The power names, scored against a paced frontier

Hey {{first_name|Investor}} -

Two things landed 8 days apart. Together they raise one question worth answering properly.

On August 14, Thiel Macro disclosed eight positions, seven of them in energy and power. On September 12, Dario Amodei called for slowing the pace of frontier AI capability, and Sam Altman and Elon Musk agreed with him. Chip stocks fell the following Monday, Nvidia by 3% and Intel by 6% (Los Angeles Times).

So: if the frontier gets paced, does the power trade still work?

My answer is yes, and for a narrow reason: pacing aims at training, while the load these companies have already signed is contracted for years. Here is that answer with the work shown. I built a card, scored every power name in that filing plus the ones Coatue and Sands hold, and wrote down what would change my mind on each.

A pacing regime, if it ever arrives, would constrain the largest training runs. It proposes nothing about models already deployed, and nothing about the electricity a data center draws while serving customers.

That splits power exposure into two very different kinds of revenue.

Revenue that needs the next frontier model to be bigger than the last one. Speculative capacity, merchant generation aimed at a campus no one has signed, equipment orders that depend on the next CapEx cycle being larger.

Revenue that gets paid whether or not the frontier moves. A 20-year power purchase agreement. A regulated utility's return on assets already in the ground. A campus under construction with a signed service agreement behind it.

The first kind is exposed to pacing, to a CapEx pause, and to the "AI is slowing down" headline that moved chips on September 14. The second kind is exposed to execution and regulation.

1. Contracted or forecast? A signed agreement with a named counterparty beats a pipeline number in a slide. Ask what kind of contract, because they are not equal: AEP describes its 63 GW as contracted, and that total mixes Electric Service Agreements with Letters of Agreement, which carry different protection (AEP investor handout, May 2026).

2. Does the revenue need the frontier to keep moving? If the answer is yes, a pacing regime is a real risk. If it's deployment load under contract, pacing is noise.

3. Who carries the build risk? A regulated utility earns a set return on what it builds, with regulators as the gate. A merchant generator carries market risk and gets paid by contract. A first-of-a-kind developer carries everything.

4. How good is the counterparty, and how long is the term? Twenty years from a hyperscaler is a different asset than a letter of intent from a developer.

5. What's the gate, and when does it clear? Zoning votes, interconnection queues, safety reviews. Every name here has one, and the date is usually public.

Then one number on the other side: how much of the company's earnings this actually touches. A perfect card on 5% of revenue moves a stock far less than a decent card on all of it.

A worked example, on the name I'd expect to score highest. Vistra ($VST) is Thiel's largest power position at 14.1% of the filing, and the fund came back to it with about 373,000 shares after exiting roughly 209,000 in Q3 2025. On January 9 it announced 20-year agreements with Meta covering 2,609 MW across Perry, Davis-Besse and Beaver Valley, with deliveries starting late 2026 and ramping through 2034. That sits alongside a 20-year agreement with Amazon Web Services for 1,200 MW from Comanche Peak, delivery from Q4 2027. Card: contracted, deployment-driven, merchant with contracted offtake, investment-grade counterparties at 20 years, and the gate is plant performance and uprate execution. That's four clean answers and one operational question, which is as good as this group gets.

Two of the names score badly on the question the headlines skipped. One of them is the smallest position in the fund.

The rest of this issue is for Insiders.

Below the line: the full scorecard on all 12 names, with the one line I check on each and the signal that would end my interest. Why the equipment makers carry more pacing risk than the chipmakers do, which is the opposite of how the market traded it. What X-Energy's card says against its story. What would prove me wrong, and the dated calendar I'm working from.