AI computing is starting to move into our neighborhoods, and the equipment a street shares — especially the distribution transformer outside your house — will help decide whether that goes well.
Demand for AI inference keeps climbing, while new 100 MW to 1 GW data centers face growing local pushback over power prices, water, noise and land. One answer is to spread the computing across thousands of homes and small businesses. That market is now forming: SPAN announced XFRA with NVIDIA in April 2026, Sunrun launched a pilot in solar-and-battery homes in July, and in September the two companies partnered to build XFRA nodes into new residential communities.
Those efforts already orchestrate fleets of nodes, manage home energy, and pair compute with batteries to ease the strain on local distribution equipment. The open engineering question is how a fleet coordinates its combined demand against the loading and temperature limits of the transformer that several homes share. That is the problem Restored Technology is focused on.
The problem: one transformer, many homes
A residential distribution transformer is commonly rated around 25–50 kVA and serves several homes, often 5–10, though this varies by utility and neighborhood. Utilities can size it that way because households rarely peak at the same moment. Engineers call this load diversity.
A computing node drawing 1–1.3 kW around the clock removes that diversity. Put one on several homes behind the same transformer and the valleys fill in. Add an air-conditioning peak on a hot afternoon, and the transformer runs hot. Heat ages its insulation faster, and an overloaded transformer fails early.
Each home's own panel can show plenty of headroom while the shared transformer is running hot. Premises-level limits and shared-network limits are different constraints, and the second one needs coordinating across every node on the transformer.
Our approach: govern the neighborhood, not just the house
Our system treats every node served by the same transformer as one group with one shared power budget. It is patent pending (U.S. provisional application filed November 2026) and has four parts:
- Transformer-aware load governance. A coordinator estimates each transformer's loading and temperature and sets a group budget from that, after reserving room for the homes' normal use. Each node gets a power cap and is held to it. If a node loses contact, it drops to a conservative fallback power limit, and it shuts its computing off automatically when reliable operating limits can't be maintained.
- Grid-event-aware AI placement. Large AI models are split into pieces spread across homes, so no single transformer holds too much of any one model. When a utility announces a curtailment ahead of time, the system moves work off the affected homes before it starts, where unaffected capacity exists. If capacity runs short, it falls back to a smaller model, smaller batches, or queuing.
- Layered safety. In order of precedence: a lockable disconnect, then a hardware safety chain that works without software, then utility control, then workload control.
- Optional heat recovery. The node's heat could preheat a home's water heater, with computing scheduled when the grid, the tank and the transformer all have room. This will need thermal and plumbing validation.
These are designs. Performance, safety and code compliance still have to be proven through testing, which is the point of the pilot below.
Why it matters
| Who | What the design aims to offer |
|---|---|
| Utilities | New load managed against estimated transformer capacity, responsive to curtailment signals, with authenticated metering of what each node actually drew |
| Homeowners | Compensation under a hosting agreement, optional preheated hot water, and a lockable outdoor disconnect for electrical isolation |
| AI compute buyers | Distributed inference capacity, with work rescheduled or moved during grid events where unaffected capacity exists |
| Electricians and installers | A defined installation: dedicated circuit, outdoor lockable disconnect, sealed weatherproof enclosure |
What we're doing next
Our plan is to prove this in three steps, each small enough to finish and measure:
- Simulation on real data. Run the governance software against actual smart-meter and weather data for a few transformers. Show that compute is reduced or shed whenever estimated transformer thermal headroom runs short, and measure the useful compute hours that remain.
- One working node. A single outdoor node with the hardware safety chain, power capping and metering. Prove it trips without software, obeys a curtailment signal within seconds, and holds its power cap.
- One transformer, a few homes. Three to five nodes behind one shared transformer, run with a utility partner and subject to its approval and each site's electrical suitability. Measure transformer loading and temperature, how well nodes stay under their caps, how much work moves before a curtailment event, and hot water delivered.
Our goal is AI compute added to a street with no transformer overload, every curtailment honored, and hosts who would do it again. We'll measure that through curtailment compliance (how often and how fast nodes reach the requested level), transformer loading and estimated temperature against its rating, and availability (compute hours delivered versus planned).
Alongside the technology, Restored Technology is developing a standardized assembly and deployment model. Our vision is to manufacture and integrate residential computing nodes to infrastructure partners' specifications, then deploy and maintain them through trained regional installation teams. As the model proves itself, this could expand into a nationwide network of authorized service providers and franchise operators.
Who we're looking for
Restored Technology is a South Carolina company building AI software for the built environment. Our founder brings 26 years of experience in property restoration and building systems. To take this from patent filing to pilot, we want to talk with:
- Utility innovation and grid-edge teams willing to host a one-transformer pilot and share transformer data
- Homeowners in the Charlotte region interested in hosting a pilot node
- Licensed electricians and solar or battery installers who want to help design the installation
- Power-electronics, thermal and distributed-systems engineers to build the first nodes
- Investors and grant programs focused on grid flexibility and AI infrastructure
If that's you, reach Joshua Aponte, Founder & CEO of Restored Technology, at japonte@restoredtechnology.com or through our LinkedIn page.