answer is vertical integration (including electrical manufacturing) to compress timelines and avoid being stuck behind supplier queues.
The loudest critiques (water use, energy prices) are often directionally wrong for modern builds.
He argues community concerns are often driven by misinformation: modern “AI factory” designs can be near-water-neutral via closed-loop cooling, and data centers can correlate with lower local electricity prices by catalyzing new generation and spreading grid fixed costs over more load.
“Exxon of AI infrastructure” is a portfolio/hedging strategy, not just a branding line.
Crusoe aims to monetize across three layers—data centers, GPUs, and tokens—analogous to an oil & gas supermajor. The idea is that where margin accrues will rotate over time, so being present across the stack creates a natural hedge against commodity-like pricing swings.
Revenue design is a portfolio of contract durations and risk profiles, not one bet.
They mix long-term, creditworthy take-or-pay GPU rentals with shorter-term, higher-margin managed services (managed inference, serverless fine-tuning). This creates blended returns while keeping upside exposure to spot shortages and new product value.
WORDS WORTH SAVING
5 quotesThe infrastructure to support AI wasn't gonna be centralized, it was gonna be distributed where energy was low cost and abundant.
— Chase Lochmiller
Where it's actually manifesting is there are not places to plug in GPUs, so that's ultimately the supply constraint.
— Chase Lochmiller
The data has shown that d- energy prices actually come down. It's actually the opposite of the narrative that's being told.
— Chase Lochmiller
Look, I, I think when we started to make really substantial purchases with, with Hoppers, um, and this was in, you know, 2023, the feedback we got was like, "Well, we don't even know if this is gonna be valuable after year three," right? Here we are three years later, and the prices being charged for utilizing Hoppers is higher than the rates that were being charged three years ago when they were, when they were brand new.
— Chase Lochmiller
I, I think what I've changed my mind on is, like, most moats are a illusion. Most moats don't exist.
— Chase Lochmiller
High quality AI-generated summary created from speaker-labeled transcript.
