AI computing power from the front yard: Start-up relies on decentralized servers

Contents

A California-based start-up called SPAN aims to meet the immense demand for AI computing power with a decentralized concept. Private households are supposed to hang white boxes on their house walls for this purpose. The power consumption will thus be distributed over a large area, and homeowners can expect financial benefits. A pilot project with 100 households is scheduled to begin later this year.

The business idea is intended to solve the problem of the slow expansion of AI data centers. These require suitable locations and, above all, a sufficiently dimensioned connection to the power grid, which, due to bottlenecks and high demand, already leads to years of waiting times for grid connection.

In addition, there is growing resistance to new data centers in the USA and other countries. Residents fear, among other things, disruptions to their power supply – and rightly so, as experts warn that the costs for grid expansion will ultimately fall on consumers. In the USA, for example, Maine tried to be the first state to limit the expansion of data centers – however, the governor vetoed the plan.

Utilizing idle grid capacity

The concept of the “Distributed Data Center Solution” from the San Francisco-based start-up, on the other hand, involves interconnecting thousands of decentralized, small server cabinets to form a total computing capacity. The liquid-cooled units are to be mounted on the outside of houses and operated with unused grid potential and solar power in the respective residential areas. According to the provider's surveys, only 40 percent of the peak load capacity is used on average.

Homeowners benefit from a favorable flat rate for electricity and internet of 150 US dollars per month. For nodes with particularly high utility value, homeowners can even receive electricity and internet free of charge. SPAN handles the installation and sells the computing power to AI customers.

First larger test in the second half of the year

Each node contains Dell PowerEdge servers with 16 Nvidia RTX PRO 6000 Blackwell GPUs, 4 AMD EPYC CPUs, and 3 TB of RAM, connected via a 24-port Gigabit switch. Hyperscalers and AI cloud providers can access the distributed network just as they would a classic data center.

Whether the concept will actually work remains to be seen. So far, it has only been tested in one house. In the second half of 2026, initial tests with 100 households are planned. Large-scale expansion could already occur in 2027, for example, in cooperation with construction companies.

Cheaper than traditional data centers?

For buyers of computing power, the distributed network is expected to be profitable: The company promises six times faster expansion of computing power at one-fifth the cost of a comparable central 100-megawatt data center. Specifically, this would cost around 15 million US dollars per megawatt and take three to five years to build. With 8,000 households, the same computing power could be achieved in about half a year for only 3 million US dollars per megawatt.

The connection of housing and data centers is not new. Alternatives are also being sought at sea: The start-up Panthalassa, for example, wants to operate floating computing nodes with wave energy. So far, however, the focus has been on using waste heat to heat houses. In Finland, for example, 250,000 households are to benefit from the waste heat of a Microsoft data center. In Great Britain, the start-up Heata has installed cloud servers in private homes, the heat from which is fed into hot water storage tanks. Homeowners receive free hot water.

(mki)