Racing to “mine” residential power capacity shortages? NVIDIA partners with the smart-meter startup Span—mini AI data centers to power your home’s exterior walls

As global demand for AI infrastructure is rising rapidly, chip design leader Nvidia announced a partnership with California startup Span, planning to install mini AI data centers directly on the exterior walls of ordinary homes and small commercial buildings. The innovative proposal is seen not only as a new way to address the compute shortage, but also as a chance to reduce the energy and environmental controversies associated with traditional large data centers.

（Morgan Stanley warns: AI demand reveals a U.S. “power shortage”—could real-time power solutions be the next investment opportunity?）

From smart meters to mini data centers, Nvidia and Span take a key step together

Founded in 2018, Span initially entered the market with “smart power panels,” helping homeowners manage electricity more efficiently and reduce electricity bill expenses. The company has since expanded its product lineup by launching remotely accessible mini data center node devices, officially moving into the field of AI infrastructure.

Its core product is called the “XFRA node.” It has the appearance of a small white box and is installed side by side on the building exterior wall alongside air conditioners and electrical equipment. The complete system includes a smart power panel, an XFRA node, and a backup battery, and in some cases is also paired with solar panels. Each device is reported to be equipped with 16 Nvidia Blackwell GPUs, 4 AMD EPYC CPUs, and 3TB of memory.

The XFRA node works by having Span’s smart panel absorb idle power capacity from the power grid, so that AI cloud service providers can draw energy, connect, and tap into the compute network. Homeowners only need to pay a fixed fee in exchange for power and Wi-Fi, and can receive corresponding compensation based on the extent of energy and network usage in practice—effectively creating a source of passive income for households.

One-fifth the cost, six times the speed—distributed architecture overturns the traditional model

Span CEO Arch Rao emphasized that the essence of the plan is infrastructure layout. He said the company has unique advantages to meet the market’s massive demand for compute at a lower cost, while also enabling individual consumers to genuinely benefit.

On the data side, the advantages of this distributed solution are significant. Span claims that deploying 8,000 nodes is six times faster than building centralized data centers of the same scale (100 megawatts), and costs only one-fifth as much. Once the node network reaches a sufficient scale, its overall compute capacity can match that of traditional mid- and small-sized data centers, greatly reducing the need to build new large server rooms. According to a report from the Pew Research Center, the electricity consumption of a traditional data center is equivalent to that of 100,000 households, highlighting the pressure on power grids and the environment.

Teaming up with PulteGroup, one of the largest homebuilders in the U.S.—new community trials first

To accelerate adoption, Span further partnered with PulteGroup, one of the largest homebuilders in the U.S., to pilot the deployment of XFRA nodes in new residential communities. Multiple communities have already completed installations and entered the practical operations stage.

This strategy not only helps quickly scale up the node network, but also allows new home buyers to participate in the energy-sharing mechanism from day one after moving in. Span said that if the technical validation succeeds, the system could also help avoid overloads on local power grids, while preserving land resources for housing or other uses rather than continuing to build massive traditional data centers.

（Trump pushes “emergency power auctions,” requiring AI tech giants to contribute $15 billion to build new power plants）

Miniaturized and decentralized—it could become the new norm for AI infrastructure

Traditional large data centers have long raised concerns from surrounding communities about water use, electricity use, and heat emissions. Span’s distributed mini data center solution targets exactly these issues by spreading compute demand across thousands of residential nodes, maintaining the massive compute capacity needed to sustain AI industry expansion without concentrating environmental burdens.

As the scope of AI applications continues to expand, the way compute infrastructure is built is bound to change. Nvidia and Span’s collaboration may be only the beginning, hinting that in the future, every building could become a node in the global AI network.

Has it stolen the compute from residential power shortages? Nvidia teams up with smart-meter startup Span—mini AI data centers installed on your home’s exterior walls. First appeared on Chain News ABMedia.