18 GPU Developers in the World Today: The Rise of China's AI and Tech Sovereignty-Driven GPU Makers

In recent years, there has been a surge in the number of GPU makers in China as the country aims to boost its artificial intelligence capabilities and gain semiconductor sovereignty. This growth is not limited to China, as the demand for AI, high-performance computing, and graphics processing has risen significantly worldwide. When it comes to discrete graphics for PCs, AMD and Nvidia currently hold the lead, with Intel attempting to catch up.

According to a report from Jon Peddie Research, the number of GPU startups in China is particularly impressive as the country strives to become a dominant player in the AI and tech industries. These new entrants are driven by a desire for tech sovereignty, as well as the increasing demand for advanced processing capabilities in a variety of fields.

The rise of GPU makers in China is indicative of the country's broader efforts to become a leader in the tech sector. With strong government support and a large pool of talented engineers and researchers, China is well-positioned to make significant strides in the field of AI and other cutting-edge technologies.

The growth of GPU makers worldwide is a testament to the increasing importance of advanced processing capabilities in a variety of industries. From self-driving cars to complex financial analyses, the demand for powerful graphics processing is only expected to continue growing in the coming years. As such, the competition among GPU makers is likely to remain fierce as they strive to meet the needs of a constantly evolving market.

In the 1980s and 1990s, there were tens of companies that developed graphics cards and discrete graphics processors. However, intense competition for high performance in 3D games led to the demise of most of these companies, leaving only AMD and Nvidia as viable standalone GPU options for gaming and compute by 2010. While other companies continued to focus on integrated GPUs or GPU IP, the mid-2010s saw a rapid increase in the number of China-based PC GPU developers.

According to Jon Peddie Research, there are currently 18 companies developing and producing GPUs. Two of these companies primarily develop SoC-bound GPUs for smartphones and notebooks, while six are GPU IP providers. The remaining 11 companies are focused on GPUs for PCs and datacenters, including well-known names like AMD, Intel, and Nvidia. In fact, if we were to include other China-based companies like Biren Technology and Tianshu Zhixin, the number of GPU designers would be even higher. However, these companies are currently solely focused on AI and HPC, so they are not considered GPU developers by Jon Peddie Research.

China is competing with other well-developed countries, including the United States, in various sectors, including technology. As the world's second-largest economy, China has made efforts to attract engineers from around the globe and establish various chip design startups within its borders. In fact, hundreds of new IC design houses are created in China every year, developing everything from tiny sensors to complex communication chips in an effort to achieve self-sufficiency from Western suppliers.

However, to truly advance in the fields of AI and HPC, China needs CPUs, GPUs, and specialized accelerators. While it may be difficult for Chinese companies to catch up to long-time market leaders in the CPU and GPU sectors, it may be easier and potentially more beneficial to develop and produce a competitive GPU. According to Jon Peddie, the head of Jon Peddie Research, the main motivators for Chinese GPU companies have been "AI training, avoidance of Nvidia's high prices, and (maybe mostly) China's desire for self-sufficiency."

GPUs are inherently parallel, which means they have multiple compute units that can be used for redundancy, making it easier to get a GPU up and running (assuming per transistor costs are low and overall yields are decent). Additionally, since GPUs are fundamentally parallel, it is easier to scale them out. Given that the Chinese company SMIC does not have production nodes as advanced as those of TSMC, this method of performance scaling is sufficient. Even if Chinese GPU developers lose access to TSMC's advanced nodes (N7 and below), they may still be able to produce simpler GPU designs at SMIC and address the AI/HPC and/or gaming/entertainment markets.

From China's perspective as a country, AI and HPC-capable GPUs maybe even more important than CPUs, as they can enable new applications such as autonomous vehicles and smart cities, as well as advanced conventional weapons. The U.S. government restricts exports of supercomputer-bound CPUs and GPUs to China in an effort to slow or hinder the development of advanced weapons of mass destruction. However, a sophisticated AI-capable GPU can enable the development of autonomous killer drones, which could represent a formidable force.

It is important to note that while there are many GPU developers, only a small number are able to build competitive discrete GPUs for PCs. This is likely because it is relatively easy to develop a GPU architecture, but much more difficult to properly implement it and design proper drivers.

According to Jon Peddie, the head of Jon Peddie Research, CPU and GPU microarchitectures sit at the intersection of science and art. They are sets of sophisticated algorithms that can be developed by a small team of engineers but may take years to perfect. Peddie explains that microarchitectures are often developed on napkins and whiteboards, and the process can be complex as it involves anticipating where manufacturing processes and standards will be in the future, making cost-performance tradeoffs, and deciding which features to include and which to omit.

The hardware design process is also expensive, as it requires specialized equipment and a large team of engineers to bring a product to fruition. While the microarchitecture team may be small, the hardware design team is typically much larger and more complex. Overall, the process of developing a CPU or GPU microarchitecture is time-consuming and requires careful consideration of a wide range of factors.

Developing a CPU or GPU microarchitecture can be a lengthy and expensive process, which is why many companies opt to license an off-the-shelf microarchitecture or GPU IP from companies like Arm or Imagination Technologies. This allows them to save time and money and get their products to market faster. For example, Innosilicon, a contract developer of chips and physical IP, licenses GPU microarchitecture IP from Imagination for its Fantasy GPUs, while Zhaoxin uses a highly reiterated GPU microarchitecture it acquired from Via Technologies, which inherited it from S3 Graphics.

The hardware implementation and software development process is also costly, particularly with new production nodes. The design costs for a complex device using 5nm-class technology are estimated to exceed $540 million, and these costs are expected to triple at 3nm. These expenses are exacerbated by the time and effort required for layout and floor plans, simulation, verification, and driver development.

There are only a few companies in the world capable of developing a chip with the complexity of modern gaming or compute GPUs from AMD and Nvidia, which contain 46 billion to 80 billion transistors. However, China-based Biren may be able to produce a similar product with its BR104 and BR100 devices, which are speculated to contain 38.5 billion transistors. Overall, the process of developing and producing a GPU is complex and requires a significant investment of time and resources.

Despite the significant costs and challenges involved in developing and producing a GPU, eight out of the 11 PC/datacenter GPU designers are based in China. This demonstrates the country's determination to become a competitive player in the tech industry, including in the field of graphics processing. While it remains to be seen whether China will be able to produce competitive discrete gaming GPUs, the country's efforts should not be underestimated. The process of developing a GPU is difficult and time-consuming, and the hardware implementation costs can be prohibitively high for these high-complexity products. It is possible that we may not see a competitive discrete gaming GPU from anyone other than large American companies in the near future due to these challenges. However, China's efforts to enter the market should not be underestimated.