Northwest Power Plan Calls for 11 GW as Data Centers Drive Demand

A grid built for flat demand meets a surge

For about a decade, the Pacific Northwest's power system barely grew. From 2010 through 2020, annual electricity use across Idaho, western Montana, Oregon and Washington stayed between roughly 20,500 and 21,200 average megawatts (aMW). Per-person consumption fell until 2020 and has risen since 2021, in part because data center load does not depend on population growth.4 That period is over. The region's planners are now preparing the largest grid build-out the Northwest has attempted in decades.

The main document is the draft Ninth Northwest Power Plan. The Northwest Power and Conservation Council voted unanimously on Aug. 11 to release it for public review.4647 It forecasts that the regional power system will grow by 50 to 100 percent by 2046. In the near term, which the plan defines as before 2032, data centers and the tech sector are expected to be the largest source of new load.11 Public comments are due Oct. 16, and hearings are still being held across the four states.1241

The numbers behind the build-out

The council expects average annual demand to rise from about 22,000 aMW today to between 31,000 and 44,000 aMW by 2046. At the high end, that is roughly double current use.4214 Peak demand is now about 35,500 megawatts in winter and 33,300 in summer. It could reach 47,000 to 60,000 megawatts.15

To meet the need through 2032, the draft calls for about 9,000 megawatts of wind and solar, roughly 5,000 to 5,200 megawatts of battery storage and 2,100 megawatts of natural gas generation.1247 Several smaller items fill out the portfolio:

• Efficiency: 1,060 aMW of energy efficiency, split between Oregon and Washington (860 aMW) and Idaho and Montana (200 aMW). This is the first time the council has set a two-part conservation target.47
• Demand response: 590 MW.12
• Voltage regulation: 220 MW.12

The region currently has about 70,000 megawatts of generating capacity.12 The council estimates the fixed cost of the 2032 portfolio at $2.3 billion and describes that as about 0.15 percent of the four-state region's 2025 economic output.49 Coverage has reported the headline forecast slightly differently. Some reports say the region may need 40 to 100 percent more electricity, while the council's own summaries describe 50 to 100 percent system growth.1311 The gap appears to come from measuring energy versus system size, but either way the range is very wide.

Data centers: the biggest source of uncertainty

The reporting agrees on one central point: data centers are the region's biggest unknown. The council's forecast says new Northwest data centers could use about as much power on average as Seattle does by 2046, or five times as much.13 The council has said its forecast uncertainty is higher than ever before.13

Shorter-term estimates vary just as much. A Washington state workgroup, citing the council, said data centers and chip fabs could add 2,200 to 4,800 aMW of load by 2030.9 An earlier council forecast ran from about 1,500 aMW in a low case to 5,000 aMW in a high case.1 Part of the reason is speculative demand. Developers often pitch several sites at once and submit requests that may be duplicates or never get built. Regulators are considering requiring a larger financial commitment before a utility promises service.20 Back in 2024, consultant Robert Cromwell warned the council that underestimating load carried more risk than overestimating it.7

The Seattle Times analysis, syndicated widely, highlights a less obvious problem: timing. Data centers are expected to connect first, as tech companies race to meet AI demand. That could give them first access to cheap renewables, available transmission and easy sites before electric vehicles and heat pumps arrive at scale.20 Council member KC Golden argued that requiring data centers to bring their own power would not fully solve this, because they would still compete for the same limited supply of power lines, equipment, workers and sites.20

The main point here, in this analysis, is that data centers are the easier load to serve. Their demand is steady. Heat pumps and evening EV charging create sharp peaks, and heat pump demand spikes in cold weather when wind and solar produce less.20 The infrastructure that data centers claim now will be needed later for that harder load.

Transmission is the real bottleneck

New generation does little if the power cannot reach customers. Bonneville Power Administration (BPA), the region's largest high-voltage transmission provider, has about 23 major projects in its Grid Expansion and Reinforcement Portfolio, with an estimated $5 billion-plus in investment through 2035.21 More broadly, BPA says it is investing up to $25 billion in transmission projects and reinforcements over the next decade.24 It is also replacing first-come, first-served handling of generator interconnection requests with a cluster process that puts the most ready projects first.21

The scale of the backlog is long-standing. BPA's own figures show the 1960s and 1970s accounted for thousands of miles of new 500-kV line, compared with a small fraction of that in each decade since. BPA had also studied more than 17,000 MW of requested transmission service since 2019.26 A Washington state analysis calls transmission the most critical limiting factor for clean energy development in the state.20

Utilities are also trying to get more out of existing lines. Puget Sound Energy received a $40.2 million federal grant, which it will match, for Project PULSE. The project will upgrade about 23 miles of transmission and add 178 monitoring sensors. PSE CEO Mary Kipp says the work will move more power over existing lines while avoiding costly new infrastructure.18 Programs that let flexible data centers connect faster follow the same logic. The Electric Power Research Institute reported that an Arizona data center cut its power use by a quarter during peak hours without degrading AI computing. Portland General Electric is testing similar flexibility measures in Hillsboro.1

Where nuclear fits, and where it doesn't

For readers following the push for nuclear-powered data centers, the draft plan's near-term portfolio is notable for what it leaves out. Its recommended resources through 2032 are renewables, storage, gas, efficiency and demand-side measures, with no nuclear.4147 The council treats small modular reactors (SMRs) as a stand-in for "clean baseload" in its modeling. That category assumes availability from 2035, a five-year build time, a cap of five units and an overnight capital cost of $9,000 per kilowatt. By comparison, the council assumes $2,000 per kilowatt for utility-scale solar and $1,000 to $1,800 per kilowatt for gas peakers.4442

The main corporate nuclear project in the region follows a similar timeline. Energy Northwest's Cascade Advanced Energy Facility near Richland, funded by Amazon, would start with four X-energy Xe-100 reactors totaling 320 MW and could expand to 12 units and 960 MW. Construction is expected to begin by the end of the decade, with operations in the 2030s.3137 Amazon has the right to buy power from the first phase.33 Cascade's schedule depends on X-energy's first plant at a Dow site in Texas, which now comes first.3740

The nearest new nuclear output will come from an existing plant. BPA and Energy Northwest are upgrading the Columbia Generating Station, still the Northwest's only commercial reactor, to add about 160 MW by 2031.2432

Overall, nuclear is a 2030s option in the Northwest, not the answer to the AI load expected over the next six years. Gas, batteries and transmission will have to cover the 2020s.

The reliability stakes

Outside analysts see the timeline as tight. In its 2026 long-term assessment, the North American Electric Reliability Corp. (NERC) flagged parts of the Northwest, where near-term additions are mostly solar, for growing unserved-energy risk in both summer and winter.2 Modeling by the consulting firm E3, cited by the Northwest Gas Association, found a 9-gigawatt gap in effective capacity by 2030, rising to 14 to 18 gigawatts by 2035. The worst risk is extended cold during low-water years.14 The council's own needs assessment likewise found the largest and longest shortfalls in winter.42

That winter risk explains why the 2,100 MW of gas is in the plan. The council says it is meant for low-water years when hydropower falls short.12 Not everyone agrees on the amount. Jennifer Light, the council's director of power planning, said it has heard arguments both that the draft includes too much gas and that it includes too little.20

The bottom line

The council plans to reassess supply and demand every year and adjust course as needed. Light has noted that scaling back is easier than ramping up.20 She also said that if data center growth comes in lower, batteries would be the main part of the plan to shrink, since renewables would still be needed.3 Golden described a possible "self-leveling" effect, in which some proposed data centers never get built if the grid cannot keep up.20

That may be the most realistic description of what happens next. The region's capacity to build power lines, rather than AI demand forecasts, is likely to set the pace of data center growth. Meanwhile, the slower, more uneven demand from electrification will need the room that data centers are moving to claim first. The final plan is expected between November 2026 and early 2027.4711 Its most important decisions may be about who gets grid capacity first, rather than how many megawatts to build.