Chip manufacturing is the rate limiting step by far.
A terawatt is a massive amount of power (2x the average electricity demand of the United Sates), but it could be brought online by 2030 using solar PV and lithium ion batteries.
Most of the necessary supply chain already exists, and the lagging areas could be built well within that timeframe.
This would be ~28% of today's global electricity demand. China is the only country on Earth that uses more electricity, and just barely at ~1.1 TW.
Solar is the only energy source that can grow to this scale this quickly. In fact, at last year's growth rates (28.6%), solar will generate an entire China worth of electricity in 2030.
(although China's demand will be higher by then)
Napkin math on energy:
1. Natural gas
Total U.S. nat gas production is ~11,500 TWh/year on a chemical energy basis. With high efficiency combined cycle gas turbines you'd get 60% of that as electricity (~0.8 TW of power).
If we redirected 100% of our national gas production to this cluster, it still wouldn't be enough.
We would need to increase gas production by ~2.5x as much as we have in the past 20 years.
The gas turbine capacity is an even bigger issue. Total installed gas generating capacity in the U.S. is 500 GW.
(and only 293 GW of that is combined cycle, much of which is older and lower efficiency)
We're backlogged for years with a few 10's of GW of orders. 1,000 GW of gas turbines is not going to happen quickly.
2. Nuclear Fission
Nuclear fission has the potential to supply well over 1 TW of power, but it's not going to be fast. The U.S. has a total of 97 GW of nuclear capacity. The entire world has 399 GW.
At our peak, the U.S. built 9 GW in a year.
We would have to build our entire nuclear fleet every single year for a decade.
This would also be a 3.5x increase in global nuclear fuel consumption and we would have to build out new enrichment and fuel fabrication to meet that.
Our most recent nuclear plants, Vogtle 3 & 4, took 15 years to build. If we could build new plants in 5 years, we'd have to start construction on an entire U.S. fleet every 6 months to have 1 TW of capacity ready within 10 years.
It is possible that we could scale up to that level of construction in the future, but it's not feasible in the near term.
3. Solar Batteries
5 TW (DC) of PV using single axis tracking in West Texas with 16 TWh of storage would give us 1 TW of power at a 92% capacity factor (equivalent to U.S. nuclear).
600 GW (DC) of solar was installed globally in 2024 (35% more than in 2023). At that rate (linearly), 5 TW could be built in just over 8 years.
Realistically, we couldn't use the entire global supply of solar panels, but 600 GW is not the entire global supply.
Global manufacturing capacity was 1 TW/year in 2024 and is expected to reach 1.8 TW/year by the end of 2025.
We could build 5 TW by 2030, and still have enough manufacturing capacity to supply the rest of global projected solar demand.
(although projections have been low for decades now)
Global battery cell manufacturing capacity was 3 TWh/year in 2024, and 7.9 TWh/year is announced for the end of 2025.
Analysts are saying that battery cell capacity is massively oversupplied (relative to 1.6 TWh/year of projected demand), but it works out quite nicely if you're trying to build a 1 TW compute cluster by 2030.
16 TWh over 5 years is 3.2 TWh/year, not even enough demand to account for the oversupply in current projections.
We would need to build new pack and module capacity to integrate all those cells, but that's very doable.
With 24% efficient solar modules and a 0.5 ground coverage ratio, 5 GW DC would require ~42,000 square kilometers of land, less than 0.5% of U.S. land area.
It would be a gargantuan effort, but a 1 TW solar-powered compute cluster could feasibly be brought online by 2030.
(from an energy perspective at least)
Been thinking about the fastest way to bring a terawatt of compute online.
That is roughly equivalent to all electrical power produced in America today.