DeepSeek open-sources the inference engine behind its API. And Huawei gets equal billing with Nvidia.

DeepSeek's decision to open-source its inference engine makes the technology it uses for serving models in production available to a wider audience. This can reduce the barriers for other developers and companies looking to deploy large language models, as they can now utilize DeepSeek's proven infrastructure rather than developing their own from scratch.

A notable aspect of this release is the explicit support for Huawei's Ascend hardware. By including Ascend alongside Nvidia's GPUs, DeepSeek is acknowledging and potentially broadening the viable hardware options for AI inference. This could be particularly relevant in regions where access to Nvidia hardware might be restricted or where there is a strategic interest in diversifying chip suppliers.

The open-sourcing of such a core component can foster greater collaboration and innovation within the AI community. When the underlying inference technology is transparent and accessible, developers can inspect, modify, and improve it, potentially leading to more efficient or specialized deployments. This also provides a practical demonstration of how a major AI player is handling its production-level inference needs.