Long Strange Trip

Chase Lochmiller on Why AI Compute Is an Energy Problem

Chase Lochmiller· Co-founder and CEO of Crusoe at Crusoe
·~61 min·English·Sequoia Capital
AI InfrastructureGPUTrainingBusiness StrategyAI Company
TL;DR

Crusoe's Chase Lochmiller argues that AI's real scaling limit is energy, and lays out how the company vertically integrates power, data centers, and cloud into an AI supermajor modeled on oil and gas.

01Core Mental Model

Compute Becomes an Energy Problem

At scale, AI's binding constraint is not chips or data but energy, and Lochmiller's whole thesis starts by treating compute as an energy problem.

if data and compute were the bottlenecks, ultimately compute becomes an energy problem.

— Chase Lochmiller, Long Strange Trip
Key Insight
The framing flips the industry's scoreboard: if the binding constraint is joules, then whoever controls cheap, dispatchable power, not just the best model, sets the ceiling on how far scaling can go.
02Origin Thesis

Bring the Market to the Energy

Instead of moving energy to where demand is, Crusoe moves demand to stranded energy, so flared gas that nobody could sell became power for on-site compute.

Instead of trying to bring that stranded energy to a market where it can be bought and sold, bring the market to the energy.

— Chase Lochmiller, Long Strange Trip
Key Insight
Moving compute to stranded energy turns a disposal cost (producers paid to flare the gas off) into revenue, which is why the model already worked as a business before any AI demand existed.
03The Strategy

The AI Supermajor

Crusoe is built like an oil and gas supermajor, owning energy generation, data centers, and cloud in one vertically integrated stack from the electron to the token.

we're trying to build an AI supermajor in the same way that Exxon and Chevron are supermajors in the oil and gas industry.

— Chase Lochmiller, Long Strange Trip
Key Insight
Vertical integration is not just scope, it is a built-in hedge: Lochmiller notes Exxon does not hedge its oil because when prices fall, upstream wellhead margins compress while downstream refining margins expand. Owning the whole stack from power to cloud gives Crusoe that same kind of natural balance across layers, steadier than betting on any single one.
04Technical Nuance

AI and Bitcoin Draw Power Differently

Bitcoin mining can hold a flat power draw and adjust it to demand, while AI workloads and cooling create more variable power needs, so the shape of the demand changes how you build the data center.

you can almost give it like a fixed draw amount of power and say like stay at that power level and just like consume that. And it will be like very flatline.

— Chase Lochmiller, Long Strange Trip
Key Insight
AI's variable, spiky load, with training bursts and the daily PUE swings Lochmiller describes (chillers working harder in the day heat than at night), is more variable to serve than Bitcoin's steadier draw, so an energy-first builder has to design for peaks and seasonality, not a flat baseline. The power profile, not just the megawatt total, shapes the facility.
05The Timing

The Magic Moment Was ChatGPT

Energy-first was never a pivot: ChatGPT arrived and made the infrastructure Crusoe had been quietly building into exactly what the market suddenly demanded.

really, you know, the launch of ChatGPT. So, November 30th, 2022, I think that was just like the moment that changed everything

— Chase Lochmiller, Long Strange Trip
Key Insight
The deeper lesson is not foresight but sequencing: an existing cash engine, Bitcoin mining on stranded gas, de-risked the infrastructure bet, so the company could wait for demand to show up instead of betting the firm on when it would.
06On Competition

Suicide, Not Homicide

Startups almost always die from self-inflicted failure, not from competitors, which is why Lochmiller spends his worry on execution and culture rather than on rivals.

most startups die from suicide, not from homicide. It's not other businesses that kill you.

— Chase Lochmiller, Long Strange Trip
Key Insight
If competitors rarely kill you, guarding against them is wasted energy, and in a non-winner-take-all market like infrastructure a rival is often a future customer or partner, which is why Lochmiller says Crusoe will partner with the very firms it competes with, the way oil majors share a pipeline with rivals.
07On Risk

Think Like a Mountaineer

Lochmiller's Everest climb became Crusoe's model for risk: separate the risk you can control from the risk you cannot, then prepare hard for one and cut your exposure to the other.

it's like a serac falls or an avalanche occurs and, you know, oops, you die.

— Chase Lochmiller, Long Strange Trip
Key Insight
Splitting controllable from stochastic risk is a general operating principle, not a climbing tip: you spend preparation where it changes the odds and cut exposure where it does not, which is how a capital-intensive, multi-year buildout survives shocks it can never forecast.