ClaimAudio · 102:34 — 104:04
Power will not be the binding constraint on US AI compute scaling, because unconventional generation (behind-the-meter turbines, reciprocating and ship engines, fuel cells, solar-plus-battery) can unlock roughly 20% of the terawatt-scale US grid that today sits idle except during rare peak-demand spikes.
Dylan argues the US can scale AI power well beyond current levels by tapping unconventional generation sources to absorb grid capacity normally reserved for rare peak demand, meaning power — unlike chips — isn't fundamentally supply-constrained. ✦ AI generated
Dylan Patel · Dwarkesh Podcast · 2026-03-13 · original ↗
plays this moment only · 102:34 — 104:04
Elicited by
“If in 2030 we have enough logic and memory to do 200 gigawatts a year, do you just think that these things are on a path to ramp up to more than 200 gigawatts a year, or what do you see?”
Today, data centers are only 3-4% of the power of the US grid, and by 2028 they'll be 10%. But if you can unlock 20% of the US grid like this, it's not that crazy. The US grid is terawatt-level, not hundreds-of-gigawatts-level. So we can add a lot more energy.
verbatim transcript · starts at 102:34
Transcript · around this moment
102:34– Scaling power in the US will not be a problem
102:34– Scaling power in the US will not be a problem
Around this claim
Counterpoint · 3
Compute is critically scarce through at least 2027, with choke points everywhere — energy, land, chips, talent, and trust — requiring OpenAI to invest aggressively years ahead of demand.Sarah Friar · All-In Podcast · conf 80%The US power market faces structurally tight supply for the next 20 years driven by a new technological demand cycle, and AI demand only turbocharges that — it is not the root cause.Dan Dreyfus · All-In Podcast · conf 80%Compute is a critically scarce resource in 2026 and will remain limited through 2027, forcing OpenAI to make multi-year bets on infrastructure that won't deliver until 2028 or later.Sarah Friar · All-In Podcast · conf 80%
This moment responds to
rebuts → Microsoft's real supply constraint today is power and the ability to build out data centers fast, not a shortage of chips — chips are actually sitting in inventory unable to be plugged in.Satya Nadella · BG2 Podrebuts → The AI industry is currently supply constrained rather than demand constrained — scarce compute, memory, data centers, and power — which makes a bubble less likely right now, though that could change with an algorithmic breakthrough enabling far smaller, more efficient models.David · a16z Podcastprovides context → By 2028-2029 the single biggest constraint on scaling AI compute will be ASML's EUV lithography tools: ASML can only build about 70 this year, 80 next year, and just over 100 by 2030 even under very aggressive supply chain expansion.Dylan Patel · Dwarkesh Podcast