Loading page…
The company is making its stack easier to use with prominent Chinese open models, but the potential limits it flagged have not been defined as a final rule.
Listen to this story
The policy risk is now attached to Nvidia’s optimization layer, not merely whether Chinese models can be downloaded. Nvidia has added or announced support across RTX and DGX systems for models including Qwen3.8-27B, DeepSeek V4 Flash, and GLM 5.2, helping developers run open-weight software on Nvidia infrastructure.
August hardware-stack update gave Qwen3.8-27B day-zero support on RTX GPUs.
DGX Spark updates simplify clustering systems for DeepSeek V4 Flash and Z.ai’s GLM 5.2.
Nvidia specifically named DeepSeek, Qwen and Kimi when discussing possible U.S. support limits.
Nvidia is deepening support for Chinese open AI models while warning investors that prospective U.S. restrictions on such support could materially affect its business.
Nvidia highlighted a local AI initiative with optimizations for open models including DeepSeek V4 Flash and Alibaba’s Qwen 3.8, alongside systems from Google and Nvidia. Its August hardware-stack update also provided day-zero support for Qwen3.8-27B on RTX GPU systems.
The distinction matters because open models can be downloaded, modified and self-hosted. That differs from leading closed systems made by OpenAI and Anthropic, and makes model compatibility a practical consideration for organizations running models on their own infrastructure.
In its second-quarter disclosure, Nvidia said the U.S. is considering limits on its ability to support third-party applications and models built on China-originated open-source foundation models. It named DeepSeek, Qwen and Kimi as examples. No final restriction was identified, so the affected products, services and applications remain undefined.
Chinese chipmakers including Huawei and Alibaba have also announced recent optimizations for open models in the DeepSeek and Qwen series. An unnamed Nvidia employee told CNBC that developers tend to choose the technology stack optimized for their model, framing Nvidia’s work as an effort to keep those developers on an American stack.
Forrester analyst Charlie Dai said the work reflects the growing influence of Chinese models and strengthens Nvidia’s position as an infrastructure layer regardless of where leading models originate. Nvidia, Microsoft, Meta, Palantir and more than 20 other companies also urged policymakers in July to avoid premature restrictions on open-weight models.
Editorial analysis
Our view: Nvidia is making a platform bet under a policy cloud. Day-zero RTX support for Qwen3.8-27B and easier DGX Spark clustering for DeepSeek V4 Flash are concrete ways to make Nvidia systems more attractive to teams selecting those models. Yet Nvidia’s disclosure frames the prospective constraint broadly, around products and services that support third-party applications and models, without identifying a final rule. The next meaningful signal is whether any U.S. action specifies which support functions, products, or applications it would cover.
Loading discussion...
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.