Old Phones Deemed Useless, Suddenly Repurposed as Data Centres

Researchers from the University of California San Diego (UCSD), in collaboration with Google, are recycling old Pixel phones into low-cost data centres. The project aims to extend the lifespan of unused devices and reduce electronic waste. Google Research explained that retired phones represent embodied carbon, which is the carbon emissions generated during the device manufacturing process. Given that many people replace their phones every few years, this habit is a major contributor to electronic waste. Consequently, the UCSD team is striving to give these devices a second life as versatile computing platforms. Research findings indicate that smartphones released around three years ago still possess higher single-core performance than servers like the Asus RS720A-E11, which can be fitted with Nvidia H200 or Nvidia RTX Pro 6000 GPUs and two AMD EPYC server processors. Although the server’s overall performance is much higher, SPEC benchmark scores show that the processor cores in old phones are still quite capable of handling various computing tasks. To achieve this, the researchers removed unnecessary components such as the screen, battery, camera, speakers, and casing. Only the motherboard was retained, as it houses the system-on-chip (SoC) that performs the computing processes. The Android operating system was then replaced with a Linux distribution commonly used in data centres, eliminating unnecessary pre-installed applications and enabling the use of orchestration software. Testing showed that approximately 25 to 50 used phones can produce computing power equivalent to a single dual-socket server processor. UCSD also found that a cluster of 20 phones can run an application required by a class of more than 75 students. This approach eliminates the need to run applications on cloud services, which incur additional costs and require larger data centre resources. The research team plans to utilise around 2,000 used phones to build a local data centre capable of supporting a hundred classes simultaneously. Besides enabling applications to run locally on owned hardware, the team stated that the construction cost is only a fraction of the usual expense, especially amidst rising memory and storage chip prices. They are targeting a full launch of the system by the end of the year, and the research will also test the durability of consumer phone components when operated continuously as part of a data centre. However, the researchers assessed that giant AI data centres are unlikely to switch to servers based on used phones, as large operators generally prefer specialised hardware that is more reliable and has fewer components. Instead, this solution is considered highly suitable for universities, educational institutions, and small organisations with limited budgets that struggle to purchase new servers or compete with large technology companies.