AI doesn't need another narrative. It needs power.
The interesting claim is not that solar powers AI.
- Published
- 07 SEPT 2026
- Reading time
- 4 min
- Sources cited
- 02
Two independently documented facts sit either side of this argument. Neither of them is about Sunrun.
First, demand. The Department of Energy reports that US data centres consumed approximately 4.4% of total US electricity in 2023, with modelling estimating approximately 6.7% to 12% by 2028.
Second, supply timing. Berkeley Lab reports roughly 8,200 active projects seeking interconnection at the end of 2025 — 1,312 GW of generation and approximately 749 GW of storage, more than 2,060 GW combined. For projects reaching commercial operation in 2025, the median time from interconnection request to operation exceeded five years in regions with available data.
AI moves faster than substations.
Distributed resources do not remove the need for transmission, power stations and grid upgrades. They may help bridge part of the gap using assets that already sit behind meters. The framing we would defend is not that solar powers AI, but that an existing distributed fleet might create usable capacity faster than new infrastructure can be built.
Sources
Source 01
DOE Data centre demandReport Evaluating the Increase in Electricity Demand from Data Centers
Dec 2024Government
Source 02
LBNL Queued Up 2026Queued Up: 2026 Edition - Characteristics of Power Plants Seeking Transmission Interconnection As of the End of 2025
2026 editionNational laboratory