Rethinking the AI Power Bill
Innovative AI business models may head off the AI energy crisis
The Constraint in Plain Sight
Every major infrastructure buildout in modern history has had a constraint hiding inside it — one that was never really external, even though it may have felt that way. The railroads needed land. The first commercial internet wave needed bandwidth. In both cases, the constraint was always embedded in the economic architecture from the beginning. The people building the thing just didn’t treat it that way. They built first and reckoned with the constraint when it arrived, which it always did, and always at a greater cost than if considered sooner.
We are watching this pattern again, in real time, with AI and energy.
Salesforce Ventures recently published a rigorous analysis of what that constraint looks like right now. Every new model, every inference run, every data center expansion is an energy demand event, and the U.S. grid wasn’t built for this moment. Their essay maps five generation technologies — renewables, batteries, geothermal, nuclear fusion, fission — positioned to meet the demand, each at a different stage of maturity, each with a different timeline and risk profile. It’s exactly the infrastructure thinking the moment requires.
Their conclusion: energy innovation is AI infrastructure innovation.
That framing matters. It means the energy question has moved from operational concern to strategic foundation. The companies treating it that way from the beginning are building cost structure, sustainability, and competitive position into the same early decision — before scarcity forces anyone's hand.
The Inversion
The conventional model for AI infrastructure companies is legible enough: identify the compute demand, build the infrastructure to serve it, then secure the energy required to run it. Find the electricity, lock the contracts, manage the cost. Reasonable sequence. Almost everyone follows it.
Crusoe didn’t.
Instead of starting with compute and scrambling for power, they began by securing stranded power assets — including renewables sitting outside the grid or at its edges — and built their AI infrastructure around those assets. Breaking from grid-bound convention isn’t primarily a financing innovation or an operational efficiency. It’s a different theory of what the company is, arrived at before the constraint became expensive.
Energy isn’t a resource Crusoe consumes to run their product. It is the foundation the product is built on.
What I find elegant about this is that the design choice does two things simultaneously. It locks in a structural competitive advantage — a different cost structure, a different sustainability profile, a different position as energy scarcity tightens. And it produces a business model that is honest about what it’s building toward. The impact isn’t layered on. The efficiency isn’t retrofitted. Both emerge from the same early decision about what goes in the foundation.
The constraint that will squeeze everyone else is already inside its architecture, already answered.
That’s not just good operations. It’s good design.
The Opportunity Inside the Constraint
The SFV piece is optimistic, and the optimism is earned. AI is both the cause of the energy demand problem and the most powerful tool available to solve it. The same technology driving unprecedented load on the grid is accelerating materials discovery for fusion, optimizing distributed energy resources, improving grid integration for renewables, and enabling predictive management across the entire supply chain. Five generation technologies, each at a different point on the maturity curve, each advancing faster because of AI.
The constraint is large. So is the opportunity it’s created.
What founders and investors do with that opportunity in the next few years will substantially determine what the AI energy problem looks like a decade from now. The companies being built today — the ones deciding right now whether energy shapes their foundation or gets bolted on later — are making decisions that will compound across the entire infrastructure buildout. The Crusoe model is one answer to what that looks like. There are others waiting to be built.
SFV is right to shine a light here. The opportunity space they’re mapping is real, the technologies are moving, and the capital is following. What the analysis points toward, and what I’d add to it, is that the most durable companies in this space won’t just have access to the right generation technologies at the right moment. They’ll have designed around the constraint early enough for that access to be structural rather than circumstantial.
The founders and investors who understand that distinction are the ones who will look back at this moment as the window when the right decisions were still available to make.
Inspiration:
Claudine Emeott, Enki Toto, and Adrianna Alterman, “Energy Abundance in the AI Era,” Salesforce Ventures, April 22, 2026. salesforceventures.com/perspectives/energy-abundance-in-the-ai-era/