California’s Grid Can’t Handle OpenAI’s Next Expansion—And Lawmakers Know It

The state Capitol’s hearing rooms in downtown Sacramento can feel oddly quiet in the late afternoon. Staffers whisper in the back rows, lawmakers shuffle papers, and the intermissions between speakers are occasionally filled with the buzz of fluorescent lights. However, the current topic of discussion—electricity and artificial intelligence—has a sort of unspoken urgency.

More precisely, the energy appetite of OpenAI and similar companies. It’s becoming more and more obvious that California’s aging power grid—something far less glamorous than algorithms—may collide with the infrastructure driving the AI boom. Furthermore, analysts and regulators claim that the issue is no longer theoretical. The figures are beginning to seem daunting.

Over the next ten years, data centers requesting new electricity connections could add about 10 gigawatts of demand, according to state utilities. That is an astounding amount. To put it in perspective, it is about four times the output of Diablo Canyon Nuclear Power Plant, the last nuclear facility in California.

Standing outside one of the sprawling data facilities south of San Francisco, the issue becomes easier to picture. Behind security fences, the enormous, windowless rectangular buildings hum softly. They appear nearly unremarkable from the outside, resembling enormous warehouses. On the inside, however, thousands of servers run nonstop, cooling systems roar, and electricity flows at a rate more akin to small cities than office buildings.

The demand for energy has skyrocketed due to artificial intelligence. Large computing clusters are needed to train and run sophisticated models, such as those created by businesses headed by Sam Altman. Strong GPUs, continuous electricity, and industrial cooling are necessary for those clusters. To put it another way, more power than many areas were intended to provide.

California’s energy planners might have misjudged the rate at which AI would grow. Despite increasing internet traffic, data-center electricity consumption stayed largely constant for many years. Improvements in efficiency assisted in controlling consumption.

However, AI altered the situation. Computing resources are consumed by machine-learning models, and demand is rising sharply due to the new generation of generative AI systems. According to analysts, data centers currently account for a sizable portion of the nation’s electricity consumption, and as AI permeates commonplace devices, that percentage may rise significantly.

Policymakers are feeling a little tense as they watch this develop. California lawmakers are currently arguing over who should foot the bill for the power infrastructure needed to support the growth of AI. Advocates for consumers contend that just because tech companies wish to expand their computing facilities, regular households shouldn’t have to pay more for electricity.

In a recent report, Pedro Nava, chair of California’s Little Hoover Commission, stated unequivocally that data centers should bear the entire cost of the strain they put on the grid. Although the argument seems straightforward, putting it into practice could be challenging.

In addition to upgrading substations and possibly building more generation capacity, utilities would need to build new transmission lines. It may take years or even ten years to finish those projects. In the meantime, the AI sector is developing at a speed that is more like months.

Whether the two timelines can coexist is still up for debate. Environmental issues are also present. California wants to drastically cut carbon emissions over the next several decades. However, in order to keep the lights on, utilities may be forced to use quicker, dirtier solutions, such as natural gas power plants, due to the abrupt increase in energy demand from AI facilities.

Researchers have also flagged another issue: water use. Large data centers frequently depend on substantial water supplies for cooling systems, which could lead to conflicts in areas that are already experiencing drought cycles.

As the discussion progresses, it becomes clear that the AI revolution is more than just a software story. It is also a tale of physical infrastructure, including transmission lines, cooling towers, electric grids, and land use choices. Some policymakers have been unprepared for this reality.

It seems that even tech executives are aware of the issue. Numerous businesses have begun investigating alternative energy sources, such as nuclear power collaborations and renewable energy initiatives created especially for data centers.

However, those answers are still being developed.

California is currently in a somewhat awkward situation. The state continues to be the center of the AI sector. Businesses developing cutting-edge systems frequently choose to stay near Silicon Valley’s talent and venture capital ecosystem.

However, the ecosystem’s very success is putting additional pressure on the infrastructure in the area.

The irony is difficult to miss. The technology sector frequently takes great pride in using software and creativity to solve challenging issues. However, the problem arising from the demand for AI power is stubbornly physical. A transmission bottleneck cannot be circumvented by coding.

You must construct something. It’s unclear if California will be able to grow its grid fast enough. Energy planners are trying to map future demand while admitting that many AI projects are still speculative. It’s possible that some planned data centers won’t ever be constructed. Others may operate at less capacity than anticipated.

However, the general course appears to be clear. Artificial intelligence is transitioning from cutting-edge technology to commonplace infrastructure. Furthermore, the systems that support AI will need massive investments that go well beyond Silicon Valley campuses, just like highways, airports, or electrical grids before it.

It seems like the discussion is just getting started as I watch lawmakers struggle with spreadsheets of energy forecasts in the Capitol hearing room.

Because creating smarter machines may not be California’s greatest technological challenge if the AI boom keeps up its current rate. Finding enough electricity to power them could be the problem.