Oracle's 1.3GW Wisconsin AI campus faces power delays as grid review restarts

[Scholar]

The grid approval process for Oracle and Vantage's 1.3 gigawatt AI data centre campus at Port Washington, Wisconsin, has had to restart, TechRadar reports. American Transmission Company filed 564 additional documents that changed the scope of the project. In response, the Wisconsin regulator withdrew its finding that the application was complete, closed the case on 10 September and saw it refiled on 18 September

The knock on effects for the timeline are significant. The 310 megawatts of partial service planned for 2027 may now slip to June 2028, and the full 1.3 gigawatts to April 2029. For comparison, 1.3 gigawatts is roughly the output of a large nuclear power station. The article quotes an analysis saying: "Based on the public record, we see a meaningful risk that load arrives later than Oracle's current guidance."

This is a theme we keep returning to. AI data centres need enormous amounts of electricity, and building the grid connections to supply them takes years. Regulators, utilities and local communities all have a say, and objections or paperwork problems can add long delays. Earlier this month we discussed the growing backlash against AI data centres and whether it could affect quantum computing too

Power is becoming the real bottleneck for AI, more than chips. Companies can order GPUs, but they cannot easily order a new transmission line. That helps explain the interest in unusual ideas we have covered, from underwater data centres to Google's experiments with AI chips in orbit. It also raises questions about who pays for grid upgrades and whether local people benefit

Should communities get more say over giant AI data centres near them? And is the power grid now the biggest limit on AI growth?

Why Oracle's 1.3GW Project Lighthouse may not see full power until 2029www.techradar.com

[SortedIt]

Who pays for the grid upgrades is the big question. If ordinary bill payers end up covering it, there will be a backlash

[Megan]

Power is absolutely the bottleneck now. Chips can be built in months, but grid upgrades take years

Every big AI company is chasing the same limited capacity. Nuclear deals by the big tech companies show how desperate they are for reliable supply

[MurkyInlet]

Delays like this might be healthy. They force companies to think about efficiency rather than just building bigger. Smaller, more efficient models need less power

Efficiency gains are already happening. Necessity speeds them up

[William80]

Communities should have a say. These sites use huge amounts of power and water, and the benefits do not always reach local people

A few hundred jobs is not much for a project that size. Water use for cooling is often overlooked, and it matters a lot in dry regions