AI Chooses Winners

Liberty Energy’s former boss Chris Wright, now US Secretary of Energy, has likened the rollout and dominance of American AI to a new Manhattan Project. After listening to four power / AI experts in the Liberty + Energy Speaker Series on AI, my assessment is that the AI project is harder. The Manhattan Project was lucky to mostly rely on physics, which may be complicated, but is logical and consistent. The successful rollout of AI relies almost entirely on government & regulation - way more complicated, irrational and inconsistent.

After an introduction by Ron Gusek, four experts (Bill Barnes, Judith Judson, Sandra Hagen Solin, Will McAdams) moderator Gregory Thurnher discussed important criteria for data center site (state) selection:

• Choices / Options: As per Will McAdams, Texas has not politicized the grid and provides a la carte choices for generation. In this sense, a deregulated market for generation and retail with the natural monopoly of the grid, is a preferred market. In Texas, people can fire their generator and shop for another “gen-tailer” - an environment that strengthens the best.

• “Certainty”: Apparently, a regulated market is “guaranteed” - by rule of law - to build what it signs up to do without capital constraint. Projects may be delayed for a long time, but they will be built. Our colleague Roy Aune pointed out during questions that only about 1% of building projects in these markets are finished on time.

• Tax incentives: Everyone likes to minimize what they pay.

• And many other smaller things: availability of a fiber network, low price of power, the ability to bring onsite generation, the allowance to use thermal generation, the availability of natural gas, etc.

• Hedging bets: Mitigate risk by taking on a variety of projects so that most will succeed.

Several speakers highlighted just how great Texas and ERCOT were doing. The graph below shows current requests to connect new large loads to the grid. The 189 GW of generation capacity waiting in line through 2030 is about 2x ERCOT’s current peak demand, with 69% of the additions coming from data centers. Granted, some studies still have to be submitted and some of these will fail. But the number is huge.

And then there is Colorado… Colorado is missing the boat on AI, mostly for three reasons: (1) not providing tax incentives to compete with other states; (2) not providing options such as behind the meter generation; and, (3) accelerating Governor Jared Polis’ 100% renewable energy mandate.

The Governors Office (https://governorsoffice.colorado.gov/governor/) should realize these businesses will not operate on intermittency, the key feature of renewables. That’s why there is no graph like the one below for Colorado.