Anyscale vs raw cloud GPU pricing - ACU rates vs per-GPU-hour

Fetched 2026-10-04. Anyscale ACU rates vs raw cloud providers: A100 $4.9591 ACU/hr (Anyscale) vs CoreWeave A100 $2.70/GPU-hr (on-demand) vs AWS p4d $4.10/GPU-hr. A10G $1.3635 ACU/hr (Anyscale) vs CoreWeave A10G not listed vs AWS g5.12xlarge $6.55/4 = $1.64/GPU-hr. L4 $0.9542 ACU/hr (Anyscale) vs CoreWeave L4 $1.25/GPU-hr. T4 $0.5682 ACU/hr (Anyscale) vs CoreWeave T4 not listed. Anyscale's value proposition is the Ray platform orchestration, not raw price competitiveness. The ACU model makes direct comparison difficult - an ACU is approximately 1 vCPU + 2GB RAM, so a GPU instance with 10 vCPU + 144GB RAM would be ~82 ACU, but the published rate is a single ACU/hr figure for the whole instance. Case studies claim 50-99% savings, but these are self-reported and likely compare against unoptimized cloud usage (idle GPUs, over-provisioning) rather than against optimized raw cloud rates.