Researchers from UC San Diego and Google have found a surprising second life for discarded smartphones: clustering them into functional data centers that rival traditional server performance on a per-core basis. By stripping away non-essential components and replacing Android with Linux, a group of 25 to 50 old phones can match the computing power of a single dual-socket server-class CPU, according to a new study.

Why Are Old Phones Better Computing Platforms Than You'd Expect?

The research team discovered that smartphones from just three years ago still deliver higher single-core performance compared to enterprise servers like the Asus RS720A-E11, which can be equipped with Nvidia H200 or Nvidia RTX Pro 6000 GPUs and two AMD EPYC server processors. While these servers excel at parallel processing tasks that mobile devices cannot handle, the per-core efficiency of older phones remains surprisingly competitive when benchmarked using the SPEC suite, a standard industry measurement tool.

This finding challenges the assumption that older hardware is simply obsolete. The researchers determined that a cluster of just 20 old phones could support one application that a class of 75 or more students requires, eliminating the need to host it on cloud infrastructure with its associated costs and energy consumption.

How to Build a Local Data Center From Recycled Smartphones

• Hardware Preparation: Strip the phones of non-essential components including displays, batteries, cameras, speakers, and chassis, leaving only the motherboard that hosts the system-on-chip (SoC) needed for computing tasks.
• Operating System Replacement: Replace the Android operating system with a general-purpose Linux distribution used in data center applications, which removes unnecessary bloat and allows deployment of orchestration software like Kubernetes.
• Clustering and Deployment: Connect multiple phones into a cluster and deploy applications locally rather than on cloud infrastructure, reducing both costs and reliance on external data centers.

The UC San Diego team plans to launch a full system later in 2026 using 2,000 phones to build a local data center capable of supporting "a hundred such classes at once," according to the research. The group is also investigating how consumer-grade components can withstand continuous use in data center applications, a critical question for long-term viability.

What Makes This Approach Cost-Effective for Educational Institutions?

The financial advantage is substantial. Building a local data center from recycled phones costs only "a fraction of the usual cost" compared to constructing one from new components, particularly relevant given today's elevated pricing for memory and storage chips. For universities and smaller organizations without the resources to compete with tech giants spending billions on infrastructure, this approach offers a practical alternative.

Beyond cost savings, institutions gain ownership of their hardware and can run applications locally without ongoing cloud service fees. This is especially valuable for educational use cases where computational demands are predictable and moderate, rather than the massive-scale workloads that hyperscalers like Google, Amazon, and Microsoft manage.

The environmental angle adds another dimension to the appeal. Retired smartphones represent significant "embodied carbon" from manufacturing, and humanity's habit of replacing devices every few years contributes substantially to electronic waste. Repurposing these devices as computing platforms extends their useful life and reduces the need to manufacture new server hardware.

Will Major Tech Companies Adopt This Model?

Despite the compelling economics, large AI hyperscalers are unlikely to switch to servers built from used phone components. These companies prioritize working with fewer, more specialized hardware components and the reliability guarantees that come with purpose-built enterprise equipment. The scale and consistency requirements of training large language models and running production AI systems demand hardware engineered specifically for those tasks.

This is not the first time researchers have explored giving old phones a second life. Another group of researchers converted old phones into "tiny data centers" last year, even using a set of four devices for underwater monitoring. NASA repurposed the Qualcomm 801 SoC, a mid-range chip from 2014 found in the Ingenuity Mars helicopter, to help the Perseverance rover navigate the Red Planet like a makeshift GPS system.

The UC San Diego research demonstrates that the computing performance gap between consumer devices and enterprise hardware is narrower than many assume, particularly when measured on a per-core basis. For institutions willing to invest engineering effort in repurposing old phones, the combination of cost savings, environmental benefits, and adequate performance makes a compelling case for local data center deployment.