AES is bringing together solar, storage, and machine learning to power Amazon data centers in CAISO. But delays have been tough.
Photo credit: Shutterstock
Photo credit: Shutterstock
More than seven years after AES first filed a request for interconnection in CAISO, Baldy Mesa, a massive solar-plus-storage project in southern California, came online this summer to power Amazon’s data centers in the region.
The 150-MW solar, 75-MW storage project is also leveraging machine learning to optimize battery charge and discharge. In theory, the four-hour duration batteries will charge during the cheapest four hours in the market, and discharge during the four most expensive hours, thanks to bidding conducted by Fluence Mosaic.
But that added AI layer isn’t something that deeply impacted how AES modeled the project, nor how it’s modeling the rest of its pipeline quite yet. Though Baldy Mesa is one of three total AES projects leveraging AI, it’s the only solar-plus-storage project doing so, Tanya Martinez, explained AES’ senior director of development in CAISO.
“AI is relatively new for us as a company to be implementing,” Martinez explained. “We hope it’s going to improve our [internal rate of return] but from a developer perspective it doesn’t really impact how we’re developing the project or how we model the project,” she added. “It’s definitely looking like it’s improving our returns to an extent, but it doesn’t really impact the development yet.”
A more pressing challenge for AES at the moment, Martinez said, is the five- to seven-year timeline for completing projects in California. That is much slower than many offtakers are looking for, she said. “The [commercial operation dates] we’re seeing for our greenfield projects right now are 2030 and beyond…That makes it much harder for us to be able to contract these projects.”
Baldy Mesa was subject to a slew of unrelated delays and challenges. The first was the pandemic.
Then, in 2023, California passed the Western Joshua Tree Conservation Act, which added additional layers of environmental assessments and permitting, plus requirements to relocate trees and develop mitigation plans.
“Once the decision was made to protect the Joshua Tree, there was a mad rush to get mitigation land,” Martinez explained. That caused “at least a year delay” in the project coming online.
Finally, there was a unique agreement AES needed to reach with Southern California Edison, which owned the land under a transmission line running across the planned project. Getting that easement took two years.
Given California’s long timelines — a problem in other parts of the country as well — Martinez said AES customers are starting to get creative.
“Some unique asks that we’re receiving are regarding microgrids,” Martinez said, pointing to a self-powered project AES developed at a Naval base in Hawaii that the company has considered replicating elsewhere. Although, she added, the queue for grid-connected microgrid projects is still long.
Another potential option to expedite the development schedule in California, Martinez said, is to “break off into a smaller project to see if that might move a little more quickly while we’re developing the larger project.”
AES still believes the trifecta of renewables, storage, and AI is a winning one for them. As chief commercial officer Kleber Costa told Latitude Media this summer, it’s the “clear winner.”
At the time, Costa said AES wasn’t getting a lot of requests for behind-the-meter projects from data center developers. But rising demand is now creating urgency and a willingness to explore unconventional power options.
Amazon and Microsoft are leveraging existing grid connections and turning to large-scale nuclear reactors. Others, like Elon Musk’s xAI, are reportedly circumventing the grid altogether, running off-grid, unauthorized gas turbines.