Fenghua No.3 and the Rise of Chinas Domestic GPUs: A New Era in AI Hardware

In 2025, China’s semiconductor industry has reached a critical inflection point with the emergence of domestically developed graphics processing units (GPUs) designed for artificial intelligence (AI) and high-performance computing applications. At the forefront of this movement is the Fenghua series of GPUs, developed by Innosilicon, symbolizing China’s push for technological self-reliance amid complex geopolitical landscapes and technological competition.

The Fenghua No.3 GPU: Technical Highlights and Innovations

Launched in late 2024 and gaining momentum through 2025, the Fenghua No.3 represents a significant step forward for China’s homegrown GPU capabilities. This GPU features a massive 112GB of high-bandwidth memory (HBM), a capacity that rivals cutting-edge international GPUs, enabling it to handle AI models ranging from 32 billion to over 600 billion parameters when scaled across multiple cards.

Built upon the open-source RISC-V architecture and designed with CUDA-compatible cores, Fenghua No.3 aims to lower adoption barriers by supporting software ecosystems initially developed for Nvidia GPUs. This compatibility opens doors for Chinese developers and enterprises to port existing AI frameworks to domestic hardware, a strategic advantage as China faces export restrictions on foreign technology.

Sponsored Banner

Fenghua No.3 supports the latest graphics APIs, including DirectX 12, Vulkan 1.2, and OpenGL 4.6, and it is China’s first GPU to implement hardware-accelerated ray tracing. This feature dramatically enhances graphical realism in games, professional visualization, and simulation, particularly crucial for industries like media production and CAD.

The GPU supports simultaneous output to six 8K displays, albeit at a 30Hz refresh rate, which is sufficient for many professional applications including cinematic editing and large-scale visualization. Additionally, Fenghua No.3 natively supports DICOM standards essential for medical imaging, allowing it to display high-fidelity MRI, CT, and X-ray scans on standard monitors—significantly reducing the cost of medical diagnostics infrastructure.

Strategic Importance in the Context of US-China Technological Tensions

The rise of Fenghua and similar domestic GPUs must be understood against the backdrop of growing US restrictions on China’s access to advanced semiconductor technologies—most notably Nvidia’s high-end GPUs designed for AI training like the H100 and A100.

China’s response has been a sustained effort to build indigenous alternatives. Companies such as Innosilicon, Moore Threads, Biren Technology, MetaX, and Mu Xi are racing to meet China’s burgeoning demand for AI and graphics compute power driven by cloud computing, AI development, and digital transformation initiatives.

Moore Threads and Mu Xi, which employ former Nvidia and AMD veterans, have actively pursued IPOs to secure funding for product development and scalability. Industry forecasts by IDC suggest that by the end of 2025, more than half of China’s AI compute power will rely on domestic GPUs, fueled largely by restricted access to foreign chips and geopolitical factors.

Comparing Fenghua No.3 to Global Alternatives

While Fenghua No.3 represents China’s largest milestone in domestic GPU technology, it currently lags behind Nvidia’s flagship GPUs in raw performance. Independent reports estimate Fenghua No.3’s performance roughly on par with Nvidia’s RTX 2080 launched in 2018—significant for a first-generation domestic high-performance GPU but with room to improve.

However, supporting large AI models with its massive HBM and RISC-V plus CUDA compatibility enables Fenghua to effectively serve AI inference and moderate-scale training needs domestically and support Chinese AI service providers.

Chinese-made GPUs prioritize ecosystem compatibility and vertical integration with Chinese cloud providers like Alibaba, Baidu, and Tencent. This integrated supply chain approach can accelerate localized AI adoption and reduce foreign dependencies.

Sponsored Banner

Broadening AI Hardware Ecosystem in China

Fenghua is part of a thriving ecosystem of Chinese AI chipmakers innovating in GPUs as well as related hardware such as AI accelerators and tensor processing units (TPUs). Huawei’s Ascend series competes in the AI inference and training arena, while startups like Biren Technologies emphasize energy-efficient AI computing chips.

China’s strategic government support via policies, funding, and regulatory favor has fueled this vibrant chip development economy, prioritizing semiconductor independence.

Challenges and Future Prospects

Despite promising advances, Chinese GPU makers face challenges including:

• High Research & Development (R&D) expenses with multi-billion dollar investments required.
• Longer product cycles relative to market leaders Nvidia and AMD.
• Restricted access to leading-edge semiconductor manufacturing technologies and equipment.
• Software ecosystem maturity and driver optimization still under development compared to established players.

Nevertheless, the governmental push for semiconductor sovereignty and recent ease in some export controls fuel optimism for accelerated progress. Partnerships with domestic foundries like SMIC and growing collaboration across industry players also improve prospects.

With initiatives underway to enhance memory bandwidth, transistor density, and AI-specific architectural features, the Fenghua family is expected to iterate rapidly. Upcoming models promise to close performance gaps while expanding AI-specific functionalities like tensor cores and improved multi-GPU scalability.

Impact on Global AI Hardware Landscape

The rise of Fenghua and China’s domestic GPU scale-up affects the global AI semiconductor ecosystem. It creates viable alternatives to Western-dominated market leaders, reshaping supply chains and encouraging more regional AI technology ecosystems worldwide.

Fenghua contributes to a more competitive landscape, encouraging innovations around price/performance, local optimization for Chinese language and applications, and tailored hardware-software co-design adapted to China’s unique AI development environment.

Conclusion

Fenghua No.3 is emblematic of China’s ambitions to build a sovereign AI hardware industry capable of catering to domestic and international demands. Though it does not surpass global flagship GPUs, it marks critical progress towards technological self-sufficiency. Through ongoing investments, talent acquisition, and ecosystem building, China’s domestic GPUs—including Fenghua—are poised to become major players in the AI hardware domain in the near future.

As geopolitical dynamics continue to shape technology access, these indigenous efforts will not only serve local needs but also increasingly influence the AI hardware market globally, fostering innovation and diversification in GPU technology with broad implications across AI, cloud computing, gaming, and professional graphics sectors.