Winning the AI Race – The AI-Industrial Awakening

As someone building AI agents, I’m excited to see policy take the physical requirements of AI seriously. Software still depends on power, chips, factories, and people who know how to build and operate them. I want more room for experimentation and more opportunity for people to learn valuable new work.

The July 2025 America’s AI Action Plan organizes its proposals around innovation, infrastructure, and international diplomacy and security. It is a policy agenda. Publishing it does not establish that the proposed capacity, jobs, or deployment outcomes have been delivered.

I support the emphasis on building. The useful question is which actions reduce delays while preserving the checks that protect people and make projects dependable.

Data centers need reliable electricity, grid connections, cooling, and equipment. Nuclear power can contribute to that supply. But the earlier account of U.S. nuclear retirements placed too much explanatory weight on post-Fukushima fear.

At San Onofre, the NRC documented unexpected steam-generator tube degradation; the operator announced permanent retirement in June 2013. That plant’s history should not be reduced to a generalized political motive. Each retirement has its own technical, economic, and regulatory circumstances.

Likewise, comparing national additions of electricity capacity requires the same year, technology categories, and units. Nameplate capacity is not delivered energy or dependable capacity during a local peak. The earlier unmatched capacity figures and projections are removed.

The All-In conversations around this plan pushed me to think beyond software. Chris Power’s argument about manufacturing capability resonated with me: skilled workers, equipment, and supply networks take time to develop. I read that as an argument for rebuilding capability, not proof of every number used to illustrate it.

The earlier five-versus-1,784 ship comparison was not a defensible like-for-like comparison. It also incorrectly called the ratio 200 times; the arithmetic alone would be 356.8. I have removed the comparison rather than substitute a more dramatic number with unmatched ship categories and time periods. The unsourced munitions, factory-output, and workforce counts are also removed.

Podcast commentary can help frame questions. Precise quotations and timestamps need verification against the recording before they are used as evidence. This revision removes quotation-style paraphrases and unverified quotations rather than attributing new wording to speakers. At the time of the July 2025 discussion, JD Vance was Vice President, not a U.S. senator.

The idea that someone from a different occupation could learn to operate advanced manufacturing equipment is what connects this debate to my interest in AI and education. The outcome I want is more people gaining the ability to contribute. A company’s hiring story or expansion plan cannot by itself establish economy-wide job creation.

A serious objection deserves a place here: rapid infrastructure construction can shift costs and disruption onto communities, while automation can displace workers before new roles are accessible. Training has to connect to actual work, and infrastructure proposals should explain who pays and who benefits. These are implementation questions an ambitious policy should answer.

I still want us to build with urgency. That conviction is stronger when it rests on comparable evidence and concrete opportunities, rather than a stack of impressive but unverified numbers.