From James Pethokoukis at AEI, June 11:
To fully exploit the potential of an emerging AI revolution, America is also going to need an energy revolution to power digital infrastructure, especially data centers. A new Goldman Sachs note on AI capital expenditures helps explain why. The bank notes that the consensus expects the big hyperscalers—Amazon, Google, Meta, Microsoft, and Oracle—to spend $757 billion on capex in 2026 and $920 billion in 2027.
But the bank thinks that buoyant take may actually be too conservative. If the AI buildout starts to resemble earlier infrastructure booms such as railroads, autos, or the 1990s digital expansion, spending could exceed $1 trillion. Yet money may not be the binding constraint here. The hyperscalers have the deep balance sheets and broad capital market access to keep building, Goldman argues. The harder limits may be physical: delayed data-center projects, scarce memory, labor shortages, and, crucially, power. The AI buildout is an electricity-and-grid story as much as a chips story.
One difference between these two revolutions is the role of technological innovation. Yes, it would be helpful if we could add new energy sources—including small nuclear, advanced geothermal, and even nuclear fusion—to the mix. (And I think we will.) But more important right now is enacting regulatory reform to take better advantage of the energy technologies we already have, such as solar, natural gas, and large-reactor nuclear.
A look at what’s happening inside the House of Representatives gives confidence that at least some folks in Washington understand the seriousness of the need for reform. This session, the House has passed bills meant to help an energy buildout through various changes to the current permitting system:
- Give energy projects more certainty by putting a 150-day clock on NEPA lawsuits after final agency approval, rather than letting projects remain vulnerable to late-stage legal challenges (SPEED Act).
- Let nuclear and other power plants keep Clean Water Act discharge permits for up to ten years instead of forcing renewal every five years (PERMIT Act).
- Replace fragmented permitting paperwork with a single digital dashboard that lets agencies share data, coordinate reviews, and track approvals in real time (ePermit Act).
- Give geothermal developers access to the same streamlined NEPA review process that oil and gas projects can use for certain drilling activities, helping speed up exploration for a round-the-clock clean-energy source that could serve those power-hungry data centers (Geothermal Energy Advancement Act).
- Stop holding states, and the factories and power plants inside them, responsible under the Clean Air Act for air pollution that drifts in from beyond America’s borders (FENCES Act).
- End EPA’s Clean Air Act review of major federal projects and agency actions already undergoing NEPA environmental-impact review, as well as its review of proposed federal regulations (RED Tape Act).
The next steps are unclear, unfortunately. A recent Politico piece suggests that, if the Senate does move, a broader permitting package could absorb all, or at least some, of those House-passed bills. In other words, the House bills are less likely to become law one by one than to serve as bits, bargaining chips, or starting points for a larger Senate-negotiated deal.
Of course, the Senate might not move. As Politico also explains...
....MUCH MORE
A few others from Mr. P.:
Puny Human, I Scoff At Your AI, Soon You Will Know The Power Of Artificial 'Super' Intelligence. Tremble and Weep
[insert 'Bwa Ha Ha' here, if desired]
Pethokoukis: "My chat (+transcript) with Google economist Guy Ben-Ishai on seizing the historic AI moment"
AI: AGI To Arrive During Trump's Presidency; Superintelligence and Singularity Also Possible
Pethokoukis: "Imagine a new economic architecture where growth feeds on itself"
Careful James, you're getting close to describing a cancer.
Pethokoukis: "Why the AI Dividend Isn’t Here Yet"
AI: "Are we just 18 months away from everything changing?"
And many more over the years
