AMD Spurs AI Interest With Forthcoming MI300 Chip

data centre AI server

AMD tantalises investors with details of forthcoming MI300 AI chip, expected to be direct competitor to Nvidia’s flagship H100

AMD shares rose in premarket trading on Tuesday ahead of an event at which it is expected to reveal more details around its forthcoming AI-oriented chip called the MI300.

The company has seen its shares surge more than 100 percent since the beginning of the year, amidst an investor frenzy around artificial intelligence spurred by the release of ChatGPT last November by Microsoft-backed OpenAI.

AMD is considered the closest competitor to Nvidia in data centre graphics processing units (GPUs), specialised accelerator chips used by researchers to carry out processor-intensive AI training.

Nvidia dominates the AI computing market with 80 to 95 percent market share, according to analysts.

tara winstead ai artificial intelligence robot amd
Image credit: Tara Winstead/Pexels

AI gold rush

The company last month briefly saw its market capitalisation surge to over $1 trillion (£800bn) after revealing a surge in sales of AI GPUs for data centres.

But AMD is considered a viable competitor, having already carved out 18 percent of the lucrative market for server CPUs from Intel – although competing with Nvidia, with its decades of experience working with AI researchers, may be an even more daunting proposition.

Nvidia’s other competitors include startups such as Cerebras Systems and SambaNova Systems, as well as the internal chip efforts of cloud companies such as Google and Amazon who rent out processing power to AI researchers.

AMD chief executive Lisa Su told investors in a call last month that she believes the MI300 – intended as a direct competitor to Nvidia’s flagship H100 – will begin generating sales in the fourth quarter and “will be more meaningful in 2024”.

‘Closest competitor’

AMD has already had success with its AI accelerator hardware in the high-performance computing market for specialised supercomputers and has been gaining “apparent traction” in the data centre “hyperscale” market, said Wedbush analyst Matt Bryson said in a research note on  Monday.

It “now appears more certain that AMD has established itself as the closest competitor to [Nvidia] in the accelerator/AI hardware market with apparent traction now extending beyond supercomputing into hyperscale,” he wrote.

AMD is the “candidate emerging as the most likely [second] source” to Nvidia in AI chips, he added.