Future Grid turns customer devices into AI compute

Future Grid is pitching enterprises on a new way to run AI inference by tapping the idle phones, tablets and laptops their customers already own. The London company says the approach can cut cloud costs, reward users for participation and reduce the need for new data centre capacity as AI power demand surges.

Why it matters: - Cloud AI spending is climbing fast, with the nine largest cloud providers on track to spend close to $830 billion in capital expenditure in 2026, and the four largest US tech companies guiding to about $725 billion. - Future Grid is aiming to shift some AI inference off expensive hyperscale infrastructure and onto opted-in customer devices. - The model could lower enterprise compute bills, create a new reward stream for users and reduce demand for new data centre buildout. - The approach also targets a growing power problem. The International Energy Agency projects data centre electricity demand will more than double to around 945 TWh by 2030.

What happened: - Future Grid said on July 13, 2026, that it is building a platform that lets enterprises run AI workloads on the idle devices of their own customers. - The London-based company says the system is designed for enterprises with app-centric customer bases and rising AI compute bills. - Founder and CEO Jonathan MacDonald said the company is seeking design partners across energy, transport, telecommunications, banking, retail and media. - The company said partners will embed its SDK, measure opt-in and run bounded pilots to document savings against a hyperscaler baseline.

The details: - Enterprises create a private network and invite customers to opt in inside the company’s own app. - A lightweight node runs on surplus CPU and is throttled so the device stays responsive for the owner. - The company routes inference and other batchable AI tasks to those devices first. - Work stays inside a private network by default, with the customer’s nodes, jobs and brand. - Execution is verified automatically with auditable proof of execution. - The enterprise is billed against the hyperscaler cost the system displaces, described as a fraction of the cloud bill. - Users can receive bill credits, loyalty points or direct payment funded from the savings. - Future Grid says the platform includes orchestration, verification, rewards and settlement for pilot-to-production deployments. - The company says its platform is patent-pending. - Future Grid is built on Celestial, which handles identity, verification and settlement as work moves between company systems and user devices. - For enquiries, the company listed more information. - Future Grid said its statistics come from published sources listed at the company’s sources page.

Between the lines: - The strategy depends on a large pool of idle consumer devices, not new hardware. - Future Grid argues that inference is especially suitable because it accounts for an estimated 80% to 90% of AI energy use and can be distributed across ordinary machines. - The company also says the environmental case improves because idle devices already exist, which avoids the embodied carbon of manufacturing new servers. - Future Grid cites research showing a modern smartphone processor core can match or beat a data centre server core in some cases. - The company also points to carbon-aware scheduling research showing emissions cuts of more than half versus a carbon-agnostic baseline. - That positioning frames Future Grid as both a cost play and a sustainability play, which could matter to enterprises under pressure on both budgets and emissions reporting.

What's next: - Future Grid is looking for enterprises to run bounded pilots and prove savings against hyperscaler costs. - The company expects partners to co-shape reward design and success metrics during those pilots. - Future Grid says each job will carry a measured carbon record that enterprises can use in ESG and sustainability reporting. - The company says participation will stay under the control of device owners, with privacy and control remaining with the user.