The Energy Model

Flare gas capture

The original Crusoe model. At oil wells, natural gas often comes out alongside the oil. If the gas can be piped to a market (a natural gas processing plant or pipeline), it has economic value. If it can't — because the well is too remote or the pipeline infrastructure doesn't exist — the gas is typically flared (burned off on-site) or vented (released as methane).

Flaring is wasteful (the gas's energy content is destroyed) but better than venting (methane is a much more potent greenhouse gas than CO2 in the short term). Both are operationally and regulatorily problematic.

Crusoe's mobile generators sit at the well-pad, capture the would-be-flared gas, convert it to electricity, and use that electricity to power containerized computing equipment also at the well-pad.

The advantages:

• The oil producer pays Crusoe (or pays them in kind by giving them gas for free) because Crusoe reduces flaring — a regulatory and ESG liability.
• The electricity cost basis is effectively zero (the producer values the gas at zero or negative).
• The greenhouse-gas footprint is meaningfully better than venting and somewhat better than flaring (combustion in a generator is more complete than open flaring).
• The compute happens at the energy source, eliminating transmission losses.

Behind-the-meter siting

The broader generalization: locate the compute behind the meter of any cheap, otherwise-stranded power source. Examples beyond oil patches:

• Wind farms with curtailment (more electricity available than the grid can absorb).
• Hydroelectric power in remote locations.
• Solar arrays with overcapacity.
• Industrial co-generation facilities with excess heat-to-power.

The common pattern: the energy producer can't get a good price for the electricity because of transmission constraints or local oversupply. A compute load that can colocate solves both their problem (taking the otherwise-wasted electricity) and Crusoe's (cheap power).

Stranded power sources

"Stranded" in this context means power that's produced but can't easily reach buyers. The reasons vary:

• Geographic isolation. Generation site is too far from population centers for transmission to be economic.
• Intermittency. Wind and solar produce variable output that the grid can't always absorb.
• Regulatory friction. Some power can't be sold across certain market boundaries.
• Time-of-day mismatches. Power produced at off-peak times sells for less.

Crusoe's compute load is flexible enough to absorb most of these — datacenters operate 24/7, so cheap night-time power is just as useful as cheap day-time power.

Energy economics

For an AI workload, electricity is a meaningful fraction of total cost:

• An H100 server draws ~10 kW continuously.
• At $0.10/kWh, that's $1/hour in electricity per server (~8 GPUs).
• At $0.02/kWh (cheap stranded power), that's $0.20/hour.
• The energy cost gap is $0.80/hour per server, or about $7,000 per server per year.
• Over a 3-year hardware lifetime, that's $21,000 per server in energy-cost advantage.

Multiply by tens of thousands of servers and the strategic advantage is measured in hundreds of millions of dollars per year. Even after accounting for the operational complexity of behind-the-meter siting, the energy gap is the structural advantage Crusoe leans on.

Why this is the moat

Energy advantage is harder to replicate than software or hardware advantage:

• Behind-the-meter siting requires relationships with energy producers — relationships that take time to build.
• Permitting and grid-interconnect approvals are slow.
• The specific knowledge of where flare-gas or stranded-power opportunities exist is non-trivial.
• The operational know-how to run datacenters in remote, oil-patch environments is unusual.

Competitors paying grid rates structurally can't match Crusoe's energy cost basis. The advantage is real and durable in the short-to-medium term.

Limits of the model

The energy model has real limits:

• Available stranded energy is finite. The total flare-gas capacity in the Permian Basin is in the low GW range, not hundreds of GW. Once captured, the resource is consumed.
• Site quality varies. The cheapest energy sites are often the worst for datacenter operations (remote, hot, difficult to build at).
• Customer concerns about siting. Some customers don't want their training runs happening at oil-patch datacenters for ESG-narrative reasons.
• Power demand outgrows opportunity. Frontier AI training needs hundred-MW-plus buildouts; stranded sources rarely provide that much capacity in a single location.

Evolution into broader power strategy

As Crusoe has scaled, the company's power strategy has broadened beyond pure flare-gas capture:

• Dedicated natural gas generation in oil-and-gas regions.
• Power-purchase agreements with renewable producers.
• Behind-the-meter relationships with industrial sites.
• Standard grid power in some configurations where scale demands it.

The "we burn flare gas" narrative has become less central than the broader "we have cheap energy through unique siting and procurement" narrative. The strategic advantage is still energy-cost-led, but the source mix has diversified.

Takeaway

The energy model is Crusoe's core strategic differentiator. The flare-gas thesis got the company started, but the broader behind-the-meter and stranded-power strategy is what scales it. The next chapter looks at the actual datacenter buildouts that emerged from this energy thesis.