ENERGY ECONOMICS

Energy economics studies how electricity is generated, priced, and consumed at scale. For AI infrastructure, it has become a primary strategic discipline: data centers require enormous and reliable power, and the gap between grid capacity timescales (decades) and compute demand growth timescales (months) has made energy access a durable competitive moat.

The structural problem: timescales don't match

The electricity grid was built for relatively stable industrial and residential demand. Adding significant new load — like a 100MW hyperscale data center — requires transmission upgrades, substation builds, and generation capacity that can take 5–10 years to complete. AI compute demand is growing on a quarterly basis. This mismatch is not a temporary bottleneck; it is a structural feature of how electricity infrastructure is financed, permitted, and built.

The consequence is that energy access has become a genuine scarce resource. Companies that secured power purchase agreements and grid interconnection slots in 2020–2022, when AI demand was less visible, now hold assets that cannot be replicated quickly. That secured capacity is worth more than the power itself — it represents a multi-year lead time advantage over any new entrant.

How electricity pricing works

Electricity prices vary by location, time of day, season, and contract structure. Industrial consumers large enough to negotiate directly with utilities or electricity markets face a different pricing environment than residential customers. For data centers, the relevant metrics are: the cost per megawatt-hour for baseload power, the reliability of that power (measured as uptime percentage and frequency of interruptions), and the carbon content of the generation mix (increasingly relevant for corporate sustainability commitments).

Regions with cheap, abundant power — the Pacific Northwest (hydropower), parts of Texas (wind), Iceland (geothermal), and Scandinavia (mixed renewables) — have become magnets for data center investment. The economics are straightforward: a 100MW facility running at 90% utilization consumes roughly 788,000 MWh per year. A $10/MWh difference in electricity cost translates to $7.9 million per year in operating cost difference. At scale, these differences are decisive.

The carbon dimension

AI companies have made public commitments to renewable energy that create a secondary constraint beyond raw cost. Training large models on fossil fuel-heavy grids generates significant carbon emissions. Microsoft, Google, and Amazon have all made commitments to match their electricity consumption with renewable energy certificates (RECs) or power purchase agreements (PPAs) with renewable generators.

The economic effect is that data centers increasingly compete for renewable energy capacity, which is itself constrained by permitting timelines and transmission infrastructure. In markets where renewable capacity is tight, this competition bids up PPA prices and creates a premium for facilities with existing clean energy contracts.

Nuclear as the emerging answer

Several major AI infrastructure investors have moved toward nuclear energy as a long-term solution to the reliability and carbon problems simultaneously. Nuclear provides always-on baseload power with no carbon emissions, at a cost that is predictable over long contract periods. The revival of interest in small modular reactors (SMRs) is driven partly by AI demand — the promise of factory-built, faster-to-deploy reactors that can be co-located with data centers without the transmission infrastructure requirements of remote renewable generation.

Whether SMRs will deliver on their timeline and cost promises remains to be seen. But the strategic logic is sound: for a company with a 20-year infrastructure commitment, securing a dedicated nuclear power source eliminates both the grid reliability risk and the carbon exposure in one move.

What energy economics means for competitive strategy

Energy access is now a first-order strategic input for any company building large-scale AI infrastructure. The companies that treated it as a utility procurement problem — solved once and revisited only when contracts expire — are structurally disadvantaged relative to companies that treated it as a strategic asset to be secured, developed, and in some cases built from scratch.