The growing demand for artificial intelligence has sparked a global backlash against massive data centers, with communities protesting their energy use, noise, and environmental impact. But a new startup is betting that homeowners will embrace a radically different approach: miniature data centers no bigger than an HVAC unit, installed right in their backyards.

SPAN, an intelligent power management company, has partnered with Nvidia and homebuilder PulteGroup to roll out what it calls XFRA nodes—small, distributed compute units that plug into the spare electrical capacity already available in many homes. The idea is to sidestep the huge land, power, and permitting requirements of traditional data centers by placing computing power where the people and the existing grid infrastructure are.

How It Works

SPAN’s smart panels are the key enabler. These panels monitor a home’s electrical usage in real time, identifying surplus capacity that isn’t being used. The average American home uses only about 40 percent of its electrical capacity, leaving a significant buffer. The XFRA node draws from this surplus, feeding on power that would otherwise go idle. SPAN claims it can install 8,000 XFRA units about six times faster and at five times lower cost than building a typical 100-megawatt centralized data center.

Each XFRA node is designed to be unobtrusive. It sits outside, encased in a weatherproof enclosure similar to a residential generator or air-conditioning unit. The hardware inside is anything but modest: 16 Nvidia RTX6000 graphics cards, four AMD Epyc central processing units, and 3 terabytes of DDR5 memory. The total hardware cost exceeds a quarter of a million dollars, with the memory alone costing nearly $100,000 and each RTX6000 card priced between $9,000 and $10,000. The AMD Epyc processors range from $8,500 to $14,000 each. The system uses liquid cooling to keep temperatures down and fans low, minimizing noise—a major concern for potential neighbors.

The nodes are built by Dell and serviced and maintained by SPAN. In addition to the outdoor unit, SPAN typically installs a smart panel, a backup battery, and sometimes solar panels to offset energy consumption. The company will also pay the host’s electricity and internet bills directly, charging a flat monthly fee that is typically lower than what the homeowner would pay to their utility and ISP.

Business Model and Potential

By distributing compute power across residential and small commercial locations, SPAN aims to reduce latency for AI inference tasks and bring processing closer to end users. This edge computing model is particularly attractive for applications such as real-time video analysis, autonomous vehicle coordination, and smart city infrastructure. The partnership with PulteGroup opens the door to installing XFRA nodes in new housing developments, making them a built-in feature from the start.

SPAN’s spokesperson told Realtor.com, “One big reason the XFRA model works is that the average American home only uses about 40 percent of its electrical capacity. As big data center developers struggle to find power sources and distribution capacity, XFRA uses capacity that’s already available.”

However, analysts urge caution. Alex Cordovil, senior analyst for infrastructure at the Dell’Oro Group, notes that the potential is real but the ceiling is narrower than proponents might hope. “The potential is real where homes pair smart panels with solar and battery storage,” he said. “The economics only stack up if these nodes consume locally generated surplus that would otherwise flow back to the grid at a low feed-in tariff.”

Cordovil points to several challenges. AI accelerators are an expensive ticket for the average homeowner. They perform best in tightly coupled clusters rather than single-rack islands. The hardware is iterating rapidly, making it difficult to maintain a homogeneous fleet. Servicing a dispersed fleet is costly, and the security model of compute bolted to a residential wall is very different from a Tier III data center facility. “A more instructive parallel is how telcos are positioning their existing footprint for AI inference at the edge—they already have power, connectivity, security and a distributed node structure, but still wrestle with running compute across a small number of GPUs per site,” Cordovil concluded.

Hardware and Infrastructure Details

The XFRA node is more than just a GPU box. It includes its own networking gear to connect to the broader internet and to other nodes, forming a distributed compute grid. SPAN’s smart panel acts as a gateway, managing power draw to ensure the home’s critical loads are never interrupted. The backup battery ensures the node can ride through short outages, and optional solar panels can offset daytime consumption. In sunny regions, the node could potentially run entirely on solar during peak hours, exporting surplus back to the grid when appropriate.

The liquid cooling system is a critical component. Traditional data centers rely on large chiller plants and extensive airflow management, but a residential node must fit within a small footprint. Liquid cooling allows for high-density hardware in a compact space while keeping noise levels down. The pumps and fans are engineered to be barely audible, a necessity for installations in quiet neighborhoods.

From a power perspective, each node consumes roughly 10 to 15 kilowatts, comparable to a whole-home air conditioner. However, because it draws from existing capacity, it doesn’t require new utility transformers or substation upgrades. SPAN’s software continuously balances the node’s load with the home’s usage, smoothing spikes and avoiding breaker trips.

Market Context and Competition

The distributed data center concept is not entirely new. Companies like EdgeMicro and Vapor IO have proposed small cell data centers for 5G and other edge applications. But SPAN’s approach is unique in targeting residential backyards and leveraging home electrical capacity. The involvement of Nvidia and PulteGroup gives it credibility and a path to scale. Nvidia provides the GPUs, the most critical component for AI workloads, while PulteGroup offers a built-in distribution channel through new home construction.

The timing aligns with growing resistance to large data centers. In the past year, multiple communities in Virginia, Texas, and other states have blocked or delayed data center projects due to concerns over power demands, water usage, and noise. SPAN’s distributed model could alleviate some of that opposition by spreading the load and making new capacity available in areas that already have housing.

Still, the economics depend on utilization. A single node with 16 RTX6000 GPUs could handle inference for hundreds of concurrent users, but only if there is demand. SPAN will need to sign up enough compute customers—whether cloud providers, enterprises, or AI startups—to keep the nodes busy. The company has not disclosed pricing for the compute capacity, but analysts estimate that node rental could range from a few thousand to tens of thousands of dollars per month, depending on GPU usage.

Cordovil of Dell’Oro Group sees a future for this model as a complement to large campuses, not a replacement. “They will coexist with huge clusters of thousands of GPUs optimized for training, but for inference and low-latency applications, the edge makes sense,” he said. The real test will be whether SPAN can manage the operational complexity of thousands of distributed nodes—each with unique electrical conditions, internet connections, and physical security requirements.

SPAN has not announced a timeline for broader availability beyond its partnership with PulteGroup. But if trials succeed, the idea of a data center in every backyard may shift from a futuristic concept to a neighborhood reality.

Source: Network World News