Silicon Valley AI data centers sit idle as power lags

In the heart of Silicon Valley, two vast data centers built for heavy-duty AI computing are sitting idle. Digital Realty’s SJC37 facility and Stack Infrastructure’s SVY02A are finished, but a shortage of available electricity has stalled their launch as Santa Clara struggles to expand its grid fast enough to meet data-center demand.

According to Bloomberg, Digital Realty’s four-story building is rated for a 48 MW critical load, and the neighboring Stack campus was likewise designed for 48 MW, complete with its own substation and eight server halls. Together they could deliver nearly 100 MW for servers and AI accelerators, yet they remain unused. Timelines for full activation are unclear, and the sites could sit empty for years.

Silicon Valley Power (SVP), the city utility, is working to expand infrastructure, but Santa Clara already has 57 data centers operating or under construction at the same time. To boost grid capacity, SVP is investing about $450 million in new lines and substations, a build-out that stretches through 2028. Power will be allocated in phases as the new infrastructure comes online.

Santa Clara’s bottlenecks echo a broader U.S. trend. In Northern Virginia, the country’s largest data-center hub, new connections are delayed by years. Similar issues are surfacing in the Pacific Northwest and the Southeast, where waits can run two to five years. Microsoft has acknowledged that some of its GPUs are idle due to a lack of power.

Even so, Silicon Valley remains a strategic base thanks to its proximity to AI developers and the low-latency links they need; Nvidia’s headquarters are close by. But even the biggest tech players can’t accelerate the pace of grid construction. Modern AI clusters draw hundreds of megawatts and push local networks to their limits, setting hard constraints on how quickly capacity can be deployed.

Digital Realty and Stack say they continue to coordinate with SVP, expecting connections as upgrades wrap up. Meanwhile, demand from AI infrastructure is growing faster than new power capacity arrives, widening the gap between turnkey buildings and available electricity. The servers are ready; the grid isn’t.