A 20-something with a smartphone graded the class... It wasn't large-scale AI computation verification

Researchers at UC San Diego built a cluster of 20 repurposed smartphones to grade assignments for a class of more than 75 students during submission peaks, an experiment Google Research described in a June 12, 2026 post. The phones' circuit boards were extracted and colocated to form the cluster, which handled a specific, controlled workload rather than general-purpose AI computation. Google noted that while a smartphone's performance cores can rival individual server cores, servers still offer far more cores and much larger memory. The result matters as an early proof of concept that distributed consumer devices can absorb bursty, narrowly scoped tasks, but it does not show that scattered phones can replace large AI data centers. What to watch next is whether such clusters can scale to heavier workloads and how they compare with conventional servers on cost and throughput.