SageMaker Training Sessions let you write your own RL loop without provisioning GPUs

On October 22, 2026, AWS made the SageMaker AI Training Session APIs generally available: you orchestrate forward_backward(), optim_step(), and sample() on stateful, serverless GPU sessions, while the service handles hardware, gradient synchronization, and isolation. If you do SFT or DPO, the high-level APIs still suffice; if you do custom RL, this removes the GPU-ops tax.