The generative A.I. boom is fueling an arms race among semiconductor manufacturers, who produce the engines that power large language model applications like ChatGPT. Specifically, the race right now is between two of the industry leaders: Nvidia (NVDA) and AMD (AMD) (Advanced Micro Devices Inc.).
At a product event yesterday (June 13) AMD announced it will begin shipping the MI300X, its most-advanced graphics processing units (GPUs) designed for artificial intelligence, later this year. The MI300X was first announced in June 2022 and was previously expected to ship next year.
At the event, AMD CEO Lisa Su emphasized that A.I. is the company’s “largest and most strategic long-term growth opportunity.”
“At the center of this are GPUs. GPUs are enabling generative AI,” she said.
A.I. chips, also known as A.I. accelerators, are one of the few bright spots in the semiconductor industry, which is facing slumping demand for personal computers. Su estimates today’s A.I. accelerator market is worth somewhere around $30 billion and set to grow 50 percent every year to reach $150 billion by 2027.
Currently, the market is dominated by Nvidia, which is believed to own 95 percent of the GPU market and more than 80 percent of the A.I. chip sector.
How AMD’s A.I. chip compares with Nvidia’s
AMD’s MI300X is a direct competitor of Nvidia’s H100. The MI300X can support 192GB of memory compared to H100’s 120GB.
AMD will also offer a package called Infinity Architecture that combines eight M1300X accelerators into one system. Nvidia has developed a similar system called NVLink that can combine eight or more GPUs in a single box for A.I. applications.
“Model sizes are getting much larger, and you actually need multiple GPUs to run the latest large language models,” Su said yesterday. ChatGPT, for example, has 175 billion parameters. At yesterday’s event, AMD demonstrated that one MI300X is capable of running a 40-billion-parameter model called Falcon.
AMD hasn’t disclosed how much its MI300X will cost. Nvidia’s H100 processors cost $40,000 per unit.
The two semiconductor giants are led by Taiwanese American CEOs
AMD’s Su, 53, and Nvidia CEO Jensen Huang, 60, have similar ethnic and professional backgrounds. Both executives were born in Tainan, a southern city in Taiwan, and immigrated to the U.S. at a young age with their families.
Su is an accomplished electrical engineer who worked for Texas Instruments, IBM and Freescale Semiconductor before joining AMD in 2012. She has led AMD as its president and CEO since 2014 and is credited with diversifying the company’s business away from personal computers into other segments like video gaming and embedded devices.
Su is the highest-paid woman CEO in the U.S. Her 2022 compensation was valued at nearly $30 million.
Nvidia’s Huang, who is also the company’s founder, worked at AMD as a chip designer before starting his own company in 1993.
Huang made the early bet for Nvidia to focus on developing A.I. processors back in 2012 and his vision paid off handsomely as the A.I. boom took off late last year. Nvidia’s share price has soared more than 160 percent this year, driven by booming sales of A.I. chips.
Huang is estimated to be worth $36.8 billion, according to Bloomberg’s Billionaires Index, making him one of the 40 richest people on Earth.