During a company event on Thursday in San Jose, California, Su remarked that the latest releases in AMD’s MI350 chip series outperform Nvidia’s equivalents and show significant improvements over prior models. The MI355 chips, which began shipping earlier this month, are 35 times faster than their predecessors, according to her.
While AMD continues to trail behind Nvidia in AI accelerators—the chips essential for developing and running artificial intelligence applications—it aims to narrow the gap with these new offerings. The stakes are higher than ever: Su had initially predicted $500 billion in market revenue by 2028, but now estimates it will surpass that figure.
“People previously thought that $500 billion was an enormous number,” she stated in a briefing after her presentation. “Now it appears well within reach.”
In February, AMD provided a forecast for its data center business that indicated slower growth than what some analysts had expected. AMD believes that the recent enhancements to its MI product line will restore momentum and demonstrate its capability to compete directly with a much larger competitor.
AMD indicated that the MI355 outperforms Nvidia’s B200 and GB200 models in running AI software and matches or surpasses them in code creation. Purchasers will find the cost significantly lower than what they would pay for Nvidia’s products, according to AMD.
A representative from Nvidia did not immediately provide a comment.
Investors reacted lukewarmly to AMD’s latest presentation, with shares dropping 2.2% to $118.50 at Thursday’s close in New York, effectively erasing the small gains the stock had accrued this year.
Nvidia and AMD are leading providers of advanced graphics chips, which serve as foundational components for AI development. Demand has consistently outpaced supply as some of the largest corporations have invested billions into new infrastructure, driving up chip prices, which can be tens of thousands of dollars each.
For AMD, the accelerator sector has allowed it to step out of Intel Corp.’s shadow, its long-time adversary in personal computer processors. However, Nvidia has surpassed them both. While AMD is earning billions from its AI accelerators, Nvidia is pulling in over $100 billion annually.
Su also discussed upcoming products, including the MI400 series set to launch next year. She claimed this product will achieve clear superiority over the Nvidia chips available at that time. AMD is incorporating additional memory and components that facilitate quicker information access, a crucial advantage for running AI software, Su noted.
Sam Altman, CEO of OpenAI, made a guest appearance with Su, who mentioned that the MI400 was developed with input from the engineers at the AI company.
Altman expressed surprise when he first learned about the MI400’s proposed specifications, remarking, “No way.” He noted that the demands on computing infrastructure have intensified this year as the application of AI services has transitioned from a novelty to a valuable business asset.
A critical question remains whether AMD can broaden the market for its AI chips. Similar to Nvidia, the company is restricted from exporting its most powerful components to China due to U.S. trade restrictions. AMD is vigorously advocating for the current administration to permit greater freedom in offering AI components to other nations.
A recent agreement announced in Saudi Arabia was finalized after consultations with Washington to ensure AMD’s ability to supply the chips, Su stated on Thursday. Greater flexibility is essential for maintaining the U.S. technology’s leadership in the global development of AI systems, she emphasized.
