Nvidia Optimizes Chips for Chinese AI Models Amid Regulatory Risk

Nvidia is expanding hardware support for Chinese artificial intelligence models while simultaneously cautioning investors about potential U.S. government restrictions that could harm its business, according to details first reported by CNBC.

The chip giant highlighted what it calls a "local AI initiative" that includes optimizations for DeepSeek's V4 Flash and Alibaba's Qwen 3.8 models. In August, Nvidia announced day-zero support for Qwen3.8-27B on its RTX GPU systems and made software updates to simplify clustering multiple DGX Spark systems for running models like Z.ai's GLM 5.2 and DeepSeek V4 Flash.

But in an SEC filing tied to its second-quarter earnings, Nvidia warned that the Trump administration is considering restrictions that would limit the company's ability to support "third-party applications and models built on open-source foundation models originating in China." The company said any regulatory controls limiting its ability to provide products supporting apps built on models "such as DeepSeek, Qwen or Kimi" could materially impact business.

Why it matters

This tension illustrates a fundamental conflict in U.S.-China tech competition: American companies need to support Chinese AI models to remain competitive globally, but Washington increasingly views those same models as national security threats. Chinese AI systems have made significant capability gains in 2026 and are typically cheaper than American alternatives, making them attractive to developers worldwide—particularly in developing countries where default technology choices could influence geopolitical alignment.

The race for developer ecosystems

Nvidia's strategy reflects competition with Chinese chipmakers including Huawei, which makes Ascend processors, and Alibaba—both of which have announced their own optimizations for open models in recent months.

An Nvidia employee told CNBC that providing support for models worldwide "allows developers to build on the American tech stack." The employee, who wasn't authorized to speak publicly, said developers using popular models will choose whichever stack the model is optimized for, "making the need to optimize for the American stack critical."

"Developers will play a crucial role in building the winning AI ecosystem," the employee said. "China has one of the largest populations of developers in the world, creating open-source foundation models. Every model should run best on the U.S. technology stack, encouraging nations worldwide to choose America."

Charlie Dai, VP and principal analyst at Forrester, told CNBC that Nvidia's optimization work reflects the "growing influence" of Chinese models within the global AI ecosystem. "By optimizing platforms such as DeepSeek and Qwen, Nvidia is reinforcing its position as the preferred infrastructure layer for AI developers and enterprises regardless of where leading models originate," he said.

Industry pushback on restrictions

Nvidia isn't alone in opposing government intervention. In July, Microsoft, Meta, Palantir and more than 20 other companies joined Nvidia in issuing a statement urging policymakers to avoid "premature restrictions" on open-weight models—systems that can be downloaded, modified and self-hosted, unlike closed systems from OpenAI and Anthropic.

Daniel Remler, senior fellow at the Center for a New American Security, warned CNBC that if Chinese AI technology becomes the default for developing countries, "those countries may be more likely to align themselves politically with Beijing and Chinese AI companies get a beachhead in their markets."

A White House official told CNBC that "unless officially announced by the Administration, any reporting about tariffs should be regarded as baseless speculation," responding to a Thursday report about potential new tariffs on U.S. semiconductors.

Nvidia shares rose Thursday after the company reported better-than-expected second-quarter results. The stock has climbed more than 10% in 2026.

These details were first reported by CNBC.