DeepSeek begins direct integration for large-scale deployments: those with 2,000 GPUs can contact the official team.

Beating AI News Flash: In the open-source announcement for V4.1 Flash, DeepSeek rarely made a direct call to large-scale deployers: if you have 2,000 GPUs and a storage cluster, you can contact DeepSeek directly.

But this is not a new commercial licensing threshold. V4.1 Flash is still open-sourced under the MIT License, with no additional licensing conditions set based on company revenue or business scale. Enterprises with resources can deploy it themselves, and contacting DeepSeek is more like direct technical cooperation for ultra-large-scale scenarios. DeepSeek has also not currently announced a formal private deployment product or paid service.

In the past, after DeepSeek open-sourced its models, deployment adaptation was mostly completed by inference frameworks such as SGLang and TensorRT-LLM and by hardware vendors. This time, the official team proactively left a direct contact method for large-scale deployment and said it will continue to support the open-source community in doing inference adaptation for new models, attempting to further expand deployment coverage.