Transforming a retired Pixel phone into a cluster node is a hands-on, hardware-first process. The goal is to isolate the computing essentials and strip away everything unnecessary for server operation.

• Hardware stripping: Workers disassemble each phone to remove the display, battery, camera, and outer casing . What remains is the motherboard, which houses the processor, memory, and storage—the only components needed for computation.
• Carbon rationale: The focus on the motherboard isn't arbitrary. According to Google, roughly 50% of a phone's manufacturing carbon footprint is embedded in the motherboard alone. Reusing it captures the most carbon value while discarding the rest .
• Operating system replacement: With the phone stripped physically, the Android operating system is wiped and replaced with a general-purpose Linux distribution. This removes all mobile-only constraints and lets the device act like any standard Linux machine .
• Orchestration: Once running Linux, each motherboard is containerized and managed as part of a unified cluster using Kubernetes and Docker . This software stack enables both serverless workloads and high-performance computing applications to run across the phone cluster .

This approach builds on years of prior research. Earlier prototypes, including a project called "Renée," replaced Android with Ubuntu Touch on smaller clusters of used phones to provide Function-as-a-Service capabilities, proving the broader feasibility of the idea .

Performance: Old Phones, New Server Equivalents

Despite their age, the stripped-down phones pack surprising compute density. The project uses the SPEC benchmark to measure throughput, and the results provide a clear equivalence.

By the SPEC benchmark, 25–50 phone motherboards match the compute performance of one modern server . Extrapolating from this ratio, a full 2,000-phone cluster is expected to provide roughly the equivalent of 40–80 servers’ worth of compute, all without manufacturing any new chips .

Real-world testing has already validated this concept. An early 20-phone test cluster handled grading for a 75-student class faster than a small cloud server, demonstrating that the approach works for immediate, practical tasks . Previous research also found that a small cluster of decommissioned smartphones can—at a significantly lower carbon footprint than traditional cloud computing—approximately match and sometimes even exceed the performance of a new server when running benchmarking suites with synthetic workloads .

The Sustainability Angle: Carbon, E-Waste, and a New Metric

The project's environmental argument rests on three pillars: reducing embodied carbon, cutting e-waste, and measuring tradeoffs with a new metric.

• Embodied carbon: Manufacturing new servers carries a substantial carbon cost at the factory level. By reusing hardware that still has useful compute capability, the project directly reduces the need to produce new equipment .
• E-waste reduction: Smartphones often sit unused in drawers or are shredded for recycling while their processors still work perfectly. The Junkyard Computing project extends the useful life of these devices, keeping functional silicon in service for years beyond its typical mobile lifespan . Earlier research notes that smartphone processors can run reliably for at least six years in data center conditions .
• Computational Carbon Intensity: To make these tradeoffs measurable, the research team introduced "Computational Carbon Intensity" (CCI). CCI is a performance metric that balances the continued service of older devices against the superlinear carbon cost of building new ones, providing a framework to decide when reuse makes environmental sense .

What Happens This Year: The UC San Diego Deployment

The 2,000-phone cluster isn't just a lab prototype—it has a concrete mission on a university campus starting Fall 2026.

• Coursework: The cluster will run real computer science classes at UC San Diego, including courses in Parallel Computation and Systems Programming . Students will interact with and program against a distributed system built from consumer hardware, giving them hands-on exposure to clusters at scale.
• Reliability research: Beyond coursework, the deployment doubles as a long-term reliability study. The research team will observe how consumer-grade phone motherboards hold up under continuous, full-load data center-style usage patterns . The findings could influence how companies and institutions treat the vast, growing supply of used but still-capable devices.

Transforming a retired Pixel phone into a cluster node is a hands-on, hardware-first process. The goal is to isolate the computing essentials and strip away everything unnecessary for server operation.

• Hardware stripping: Workers disassemble each phone to remove the display, battery, camera, and outer casing . What remains is the motherboard, which houses the processor, memory, and storage—the only components needed for computation.
• Carbon rationale: The focus on the motherboard isn't arbitrary. According to Google, roughly 50% of a phone's manufacturing carbon footprint is embedded in the motherboard alone. Reusing it captures the most carbon value while discarding the rest .
• Operating system replacement: With the phone stripped physically, the Android operating system is wiped and replaced with a general-purpose Linux distribution. This removes all mobile-only constraints and lets the device act like any standard Linux machine .
• Orchestration: Once running Linux, each motherboard is containerized and managed as part of a unified cluster using Kubernetes and Docker . This software stack enables both serverless workloads and high-performance computing applications to run across the phone cluster .

This approach builds on years of prior research. Earlier prototypes, including a project called "Renée," replaced Android with Ubuntu Touch on smaller clusters of used phones to provide Function-as-a-Service capabilities, proving the broader feasibility of the idea .

Performance: Old Phones, New Server Equivalents

Despite their age, the stripped-down phones pack surprising compute density. The project uses the SPEC benchmark to measure throughput, and the results provide a clear equivalence.

By the SPEC benchmark, 25–50 phone motherboards match the compute performance of one modern server . Extrapolating from this ratio, a full 2,000-phone cluster is expected to provide roughly the equivalent of 40–80 servers’ worth of compute, all without manufacturing any new chips .

Real-world testing has already validated this concept. An early 20-phone test cluster handled grading for a 75-student class faster than a small cloud server, demonstrating that the approach works for immediate, practical tasks . Previous research also found that a small cluster of decommissioned smartphones can—at a significantly lower carbon footprint than traditional cloud computing—approximately match and sometimes even exceed the performance of a new server when running benchmarking suites with synthetic workloads .

The Sustainability Angle: Carbon, E-Waste, and a New Metric

The project's environmental argument rests on three pillars: reducing embodied carbon, cutting e-waste, and measuring tradeoffs with a new metric.

• Embodied carbon: Manufacturing new servers carries a substantial carbon cost at the factory level. By reusing hardware that still has useful compute capability, the project directly reduces the need to produce new equipment .
• E-waste reduction: Smartphones often sit unused in drawers or are shredded for recycling while their processors still work perfectly. The Junkyard Computing project extends the useful life of these devices, keeping functional silicon in service for years beyond its typical mobile lifespan . Earlier research notes that smartphone processors can run reliably for at least six years in data center conditions .
• Computational Carbon Intensity: To make these tradeoffs measurable, the research team introduced "Computational Carbon Intensity" (CCI). CCI is a performance metric that balances the continued service of older devices against the superlinear carbon cost of building new ones, providing a framework to decide when reuse makes environmental sense .

What Happens This Year: The UC San Diego Deployment

The 2,000-phone cluster isn't just a lab prototype—it has a concrete mission on a university campus starting Fall 2026.

• Coursework: The cluster will run real computer science classes at UC San Diego, including courses in Parallel Computation and Systems Programming . Students will interact with and program against a distributed system built from consumer hardware, giving them hands-on exposure to clusters at scale.
• Reliability research: Beyond coursework, the deployment doubles as a long-term reliability study. The research team will observe how consumer-grade phone motherboards hold up under continuous, full-load data center-style usage patterns . The findings could influence how companies and institutions treat the vast, growing supply of used but still-capable devices.