The case for the disappearing data center

For decades, data centers purposefully hid in plain sight. Most people rarely noticed them, much less questioned how much power, water or land they consumed. AI has ended that quiet arrangement with a loud public pushback. Its massive new facilities are turning previously obscure infrastructure into a visible and often noisy public target and forcing the industry to confront a difficult question: Can a data center this large ever disappear?

“Given AI’s prominence and the recent politicization of data centers, it is unlikely that large, high-density, high-performance computing data centers will be able to fade from public view any time soon,” said Daniel Farris, data center attorney at global law firm Foley & Lardner, and co-lead of the firm’s data center and digital infrastructure team.

In fact, Farris said he expects the spotlight on data centers to only grow — and not in a good way.

“The scrutiny drawn by large AI data centers may bleed over to more traditional colocation and hyperscale data centers that have seamlessly integrated into communities across the country for decades,” he said.

Fighting for invisibility at this point looks to be a losing cause. Although there are places where these data centers are already getting smaller.

On the physical side, “the footprint is already shrinking quickly,” according to Damir Špoljarič, a partner at Altiora Partners and managing partner at Gi21 Capital. He said that Altiora , an investment and growth consultancy firm, is designing for 150 to 250 kW per rack, and the next Nvidia generation is expected to draw around 220 kW per rack. “The higher the density, the more compute you can fit into a much smaller site. That is very different from the large, low-density halls people often imagine when they hear mega data center,” Špoljarič added.

Creating a better match between supply and demand can shrink the size of new data centers and, at least theoretically, make them quieter and less of an energy and water hog.

“A large share of the physical footprint of AI data centers comes from planning for peak demand that rarely materializes,” said Srinivas Chippagiri, a senior member of technical staff at Salesforce. “Capacity gets built for the worst case, and that worst case drives how much land, power and cooling infrastructure gets committed at a given site.”

Instead of planning for the worst case and thereby reaping the worst public responses, data center developers and organizations need to “get real” about workload flexibility, according to Chippagiri. Compute can be deferred, shifted to another region or run at lower priority during peak periods, which can in turn materially reduce how much new physical build developers need to request from a community in the first place, he added.

“That is an operational lever, not an engineering one, and it is underused,” Chippagiri said.

However, shrinking the size of AI data centers isn’t enough to quell the rising anger against the fact that they exist at all.

Size matters, but it isn’t everything

Size does affect public perception, but size isn’t the biggest issue cited in public pushback.

“They [AI data centers] cannot disappear at gigawatt scale,” said James Dickey, a Texas state and local government and public affairs strategist, and founder and principal of JD Key Consulting. “The old hall was tens of megawatts, a tax abatement and a fence. An AI campus pulls hundreds of megawatts, competes for groundwater and shows up in the Electric Reliability Council of Texas queue as a transmission problem. Neighbors notice.” Dickey is also former chairman of the Republican Party of Texas.

The noise from AI data center builders rolling up with 60 or so gas turbines to power the huge structures isn’t helping the cause. Neighbors complain of falling home values, hearing problems, sleeping problems and a nonstop volume not unlike living next to an endless airport runway where jet engines are forever idling. Lawsuits and regulator rulings have added to the woes of data center developers and AI builders, too.

It’s clear that making the facilities smaller and moving them to more rural areas won’t make data centers invisible as long as water tables go down, power bills go up, and the noise is nonstop. The key to resolving public pushback is in resolving all of these issues that, separately and combined, are not annoyances but significant concerns.

“Acceptable is not invisible,” Dickey said. “Acceptable is paying for power, water and roads, cooling without draining the neighbor’s well, and making the first public notice a meeting you called.”

Rethinking the AI super data center concept

There are those who call for a total reconsideration of the basic idea of AI data centers. After all, the old ways are not always the best in an era of such profound change, most of which is brought about by the occupants of the offending data centers: AI.

“The instinct to build massive, centralized AI super data centers is a relic of legacy cloud thinking. To reduce civic friction, we must shift toward decentralized, multi-agent architectures,” said Eshaan Jain, senior product manager at Mphasis Silverline contracted to support T-Mobile’s national B2B portfolio.

“By processing lighter, intent-driven tasks at the edge and reserving heavy foundational model compute strictly for complex reasoning, we can drastically shrink the required physical footprint of any single facility, diffusing the local infrastructure burden across smaller, less intrusive nodes,” Jain added.

Repurposing existing buildings has also come to the forefront in new thinking about where to put AI data centers.

“We often prioritize adaptive reuse. Our Vint Hill, Virginia, campus occupies a former U.S. government office building, while our Hillsboro, Oregon, facility was once a pipe organ factory,” said Nishant Kapoor, senior director of data center and network operations at OVHcloud US. “We have taken a similar approach with former industrial sites in Canada, Germany and France.”

Adapting existing but often abandoned buildings can help alleviate some of the energy and water infrastructure needs of the data center itself, although not the usage amounts of either. Repurposing existing buildings can also revitalize dying or abandoned communities.

“Remote siting is central. Locating campuses near stranded or underutilized generation, transmission capacity, fiber routes and industrial land can avoid placing disproportionate burdens on residential communities,” said Justin Hood, chief operating officer at ThisWay Global, a builder of sovereign AI infrastructure for enterprises, governments and research institutions.

“In many cases, a purpose-built campus in a low-conflict industrial or energy zone is preferable to retrofitting a hyperscale facility into a constrained suburban market,” Hood added.

Communicating new ideas, plans and mutual benefits will need to happen on a regular basis so that the current contentions between providers and communities do not escalate. That will require both sides to be creative problem-solvers and willing partners. For example, data center owners and operators must accept the nuances involved, rather than pursue a forceful assertion akin to Star Trek’s Borg mantra, “resistance is futile.”

As digital infrastructure becomes increasingly important to economic growth and innovation, “that doesn’t mean communities should simply absorb the impacts,” said Steven Lim, senior vice president of marketing and GTM strategy at NTT Global Data Centers.

In the end, “the question shouldn’t be about making data centers invisible. It’s about making them better neighbors,” Lim said. “The goal is to build and operate facilities in a way that respects local priorities while delivering the digital infrastructure society depends on.”

How does your data center strategy take into consideration peak demand or actual workload patterns? Let us know at [email protected].