Google and UCSD to Build Low-Carbon Cloud Using 2,000 Retired Pixel Phones

According to Beating Monitor, Google researchers, in collaboration with a team from the University of California, San Diego (UCSD), are exploring phone cluster computing by repurposing the motherboards of 2,000 retired Pixel smartphones into a low-carbon cloud data center to provide affordable, low-carbon cloud computing resources for students and faculty. While smartphones typically have a four-year update cycle, the computational cores of retired devices remain functional, with CPU performance approaching that of modern server cores—demonstrating comparable results in certain single-threaded tests of SPEC CPU 2017. To adapt these devices for data center environments, the research team removed non-essential components such as screens, batteries, casings, and cameras, retaining only the motherboards. This is because motherboard manufacturing accounts for the majority of embodied carbon emissions in the device’s lifecycle, while also eliminating battery-related safety risks in data center settings. On the software side, the team replaced the original Android user space with a generic Linux distribution, bypassing mobile-specific memory management restrictions such as the “Low Memory Killer.” For hardware management, every 25 to 50 smartphone motherboards are grouped into self-managed clusters, with containerized applications scheduled via Kubernetes. Based on SPEC CPU 2017 test results, a cluster of 25 to 50 motherboards achieves overall CPU throughput comparable to that of a traditional server. Preliminary experiments show that a micro-cluster of 20 phones can smoothly handle peak submission loads from over 75 students in parallel computing courses, with evaluation tasks completing in approximately 50 seconds—even outperforming AWS cloud instances in grading latency. The team estimates that a system composed of 2,000 phones will deliver computational power equivalent to roughly 50 traditional servers, capable of simultaneously supporting hundreds of courses in systems programming and parallel computing. The full system is scheduled to launch in fall 2026 as a long-term test platform to evaluate the hardware reliability of consumer-grade devices under sustained high loads.