Depends on whether you measure workloads as "jobs" or "flops". If "flops", I would hazard that the bulk of computing on the planet right now is happening on GPUs.