California-based Span, in partnership with Nvidia, has deployed prototype data center “nodes” in Northern California. The cabinet-sized units, dubbed XFRA, are installed on the sides of homes and small businesses. Requiring no fans, the technology is quiet, mitigating the problem of noise pollution that has drawn the ire of residents of areas with nearby warehouse data centers. 

Ryan Harris, chief revenue officer of Span, said the company estimates XFRA will be able to generate about one to two megawatts worth of compute later this year, scaling across the country to an annual capacity of more than 1 gigawatt beginning next year. PulteGroup, among the largest homebuilders in the U.S., is testing the system. Nvidia will provide the liquid-cooled RTX PRO 6000 Blackwell Server Edition GPUs for the system.

“We do see a path to being able to contribute on an annual basis hundreds of megawatts, if not gigawatts, of scale compute capacity, while doing so in a deflationary-to-energy-price way,” Harris told Fortune.

All the while, tensions between hyperscalers and residents have been mounting over AI’s rising costs and environmental impacts. With data centers the size of dozens of football fields combined sprouting up around the country, residents have protested the construction of AI infrastructure, which McKinsey projected to touch $7 trillion in capital expenditures by 2030. The warehouses erected to store and process massive amounts of data have strained the U.S.’s already beleaguered grid system, potentially driving up electric bills by 6% over the next year, according to Goldman Sachs research. 

That’s on top of concerns that data centers are guzzling water as part of their cooling systems. Two data center developments, one in Arizona and one in Georgia, took public water without authorization, and a recent study by the Houston Advanced Research Center projected the centers would drain as much as 399 billion gallons of water in Texas alone by 2030.

“We know what a big project this is, and what a nuisance it’s going to be, and what environmental impact it’s going to have on this area,” Kathryn Haushalter, a 42-year-old former U.S. Marine living in Saline Township, Michigan, across a future data center site, recently told Fortune. “I’m just so nervous for everybody else that doesn’t realize.”

Data centers in your home

Span’s XTRA models are part of a wider distributed network of AI infrastructure, using a home’s underused electrical capacity to create something similar to a cloud of compute that can be given to service providers. Span can install nodes at six-times the speed of centralized 100-megawatt data centers and at about one-fifth of the cost of construction.

The company charges a flat monthly fee of about $150. In return, it essentially pays a host’s electricity and internet bills. The computing power generated from the nodes are distributed to customers like hyperscalers and AI companies. XFRA is not designed to replace commercial data centers, according to the company, but rather to reduce strain on the grid.

Heata, a UK-based startup, also installs servers that act as a “virtual data center,” processing cloud computing workloads. But it adds a twist by using thermal conductors to carry heat from computer processors to cylinders filled with water for home heating needs.

The startup has installed units in about 100 homes and claims to have saved about 1 gigawatt-hour of energy. About 70% of the saved energy comes from less need to use domestic gas or electric heating systems in homes, while the remaining 30% comes from less of a need to cool data center processors.

A Heata spokesperson told Fortune the company has generated 8 million liters of hot water, saving homes about $55,000 on energy bills.

Courtesy of Span

Questions around the true benefit of home data centers

While these startups may promise to save both waste heat and money, Utah State University physics professor Robert Davies warned that efforts to modestly reduce the ecological harms of data center power usage and grid strain could actually exacerbate the problem.

In a preliminary analysis, Davis calculated only 30%-40% of homes may be suitable for mini data centers or servers due to integration constraints, the need for stable internet, and participants being willing to have the technology installed in their homes.

Separately, only 2%-3% of homes could realistically be heated through alternative energy-harnessing technologies because of constraints to how much waste heat can be collected. Additionally, there’s a lot of heating energy that could go to waste because in many geographies, it’s a seasonal need.

Davies said these new technologies are truly helpful and could still benefit millions of households. But he cautioned that framing data center expansion as something that can be made more efficient could be a slippery slope in the bigger picture of the environmental impact of AI infrastructure. 

“These projects tend to be heavy on the benefit analysis and very light on the cost analysis,” Davies told Fortune. “And you don’t actually get a full sense of the cost until you do a whole systems analysis. These are multi-generational challenges, and are they solving problems that we really need solved?”

He fears that increased efficiency of repurposing data center waste will encourage even greater data center expansion that further taxes the environment. He invoked Jevons paradox, a 160-year-old theory that says as a resource becomes more efficient to use, more, rather than less, of it is used. It was based on English economist William Stanley Jevons’ observation that better steam engines made coal cheaper, subsequently increasing total coal consumption.

“We now need about 45% less energy to do the same thing that we needed 35 years ago. So that seems awesome,” Davies said. “Are we using 45% less energy than we were 30 years ago? The answer is, no. Turns out, we’re using about 70% more energy.” 

The Heata spokesperson said the company deals with substitution, not just increased efficiency, because homes must be heated whether or not one of its servers is present. If compute demand continues to rise, creating an integrated energy system to help fill heating demand becomes even more important, according to the company.

Still, Davies is worried the demand for compute far outstrips capabilities to repurpose waste heat, and could lead to more data center construction that would further burden environmental capacity versus extend it.

“The strategy that I see the sector applying here is seductive,” he said. “It seems useful, we want it to be useful. But in a whole systems analysis, it’s really not.”