That 50 GW is ~25% of projected capacity growth by 2030 sounds plausible, until you look under the hood.
The EIA projection includes a lot of intermittent renewables (solar, wind), much of it geographically scattered, slow to interconnect, and unsuited for high-density AI workloads. AGI-scale compute doesn’t just need power. It needs reliable, local, 24/7 baseload, often co-located with data centers. That’s a different infrastructure class.
Even if we had 50 GW of suitable capacity on paper, we’d still face:
Grid interconnection delays of 3–7 years
Transformer lead times of 2–3 years
Substation and transmission bottlenecks
Competing demand from EVs, heat pumps, industry, and climate resilience
So sure, 50 GW is “just” a quarter of the projected increase. But in practice, very little of that is usable for AI unless it’s purpose-built, and built fast. That’s the gap Anthropic (and others) are glossing over.
Granted. But also watch for the explosive growth of natural-gas mobile generators which have none of these constraints. Just hitch one to a semi-tractor and drive to where needed.
I think the word you're looking for is graft. Gotta use a heavy helping of hyperbole if you want to keep the private equity spigot flowing
<< 50 gigawatts of new energy capacity and $500 billion in investment by 2030.>>
Well, how does that compare to the US Energy Information Adminstration (EIA) outlook prediction for all growth in new energy?
https://www.eia.gov/outlooks/aeo/data/browser/#/?id=9-AEO2025&cases=ref2025&sourcekey=0
Total Electric Power Sector Capacity (GW):
2024: 1194 GW
2030: 1400 GW
So the US government is predicting 206 gigawatts of increased capacity by 2030 nationwide.
Doesn't seem too hard to believe that 1/4th of that could be data-centers, especially with their explosive growth in the last few years.
That 50 GW is ~25% of projected capacity growth by 2030 sounds plausible, until you look under the hood.
The EIA projection includes a lot of intermittent renewables (solar, wind), much of it geographically scattered, slow to interconnect, and unsuited for high-density AI workloads. AGI-scale compute doesn’t just need power. It needs reliable, local, 24/7 baseload, often co-located with data centers. That’s a different infrastructure class.
Even if we had 50 GW of suitable capacity on paper, we’d still face:
Grid interconnection delays of 3–7 years
Transformer lead times of 2–3 years
Substation and transmission bottlenecks
Competing demand from EVs, heat pumps, industry, and climate resilience
So sure, 50 GW is “just” a quarter of the projected increase. But in practice, very little of that is usable for AI unless it’s purpose-built, and built fast. That’s the gap Anthropic (and others) are glossing over.
Granted. But also watch for the explosive growth of natural-gas mobile generators which have none of these constraints. Just hitch one to a semi-tractor and drive to where needed.
https://voltagrid.com/