Sunrun is taking a familiar energy-industry idea—use the power you already generate and store—and applying it to one of the most power-hungry technologies on the planet: large-scale AI computing. In a move that looks less like a traditional data-center expansion and more like a bet on “distributed” infrastructure, the company has launched a pilot program designed to place AI compute nodes inside customers’ homes. The homes in question aren’t random. They’re specifically households equipped with Sunrun solar panels and battery storage, which means the compute hardware would be operating in an environment already built around managing electricity supply and demand.
The concept is straightforward, but the implications are not. Instead of building another centralized facility packed with servers, cooling systems, and high-capacity grid connections, Sunrun wants to turn participating homes into small pieces of a larger compute network. Customers would host the equipment and, in return, be compensated for participating in the pilot. Sunrun then plans to sell the resulting distributed compute capacity to “enterprise compute buyers,” including organizations that need additional resources for AI workloads.
This is not the first time the industry has talked about moving compute closer to where power is available. But it’s a notable twist because Sunrun is coming from the energy side, not the cloud side. That matters, because the biggest bottlenecks for AI infrastructure are increasingly not just about chips and software—they’re about electricity, grid constraints, and the practical challenge of keeping systems running reliably without turning every new deployment into a major construction project.
What Sunrun is effectively proposing is a hybrid model: home-based compute nodes paired with home-based energy assets, coordinated in a way that can potentially smooth out the mismatch between when electricity is generated, when it’s stored, and when compute demand needs to be met. If it works, it could represent a new kind of “infrastructure bundling,” where the same household solar and battery system that reduces utility bills also becomes part of the supply chain for enterprise AI compute.
The pilot’s framing emphasizes “distributed AI compute.” That phrase is doing a lot of work. Distributed compute can mean many things—from edge devices running small models to clusters spread across multiple locations. In this case, the compute nodes are intended to be meaningful enough to attract enterprise buyers, which suggests they’re not just running trivial tasks or lightweight inference. They’re positioned as a resource that can be aggregated and sold, implying a level of orchestration and performance management that goes beyond simply plugging in a device and hoping it contributes useful work.
Why Sunrun is choosing homes with solar and batteries
There’s a reason Sunrun isn’t targeting any home with an outlet. Solar generation is intermittent by nature, and batteries are what make it usable at the times when it’s needed. AI compute, meanwhile, tends to be power-hungry and often time-sensitive. Even if a workload can be scheduled flexibly, the system still needs a dependable baseline of electricity and a way to handle peaks.
By focusing on homes already equipped with solar and battery storage, Sunrun is starting with a power profile that can be managed. Batteries can buffer short-term fluctuations, and solar can reduce reliance on the grid during daylight hours. In theory, that combination could allow compute nodes to run in a way that is both economically attractive and operationally feasible.
But there’s another angle that’s easy to miss: batteries are also a control surface. A centralized data center is constrained by its location and its connection to the grid. A distributed system, by contrast, can potentially shift demand across many sites. If Sunrun can coordinate compute scheduling with energy availability—using stored energy when it’s available and throttling or pausing when it isn’t—it may be able to offer a more resilient service than a purely grid-dependent setup.
That doesn’t eliminate the hard problems, though. It just changes where they show up. Instead of building a single facility that must be cooled and powered continuously, Sunrun is distributing the burden across many homes, each with its own constraints: limited space, varying household behavior, different local grid conditions, and the reality that customers don’t want their living rooms turned into server rooms.
The customer experience: compensation is only the beginning
Sunrun’s pilot includes compensation for customers who participate. That’s likely necessary, because hosting compute hardware in a home introduces tradeoffs that go beyond money. Even if the equipment is designed to be quiet and unobtrusive, customers will still be dealing with installation, monitoring, and the possibility of downtime. They may also have concerns about noise, heat, internet connectivity, and whether the hardware affects their home energy system’s performance.
There’s also the question of how much control customers have. In a typical solar-plus-storage setup, the homeowner expects the system to behave predictably: charge when conditions allow, discharge when it’s beneficial, and maintain a certain level of resilience during outages. Adding compute nodes could complicate that behavior unless Sunrun’s software layer is carefully designed to prioritize household needs.
If the compute node draws power during periods when the battery would otherwise be used for backup, the homeowner might lose some of the value they bought. If the compute node is throttled too aggressively, enterprise buyers might not get the reliability they expect. The pilot’s success will likely depend on striking a balance: ensuring that the home remains comfortable and protected, while still delivering enough compute availability to make the service commercially meaningful.
In other words, compensation alone won’t determine adoption. Trust and transparency will.
Reliability and scheduling: the hidden engineering challenge
AI workloads are often described as flexible, but in practice they can be sensitive to performance and uptime. Training jobs, fine-tuning runs, and large inference pipelines all have different tolerance levels for interruptions. Some workloads can be checkpointed and resumed; others can’t. Even when interruptions are technically possible, frequent disruptions can increase total runtime and cost.
A distributed compute network must therefore solve two problems at once: it needs to deliver enough consistent capacity to be useful, and it needs to manage variability across many homes. Solar output varies by weather and season. Battery state varies by household usage patterns. Internet connectivity varies by neighborhood. Hardware health varies over time.
Sunrun’s pilot will likely require sophisticated orchestration—software that can allocate tasks to nodes based on current energy availability, predicted availability, and performance characteristics. It also needs to handle failures gracefully. If a node goes offline because a customer’s battery is reserved for backup, or because the household’s internet connection drops, the system must reroute work without causing enterprise buyers to lose time.
This is where the “distributed” approach can become either a strength or a liability. Centralized data centers are easier to manage because they’re uniform and controlled. Distributed systems are harder because they’re heterogeneous. But distributed systems can also be more scalable in a different sense: instead of waiting for new construction, you can add capacity by onboarding more homes.
Security and privacy: compute in your home is not a neutral idea
Placing compute hardware in residential environments raises security questions that go beyond standard enterprise concerns. Data centers are designed with physical security, controlled access, and hardened infrastructure. Homes are not. Even if the hardware is tamper-resistant and monitored remotely, the threat model changes.
There are also privacy considerations. If the compute nodes are processing enterprise workloads, they may handle sensitive data. That data must be protected in transit and at rest, and the system must ensure isolation between tenants. In a distributed environment, the risk of misconfiguration or leakage can increase if the orchestration layer isn’t extremely robust.
Then there’s the customer’s perspective: homeowners may worry about what’s happening inside their home. Are there cameras? Are there microphones? Is the system collecting energy usage data? Is it monitoring household activity? Even if the answer is “no” or “only for operational telemetry,” the perception matters. For a pilot to scale, Sunrun will need to communicate clearly what data is collected, why it’s collected, and how it’s secured.
The company’s ability to meet enterprise-grade security expectations will likely be a deciding factor in whether this becomes more than a novelty.
The grid and energy markets: a new kind of demand response
One of the most interesting aspects of this pilot is how it could intersect with energy markets and grid stability. Solar and battery systems are often discussed in terms of reducing emissions and lowering costs. But they also play a role in demand response—shifting energy use to times when the grid is under less stress.
If Sunrun can coordinate compute demand with energy availability, the distributed compute network could function like a flexible load. That means it might be able to ramp up when solar generation is strong and ramp down when it isn’t, potentially reducing strain on the grid. In regions where grid capacity is constrained, that flexibility could be valuable.
However, the relationship between compute demand and energy availability is not automatically beneficial. If the system is designed to maximize compute output regardless of grid conditions, it could still create peak loads. The pilot’s design choices—how it schedules workloads, how it uses batteries, and how it responds to grid signals—will determine whether it behaves like a stabilizing resource or just another consumer of electricity.
There’s also the question of incentives. Sunrun’s compensation to customers is one incentive. Enterprise buyers paying for compute is another. But the broader energy ecosystem may introduce additional incentives or requirements, such as participation in demand response programs or compliance with local regulations. The pilot could become a bridge between the AI infrastructure market and the energy flexibility market, which is a relatively new intersection.
A unique take on “where compute comes from”
The usual story about AI infrastructure is that companies need more space, more power, and more cooling, so they build data centers or sign long-term deals for capacity. Those projects are expensive and slow, and they often face permitting and grid interconnection delays.
Sunrun’s approach reframes the problem. Instead of treating compute as something that must be centralized, it treats compute as something that can be distributed across existing energy-enabled sites. That doesn’t remove the need for power—AI still needs electricity—but it changes the logistics of getting that electricity to