Energy, AI, and Markets

Part 2 of my podcast conversation with Doug Lewin

Photo by Ian Battaglia on Unsplash

Earlier this month, I joined Doug Lewin on his Energy Capital Podcast and we had so much to talk about that he broke our conversation into two podcasts! In part 1 we talked about markets, innovation, and the extent to which the regulatory framework and utility business model are outdated, and I riffed on it in a post two weeks ago.

Quarantining the Monopoly: A Conversation with Doug Lewin

In part 2 that Doug released last week, we discussed the intersection of electricity markets, risk, and the rapidly growing demand for data centers and AI. Our conversation spanned market design, the evolution of utility regulation, and the unprecedented energy needs of hyperscale computing. I’d like to reflect here on some of those themes and offer my perspective on why this moment feels different from past episodes of electricity demand growth.

Markets as Error-Correcting Mechanisms

One of the points I emphasized is that markets do not only allocate resources efficiently; they also serve as error-correction mechanisms. If investors build too many natural gas plants and can’t earn sufficient returns, that is a signal to redirect capital elsewhere. This feedback loop is vital, particularly in electricity systems, because the costs of misallocation are high and the consequences can be widespread.

But electricity markets are unusual. They are designed rather than emergent, “born in captivity” rather than arising spontaneously out of centuries of exchange (although when done thoughtfully, market design rests first and foremost on those market institutions that have evolved over centuries). Even so, one of their greatest innovations—the move to security-constrained economic dispatch—has delivered enormous benefits to consumers by ensuring that power is dispatched according to cost and system reliability constraints. Yet one area remains underdeveloped: markets for risk.

Missing Markets in Risk

Winter Storm Uri in Texas and Winter Storm Elliott in PJM underscored how poorly we allocate and price risk in power systems. Consumers bear outage risk and rate risk. Generators and fuel suppliers navigate opaque and incomplete arrangements, often falling back on force majeure. Meanwhile, utility shareholders are insulated from much of this uncertainty by rate-of-return regulation. In 2022 Jacob Mays and a group of co-authors, including me, wrote a paper digging into the problem of incomplete markets in risk in electricity with some suggestions for institutional change to improve it.

We need more complete markets for risk allocation. Allowing risks to be priced, traded, and borne by those most able to manage them would improve efficiency and resilience. This is not a new observation, but it becomes more urgent as we face new sources of demand stress in power systems. Risk management cannot remain an afterthought. It must be central to how we design the next phase of electricity markets.

The Demand Side Opportunity

Traditionally, demand for electricity was considered inelastic. Flipping a switch was a mechanical action, and no consumer wanted to monitor prices or adjust behavior manually. But digital automation changes that equation. Smart devices, sensors, and programmable controls make it possible for demand to respond in real time to system conditions without requiring constant consumer attention.

Transactive energy—where devices themselves respond to price signals—illustrates this potential. Your air conditioner, water heater, or refrigerator could automatically adjust operation within your chosen comfort band, guided by price signals, informed by your trigger prices that form your devices’ bids and offers in a local market, that reflect real system costs. This kind of flexibility unlocks latent demand response that is far less inconvenient than twentieth-century approaches. But for this to succeed, regulation must shift toward performance-based incentives, rewarding utilities and third parties for delivering measurable outcomes in reliability and resilience rather than simply for investing in assets.

Performance and Regulation

Here we might learn from the UK, where utilities operate under price-cap regulation rather than rate-of-return. Utilities are allowed to keep efficiency gains so long as they maintain defined service quality levels. Such performance-based models could align incentives more effectively in U.S. distribution systems, where regulated monopolies still dominate and consumers have little recourse when service quality falters. If utilities faced penalties for poor performance, measured in outage frequency, duration, or even wildfire prevention, customers would see real benefits.

The absence of these mechanisms has costly consequences. Consider Hurricane Beryl, where outages lasted a week for many Houston households, yet the utility’s shareholders did not bear commensurate financial consequences. Aligning incentives remains one of the most pressing challenges in regulation today.

Data Centers as a Demand Shock

This brings us to AI and data centers. The growth trajectory is staggering. The International Energy Agency estimated that global data center consumption in 2022 was 460 terawatt hours. By 2030, that figure could exceed 1,000 terawatt hours. McKinsey projects a 23 percent compound annual growth rate for U.S. data center energy demand between 2023 and 2030—equivalent to adding “a Texas worth” of electricity consumption in just a few years. For an industry where even 3 percent annual growth is notable, this is unprecedented.

What makes this moment different is the combination of magnitude and speed. Utilities are not built to add generation capacity in two years to match the construction cycle of hyperscale data centers. A gas turbine might have a seven-year supply chain; a transformer can take six years to procure. Utilities historically built slowly, prudently, and with regulatory oversight. That cadence cannot meet hyperscalers’ “speed to power” requirements.

Make or Buy in Energy Supply

This mismatch is forcing technology companies to confront the make-or-buy decision in power supply. Transaction cost economics, pioneered by Ronald Coase, helps us understand this. As transaction costs fall, firms find it easier to contract for services rather than vertically integrate. In ERCOT, where the wires network monopoly has been quarantined, hyperscalers can contract directly with power providers more readily than they can in other states. They can procure on-site solar and storage, strike long-term contracts for natural gas peakers, or even partner in geothermal and nuclear projects, as Google has done with Fervo and Kairos.

Data Center Electricity Use III: Make or Buy?

In other states, however, regulatory obligations to serve can leave even the largest customers captive to their utility. This tension will only grow sharper as AI infrastructure proliferates. Whether regulators allow greater flexibility for large customers to contract directly for power will shape the geography of future data center investment.

Innovation in Energy Procurement

Some companies are already innovating at the intersection of economics and environmental performance. Crusoe, for example, sites data centers in the Permian Basin and uses otherwise-flared natural gas as fuel. This turns wasted, polluting gas into productive energy. By aligning economic and environmental incentives, firms like Crusoe illustrate how markets can foster creative solutions that regulation alone would not design. It is an encouraging sign of how entrepreneurial approaches can meet new challenges.

From Capacity to Capability

Perhaps the most important conceptual shift we must make is from capacity to capability. Utilities and regulators are habituated to asking, “What is your peak demand?” Data centers provide the maximum number they might ever reach, even though they typically operate well below that. What matters is not capacity alone but the range of capabilities a customer can bring to the grid: flexibility, responsiveness, and the ability to adjust load dynamically in ways that benefit the system.

If we remain locked in a capacity mindset, we risk misallocating resources and overbuilding infrastructure. A capabilities approach, by contrast, recognizes that demand can be a resource in itself. This reframing will be essential if we are to integrate hyperscale computing into electricity systems without destabilizing them.

A Moment of Transformation

The rise of AI and hyperscale data centers is not just another episode of demand growth. It is a stress test for our market designs, regulatory frameworks, and investment models. If we get it right, we can harness these demands to spread system costs, drive innovation in distributed energy resources, and accelerate the transition to a more flexible, resilient, and decarbonized grid. If we get it wrong, we risk reinforcing rigidities, misallocating risks, and slowing the very innovation that could help us adapt.

Markets are not, and cannot be, perfect, but they are powerful tools for discovery, coordination, and error correction. At this point, we should lean into those strengths. By rethinking risk allocation, embracing performance-based regulation, and recognizing capabilities rather than simply capacity, we can position electricity systems to meet the challenges of AI, data centers, and beyond.