Nvidia wants to turn American homes into AI data centers with new ‘garage server farm’ plan

For years, giant AI data centers have swallowed farmland, strained power grids, and sparked fights in communities across America. Now Nvidia has a different idea: move some of that compute into suburban neighborhoods.

The company is teaming up with California startup Span and homebuilder PulteGroup to install small residential AI compute units directly onto newly built homes. The units, called XFRA nodes, function like miniature data centers attached to houses and small commercial buildings, tapping unused electrical capacity through Span’s smart electrical panels.

The concept sounds unusual at first. A home that doubles as part of an AI infrastructure network feels closer to science fiction than to suburban planning. Yet Nvidia and Span believe the model could solve one of the biggest problems facing the AI boom: where to put all the compute.

“Nvidia (NVDA) partnered with homebuilder PulteGroup (PHM) through startup Span to deploy residential “XFRA units”—small data centers on new homes that tap unused grid capacity, with Span claiming it can deploy 8,000 units six times faster and at one-fifth the cost of a 100-megawatt centralized data center,” Yahoo Finance reported.

The timing matters. Across the U.S., resistance to massive data center projects is growing. Residents have pushed back against facilities blamed for heavy electricity use, water consumption, noise, and rising utility costs. Span’s pitch flips that model on its head. Instead of concentrating compute inside giant warehouse-sized facilities, the company wants to distribute it across thousands of homes.

“NVIDIA wants the next AI factory to sit in your garage.”

Nvidia, PulteGroup partner with Span to put mini data centers on homes

Through the partnership, PulteGroup will help bring the hardware into newly constructed communities across its national footprint. Span says the economics are compelling. The company claims it can deploy 8,000 XFRA units six times faster and at one-fifth the cost of building a comparable 100-megawatt centralized data center.

In a CNBC interview, SPAN said it could deploy 8,000 XFRA units at roughly one-fifth the cost of building a traditional 100-megawatt data center with comparable compute capacity.

“Span is a California-based startup that originally launched with ‘smart’ electrical panels designed to help homeowners save money on their electricity bills. Now, with the help of Nvidia, it has come up with something new: small, fractional data centers, or ‘nodes,’ called XFRA units that can be installed on the side of residential homes and small commercial buildings,” CNBC reported.

SPAN’s XFRA node would sit alongside houses, with a wall-mounted smart panel and a nearby backup battery. Credit: SPAN

Traditional data centers consume staggering amounts of electricity. A single large facility can use as much power as roughly 100,000 households. Span argues that its distributed approach makes use of existing idle residential grid capacity.

Homeowners would receive flat-rate power and internet service, alongside compensation tied to Span’s energy and network usage.

Span CEO Arch Rao says the approach helps “meet what is clearly an insatiable demand for more compute, much more cost-effectively, while benefiting individual consumers.”

The company plans to start small before scaling aggressively. According to Ars Technica, SPAN expects to deploy as many as 80,000 XFRA nodes across the United States starting in 2027, creating more than one gigawatt of distributed compute capacity.

The network is not intended to replace giant hyperscale facilities operated by companies like Google or Microsoft. Instead, the residential nodes are better suited for AI inference, cloud gaming, content streaming, and other workloads closer to end users.

SPAN believes the model sidesteps many of the problems attached to hyperscale data centers.

“Data centers are loud, ugly, and often drive up local electricity bills,” said Chris Lander, vice president of XFRA at SPAN, in correspondence with Ars. “[This] is quiet, discreet, and makes energy more affordable for the host and community.”

Nvidia’s next AI data centers may sit in your garage, not a warehouse

PulteGroup’s involvement gives the experiment something many startups lack: immediate physical distribution. The homebuilder operates more than 1,000 active communities across over 45 U.S. markets. That gives Nvidia a direct path into newly built neighborhoods at a time when housing construction remains at a historically elevated level.

Housing starts reached a seasonally adjusted annual rate of 1.50 million units in March 2026, according to federal housing data. Pulte itself reported 8,034 net new home orders during the first quarter, up 3% from a year earlier.

The project fits into Nvidia’s larger push to spread AI compute beyond centralized cloud infrastructure. The company has spent the past year introducing products aimed at edge AI and local inference, including DGX Spark personal AI supercomputers, Blackwell GPUs, BlueField data processors, and Jetson robotics systems.

On Nvidia’s latest earnings call, CFO Colette Kress said: “DGX Spark and Station revolutionized personal computing by putting the power of an AI supercomputer in a desktop form factor.”

CEO Jensen Huang painted a broader picture of where he thinks the market is heading.

“Enterprise adoption of agents is skyrocketing,” Huang said during Nvidia’s Q4 earnings call. He added that AI is “going everywhere, doing everything, all at once.”

“As the demand for AI and inference compute continues to accelerate, there is a critical need for low-latency solutions that are proximal to end users and can scale rapidly,” said Marc Spieler, Senior Managing Director of Global Energy Industry at NVIDIA, according to SPAN news release. “SPAN is pioneering new ways to deploy enterprise-grade GPUs in distributed environments. The XFRA solution helps meet the specific power and latency requirements of modern inference workloads while making compute more accessible and efficient.”

Nvidia certainly has the financial firepower to experiment with new infrastructure models. The company reported Q4 FY2026 revenue of $68.13 billion, up more than 73% year over year. Free cash flow climbed 124% to $34.9 billion. Nvidia guided for roughly $78 billion in Q1 FY2027 revenue, excluding China data center sales.

Investors have rewarded the company accordingly. Nvidia shares have climbed more than 83% over the past year.

The bigger question is whether consumers will embrace the idea of living beside AI hardware connected to a distributed compute network.

That answer could shape the next phase of AI infrastructure. If Span’s model gains traction, the future AI data center may no longer sit in a remote industrial park. It may sit quietly on the side of your neighbor’s house.

Image credit: SPAN