In the artificial intelligence race, tech giants have broken just about every traditional rule of infrastructure development. Google is moving upstream into development by acquiring Intersect Power. Microsoft is resurrecting a nuclear plant at Three Mile Island. Amazon has spent more than a year navigating federal regulatory hurdles to secure a direct connection to the Susquehanna plant in Pennsylvania.
But there’s one key infrastructure truism that, no matter how much tech giants may spend on behind-the-meter power, the AI boom hasn’t yet figured out a way to innovate around: local communities pose one of the biggest and most unpredictable barriers.
Despite the move towards behind-the-meter generation and energy parks, the AI boom is hitting a pretty solid wall of community opposition. According to Data Center Watch, more projects were delayed or blocked in a single quarter last year than in the previous two years combined.
Between March and June alone, the group tracked $98 billion in stalled investments across 11 states. In 2025, hyperscalers withdrew plans for data centers in several states — including a high profile Google withdrawal in Indiana — after local residents raised concerns about utility costs, water availability, and changes to rural landscapes.
In aggregate, those delays are neutralizing the tech and business gains meant to accelerate speed to power, said Helen Kou, head of U.S. power research and analysis at BloombergNEF.
“We’ve seen a significant increase in overall data center capacity…but we also analyzed how long it takes to develop a data center,” Kou told Latitude Media at the BNEF Summit in San Francisco this week, pointing to the firm’s data center market outlook for the first half of 2026. “Our original hypothesis was that, given all the new innovations, business models, and trends happening around AI, you would have seen the time it takes to move from one state to the next significantly decrease, especially given the fact that we’ve seen so much change happening to the [project] pipeline.”
Instead, the data shows that there has been no marginal change in the time it takes for projects to move from early stage to committed, from committed to under construction, and from under construction to operational.
“So despite the volume of data centers increasing, the size of data centers increasing, all of these other metrics changing, the one trend that hasn’t changed is the time it takes to develop,” Kou said. “And when we qualitatively talk to developers about it, it does come down to localized community [hurdles], like permitting, interconnection constraints. That increasingly has become the major barrier to speed to power.”
Scaling smaller solutions
BNEF’s data examines nationwide trends, but there are geographic limitations to where data center developers can utilize the “energy park” approach favored by Intersect and Google, or other “bring-your-own-capacity” approaches to speed to power.
For now, the energy park model of co-locating with data centers works best in ERCOT, where Texas’ deregulated market has established protocols that allow large loads to offset their power use with onsite generation. In most other states, doing so is complicated by universal service rules, and the lack of tariff structures to credit isolated co-location.
But even in cases where companies can get their co-located or behind-the-meter power online, community hurdles remain.
Nick Chaset, CEO of Octopus Energy U.S., said his company is seeing a widening opportunity for small-scale solutions, like aggregated residential flexibility, to help address that issue.
Octopus, Chaset said, has started to experiment with the bring-your-own-capacity framework for data centers, allowing them to subsidize residential flexibility and for solar and battery storage for impacted ratepayers, who could then see their bills go down as the data center comes in.
It’s a model Octopus pioneered in the UK to get residents on board with wind farms, with a concept they called “Fan Club,” whereby ratepayers whose homes neighbored turbines received discounts on their power bills when the wind was blowing. “Suddenly you went from a dynamic where you couldn’t permit a wind facility onshore in the UK to thousands of communities asking for wind turbines to be installed,” Chasset added. “Because what people were most worried about was the cost of power.”
As power bills rise around the U.S., and data center developers realize that even owning the chips and the generation doesn’t solve the problem, Octopus is betting a similar approach could work stateside.
