A calculation based on DeepSeek inference throughput and its off-peak API prices puts theoretical billing at about $38 per GPU-hour, according to an October 24 BlockBeats report.
Throughput and business returns
The report said DeepSeek had open-sourced the DeepInfer inference framework and disclosed that its cluster could process 155 trillion tokens per day.
The $38 figure is a theoretical billing estimate, rather than reported revenue or profit. The report identified GPU utilization, the share of paid requests, hardware and electricity costs, and service latency as factors determining actual returns.
Source: BlockBeats, 2026-10-24 05:07:28 UTC.