Phone Cluster Grades Class Work Faster Than Cloud

A 20-phone local setup handled peak grading for a 75-plus student class, with lower latency than the default Amazon Web Services backend.

Updated on Nov. 6, 2026 in Data Centers

Isometric editorial illustration of a neat grid of twenty dark smartphone blocks on a pale bench, cables gathering into one bundle, in muted oxblood and teal. — Researchers at the University of California, San Diego used 20 consumer phones linked locally to grade work for a 75-plus student class at peak demand, beating the default Amazon Web Services backend on latency. AI Illustration. Upload story photo >

A controlled cluster of 20 smartphones processed peak grading requests for a class of more than 75 students, using an organized local setup rather than phones spread across the internet. Grading latency was lower than with the default Amazon Web Services backend.

Why it matters

The test suggests that idle consumer phones could supplement conventional server capacity for certain workloads. It also shows the approach depends on local organization and stable connectivity rather than open distributed networks.

The cluster served a class of more than 75 students on 20 phones, while Acurast's network includes more than 318,000 phones across more than 175 countries. Servers still offer more processor cores and memory than phones, and the source does not quantify the latency gap in detail.

The players

University of California, San Diego

A public research university in San Diego that operates the grading cluster test described in the story.

Amazon Web Services

Amazon's cloud computing platform, whose default backend was used as the latency comparison in the test.

Acurast

A company operating a decentralized network that uses consumer smartphones to provide computing capacity.

The details

Researchers at the University of California, San Diego arranged 20 phones into a local computing job to grade student work during peak demand. Acurast, which operates a network of consumer phones, says its provider setup requires a stable internet connection and a power source.

May 2025: Controlled image-recognition tests of smartphone clusters were published.
Sept. 29, 2026: Acurast announced its network of over 318,000 phones.
Fall 2026: A 2,000-phone deployment is planned for the university project, with capacity for 100 similar classes.

The Tech Race

The project extends Acurast's decentralized network of more than 318,000 phones into a specific academic workload. It tests whether consumer hardware can compete with cloud servers, which still hold advantages in processor cores and memory.

Phone owners could eventually have idle devices contribute computing capacity while charging, though the approach requires a stable connection and power source. Students and instructors would see the effect only indirectly, through faster grading turnaround if such systems are adopted.

The takeaway

Distributed phone networks may handle specific tasks well, but they depend on careful local setup and reliable connectivity. Organizations weighing them should compare latency and reliability against conventional cloud options for their own workloads.