AMD Targets China's AI Market With Free Cloud and 3 New Tools

Advanced Micro Devices is making a direct challenge to Nvidia’s dominance in China’s artificial intelligence sector, launching a free AI developer cloud and a suite of new hardware specifically tailored for the local market. The move, announced at its AI Developer Day in Shanghai, deepens AMD’s collaboration with top Chinese AI firms and aims to provide a full-stack alternative for developers constrained by high costs and supply bottlenecks for AI chips.

“China is the most vibrant AI ecosystem globally,” AMD Chief Executive Officer Lisa Su said at the event, noting some of the “best work in the world is happening here.” AMD already has over 4,000 engineers in China, which Su described as a core force driving the company’s product roadmap.

The strategic push includes three main components: a free, public AI developer cloud with Radeon GPUs; new hardware including the Ryzen AI Max+ 395 processor designed for running large models locally; and deep software integration with China’s leading open-source models from companies like 01.AI, Baidu, and Zhipu AI. The company also announced a partnership with 01.AI, founded by Kai-Fu Lee, to create an “enterprise intelligent agent all-in-one machine.”

This strategy directly addresses the biggest challenge in the generative AI boom: the scarcity and high cost of computing power. While Nvidia controls the vast majority of the AI chip market, long lead times and high prices have pushed major AI labs to seek alternatives. AMD is positioning itself as the leading option, a strategy validated by AI startups like Zyphra, which runs its models entirely on AMD hardware, citing cost and supply-chain advantages.

A Three-Stage Hardware Attack

AMD laid out a comprehensive hardware path for developers covering three stages: local development, scale testing, and full deployment. For initial development, the new Ryzen AI Max+ 395 processor features 128GB of unified memory, allowing developers to run 200-billion-parameter models on a local machine without costly cloud services.

For larger-scale testing, the company introduced the Radeon AI Pro R9700, a workstation GPU, paired with the Threadripper Pro 9000 CPU. AMD calls the Threadripper the “world’s fastest workstation CPU,” providing 128 PCIe 5.0 lanes to support multiple GPUs. This allows teams to simulate and optimize complex, multi-agent AI systems locally before incurring deployment costs.

“The winners of the next five years won't be the teams that rent the most GPU cloud hours, but the teams that design for efficiency from day one,” said Jack Huynh, AMD’s senior vice president and general manager of computing and graphics.

Open Software and Local Models Drive Adoption

The hardware is tied together by AMD’s ROCm, an open-source software platform that now supports over three million models from hubs like Hugging Face and China’s ModelScope. The company announced Day-0 support for top Chinese models, including those from Xiaomi, DeepSeek, and StepFun.

Zhu Yibo, co-founder and CTO of StepFun, demonstrated running their 196-billion-parameter Step 3.5 model on an AMD laptop. The model ran at nearly 100 tokens per second, a speed Zhu claimed was “faster than many cloud model APIs,” with zero API costs and enhanced data privacy. This ability to run powerful models locally could save Chinese developers millions in API call fees.

The focus on an open, efficient, and localized stack is AMD’s core bet. As 01.AI’s Kai-Fu Lee noted, limited hardware resources have forced China’s AI community to excel at engineering efficiency and open collaboration. By providing a viable, cost-effective alternative to Nvidia and deeply integrating with this ecosystem, AMD aims to become the foundational hardware for China’s next wave of AI innovation. The strategy reflects a broader industry trend, with major players like Anthropic and Meta also diversifying their infrastructure to include AMD chips.

This article is for informational purposes only and does not constitute investment advice.