Can AMD Capture Nvidia’s AI Chip Market?

Insiders say Lisa Su’s visit to Taiwan aimed to promote the MI300 series and other AI solutions in the supply chain.

Chairperson and CEO of AMD Lisa Su just concluded her Taiwan visit. How does she challenge Nvidia’s CEO, Jensen Huang, for the leadership position in the AI chip market?


“Nvidia’s accelerators are in short supply, and everyone wants an alternative,” says a high-ranking executive from a major AI server system manufacturer, frustrated that they couldn’t get enough GPUs and facing trouble even producing samples. Who is their alternative? “AMD, of course. They have just launched the MI300, and we are optimistic about its potential.” 

In mid-June, AMD released its next-generation AI chip, the MI300X, featuring eight integrated processors with memory exceeding Nvidia’s and directly challenging Nvidia’s top-tier H100 in terms of speed and performance. Additionally, AMD introduced the CPU-GPU hybrid architecture MI300A, targeting Nvidia’s Grace Hopper architecture. 

The rise of generative AI and the demand for accelerated computing have propelled Jensen Huang’s Nvidia to the peak of performance and the capital market. It is widely recognized that Lisa Su, the woman who single-handedly rejuvenated AMD, is his number-one challenger. AMD has introduced a series of AI computing processors, which are Su’s weapons to break into Nvidia’s dominant territory. 

It has been four years since Su’s last visit to Taiwan. Her visit this July lasted less than a week and mainly involved meetings with over 1,000 AMD employees and supply chain ecosystem members in Taiwan. She also hosted a closed-door event called “AMD Innovation Day” to acquaint the supply chain with AMD’s plans and strategies. 

In comparison to Jensen Huang’s multiple public appearances in May at the computer expo Computex, where prominent Taiwanese companies such as Quanta, Gigabyte, ASUS, and TSMC all endorsed Nvidia’s AI supply chain, Su has remained low-key. 

This difference can be attributed to the fact that while Nvidia’s second-quarter revenue exceeded Wall Street’s estimates by over 50% – leading to a surge in its stock price – AMD’s second-quarter revenue fell short of expectations. Nevertheless, AMD remains one of the most powerful players in the non-Nvidia camp. 

“AI is still in its early stages, and there are five to ten years of prospects to look forward to,” Su said during a media meeting in Taiwan. “It represents significant growth potential for AMD.” The AI market is expected to reach a size of US$150 billion in the next three to five years, offering massive growth opportunities. 

The 53-year-old Su is about to wrap up her 10th year as CEO of AMD, overseeing a major transformation for the company. AMD originally focused on CPUs and, facing years of losses, had to sell its semiconductor manufacturing plant and graphics chip business in order to survive. 

Nine years ago, Su shattered the bamboo-glass ceiling for East Asian women in the semiconductor industry when she became AMD’s CEO. She redirected resources to focus on PC and server processors. Another crucial move was switching its manufacturer from GlobalFoundries to TSMC, making AMD a formidable competitor that Intel could not ignore. 

Last year, she further expanded AMD’s presence by acquiring FPGA manufacturer Xilinx, enabling the company to enter the 5G and automotive fields. Under Su’s leadership, AMD grew from having less than 10,000 employees to employing over 25,000 people, with the majority being research and development engineers. 

During a press conference, Su said that AI is still in its infancy, with five to ten years of prospects to look forward to.

Insiders revealed that Su’s visit to Taiwan aims to promote the MI300 series and other AI solutions in the supply chain, as well as to solidify production capacity through meetings with TSMC, the world’s largest contract chipmaker. It is understood that the MI300 uses TSMC’s 5-nanometer process. 

“MI300 utilizes Chiplet architecture, advanced packaging technology CoWoS, HBM, and other high-end technologies, making it the most powerful product in the world,” Su said in July. “Without TSMC, we couldn’t have achieved this.” She further emphasized that AMD will continue to use TSMC’s most advanced technologies, maintaining a strong and long-term relationship with the Taiwan supply chain. 

However, analysts in the industry believe that Su may not replicate the same enthusiasm Huang generated a month ago, as AMD is currently facing inventory corrections in the personal computer industry. Furthermore, AMD’s AI solutions have not yet been adopted by major customers. Nonetheless, the technology industry’s norm is that no company can monopolize the market forever.  

Market capture opportunities 

An analyst from Northland Capital Markets believes that AMD has a chance to capture 20% of the AI chip market in the long run, considering its continued product improvement and the global need for a second supplier apart from Nvidia. In fact, large cloud service providers like Google, Meta, and Microsoft are all making their own chips, prompting discussions about the “second supplier” on the Taiwan-based supply chain. 

“Now, GPUs are monopolized by Nvidia, and AMD’s entry is reasonable,” says a vice president from a Taiwan-based semiconductor company. He compares Nvidia’s AI acceleration to a car, saying, “some companies or applications don’t need a Ferrari – they only need a Toyota.” 

A director from a Taiwan-based server contract manufacturer contrasts Nvidia’s and AMD’s AI accelerator solutions, noting that “AMD takes the more affordable route but offers comparable hardware performance to Nvidia. The difference lies in the software.” 

He admits that AMD’s products are less user-friendly than Nvidia’s, requiring more intervention by software development engineers. AMD recognizes this and has begun optimizing its software to avoid being tied to Nvidia’s ecosystem. 

When asked how AMD plans to deal with Nvidia’s CUDA software ecosystem, Su responded that AMD has established its own software platform called ROCm. It is an open-source software ecosystem that already supports deep learning languages like PyTorch 2.0 and TensorFlow. She emphasized that AMD would continue to invest in software. In the AI supply chain, rumors have circulated that some companies have already tried out AMD’s AI solutions. 

Lisa Su remains optimistic about catching up with Nvidia in the AI sphere.

A high-ranking executive from a Taiwan-based brand manufacturer, who recently entered the server business, reveals that his company is actively seeking to collaborate with AMD for the MI300. “Currently, AMD’s internal capacity to support downstream customers is limited, so they will choose a few strategic partners first,” he says. 

Analysts speculate that Microsoft, Amazon, and HP are currently in discussions with AMD, while Taiwanese brands such as Acer and ASUS have also shown interest. Su confirmed that many customers have expressed interest in the MI300 over the past three months, including large cloud customers and enterprise-level application customers. However, she did not disclose specific customers and said that AMD would announce its customers when the products are launched later this year. 

With a Ph.D. from MIT and a lifelong career in the semiconductor industry, Su still feels the excitement of being a research engineer when she sees technological innovation. Despite concerns about the overheated speculation surrounding AI and its supply chain in the capital market for the past few months, she believes that AI will remain the primary growth driver for the next generation of computing. 

“We are very ambitious, following technology’s advancement,” she said. “AI will be a major growth driver, and I’m excited about it.” 

This article first appeared in CommonWealth Magazine in July 2023. It has been reprinted, with editing and updating, with permission from the publisher.