AMD is preparing to launch the Radeon AI PRO R9700, a new graphics processor designed specifically to comply with recent United States export controls while serving the unique demands of the Chinese Artificial Intelligence and research market. The GPU features intentionally reduced compute performance to stay below the export regulation´s compute throughput threshold, ensuring eligibility for sale in China. Despite its limited raw processing power, the Radeon AI PRO R9700 retains 32 GB of high-bandwidth memory and full PCIe Gen 5 support, enabling robust data transfer and facilitating strong aggregate performance when multiple units operate in tandem. This architecture is tailored for on-device inference and model fine-tuning, supporting enterprise and research workloads that prioritize data privacy and compliance.
This strategic release comes after a period of market turbulence for AMD in China, highlighted by financial setbacks linked to previous regulatory blocks on its MI308 accelerator. AMD incurred significant charges after the MI308 was barred from export, similar to competitor NVIDIA, which suffered substantial lost revenue and a dramatic reduction in its market share in China following restrictions on its H20 series accelerators. These developments have shifted the competitive landscape, allowing AMD an opportunity to appeal to Chinese buyers with a combination of price competitiveness and more dependable supply chains.
Mirroring NVIDIA´s introduction of the performance-limited B20 Blackwell-series GPU for the Chinese market, AMD seeks to address customer demand for optimal local deployment of Artificial Intelligence infrastructure while navigating regulations. The Radeon AI PRO R9700 is slated for release in the third quarter of 2025, timed alongside NVIDIA´s B20. AMD plans to reveal its comprehensive Artificial Intelligence silicon roadmap at the Advancing AI Summit on June 12, positioning the R9700 as an accessible solution for enterprises and academic institutions engaged in Artificial Intelligence research and innovation, rather than for use in large-scale cloud computing centers.
