A data center bathhouse? Architects are rethinking what AI infrastructure owes its neighbors.

• Generative AI has sparked a race among hyperscalers to build large data centers at high speed.
• Some facilities are being built near communities, attracting opposition from concerned residents.
• Architects are designing data centers that could benefit local neighborhoods.

A data center goes up in the middle of a California desert oasis, and in return, a nearby community gets a 32,000-square-foot bathhouse where people lounge in pools warmed by subterranean servers.

Forma, a New York-based architecture studio, designed the "Pink Thermal Baths" concept in 2021, before generative AI ignited a race to build computing facilities across America.

Miroslava Brooks, a founding partner of Forma, told Business Insider the concept wasn't meant to be a blueprint for turning hyperscale AI campuses into spas. Instead, it asks a question that has become more urgent as computing complexes pop up near residential neighborhoods: What can a data center give back to the place that hosts it?

"I really think that the first question is, what does this building give back? And that has to go beyond just energy and the data," Brooks said.

The cloud is getting harder to ignore

Data centers aren't new. Before frontier AI model labs, they've supported banks, websites, streaming services, and cloud storage.

Thomas McGoldrick, a managing director at Gensler who has designed data centers for about 20 years, remembers when they were treated as "back-of-house" facilities supporting individual businesses.

"Now, the transfer of data becomes more and more important," he said. "It's become part of our strategic infrastructure."

A Business Insider analysis found that by the end of 2025, more than 1,400 data centers had been built or approved across the US.

Some data centers are being developed near homes. A 2024 Virginia study found that 29% of operational data center properties were within 200 feet of residentially zoned land and said neighborhood impacts could increase as suitable land grows scarcer.

Residents have raised issues about constant noise, water use, and pressure on electricity bills.

A March Gallup poll found 71% of US adults opposed building an AI data center in their local area, including 48% who "strongly oppose" one.

The scale of data centers and the opposition around them have pushed architects into a debate over whether design can reduce those burdens or merely camouflage them.

Making gray boxes fit in

Gensler, a San Francisco-based architecture firm, has multiple hyperscaler clients, including Microsoft, and works with developers building for major cloud companies.

McGoldrick said those clients prioritize speed to market, scale, access to power, and buildings that can adapt as computing equipment changes.

"They're all trying to get their product out there as fast as they can to make their business grow as best they can," he said.

Within those constraints, Gensler tries to make data centers more than blank, industrial boxes. One approach, McGoldrick said, is to treat them like an "office building that houses computers."

For one complex, Gensler repurposed an old call center campus into a 1 million-square-foot computing facility.

The firm used Corten steel to complement the earthy textures of the local landscape and, through efficient planning, added a one-acre public park.

McGoldrick said the firm often starts with a repeatable prototype, then adjusts the materials and layout for each site.

Beyond aesthetics, data centers have faced increased scrutiny from communities for their energy use and noise. McGoldrick said there are limits to what architects can address.

"There are only so many things that we could control in that environment," McGoldrick said. "So we're openly honest about what we are doing and what we're seeing in other communities."

A data center that gives back

Arup, a UK-based architecture and engineering firm, is exploring how the standardized data center box could change when brought into a city.

Rachel Atthis, an Arup director, said the traditionally long, low building may need a smaller footprint and more height, scaling multiple stories. Bringing a data center into town, she said, means architects have to "turn it on its end."

Arup associate director Marco Mugnai said acoustic screens, landscaped buffers, and changes in site topography can address noise issues.

The firm has also imagined data centers that repurpose structures, such as redundant offshore oil rigs, or pair with tomato farms that use their waste heat.

"I suppose it's about giving back," Atthis said. "I think every building in some way should do that."

Some of the ideas require participation from local governments and developers to build supporting infrastructure, such as district heating networks. Security and backup-power requirements also mean some of Arup's more ambitious concepts may remain years away, Atthis said.

Forma's Pink Thermal Baths makes a similar proposition. Rather than a linear system, in which electricity enters and heat is expelled, Brooks, the Forma founding partner, imagines a "circular model" that turns excess heat into a public use.

"A successful data center," she said, would operate across "the ecology, the infrastructure, and the culture or civicness."

When design isn't enough

Marina Otero Verzier, an architect and Harvard Graduate School of Design lecturer, has explored another use for server heat.

Her "Computational Compost" project channels heat generated from a computer into a vermicomposting system, where worms and microorganisms thrive, creating fertile compost for a local garden.

Otero cautioned that reusing heat is not a complete solution.

"I don't think it's enough just to reuse the heat, because the heat is already a waste product," she said.

Parks, lower-carbon materials, and shared heating systems can improve data centers, Otero said, but "it's just not enough."

She proposed challenging the data center blueprint itself, raising questions about whether every type of data must be immediately available, whether it requires round-the-clock operations, and whether companies' competitive demands should determine how community resources are expended.

This could mean designing facilities for different "ecologies of data," she said — hot, cold, private, temporary — rather than defaulting to the same high-security, always-on model. It could also mean starting with what a community needs, rather than reorganizing housing, energy, and other local infrastructure around a data center.

"The needs of OpenAI, Google, Meta are not the needs of the majority of the world," Otero said. "They are the needs of those companies and their owners."