Talking to the gathered throngs at a Citi conference, Hu said the firm was still working out “if Chinese customers can be allowed to buy from US" and AMD was “not starting new wafers for MI308,” the 'China only' accelerator, and that the plan was to clear existing stock before committing to anything else.
Future spend in the region rests on whether AMD “can get the license for our next generation,” Hu said.
AMD, vice president of investor relations, Matthew Ramsay said: “I think inside of China, there's a larger demand for AI processing silicon than there is ability to manufacture that silicon in China.”
His read is that sanctions crimp local leading edge supply while demand shoots up, which leaves AMD eager but constrained. “Getting visibility in the short term as to what that looks like is -- given all the moving parts, has been the challenge.”
Hu dodged the bubble bait and stuck to adoption data, pointing to hyperscaler capex in the second quarter and “tremendous evidence for AI adoption.”
Those customers “have improved their return on their investment across not only their platforms” and AMD itself is seeing productivity gains, she said.
“We hear a lot of other companies adopting AI.”
Her stance is that “we are still at a very early stage for AI adoption” and it is too soon to size the “magnitude” of the impact, which is hardly a bearish posture.
She nudged attention towards CPUs, saying AI is lifting demand for general compute as well. In her words, AI is “probably once a lifetime opportunity we're seeing.”
Hu said every generation adds features and bill of materials, which pushes up list prices, but AMD aims to keep total cost of ownership attractive while protecting gross margins.
Conference host, Christopher Danley, pressed on the headline grabbing $500 billion total addressable market. Ramsay pointed to inference growth, data set size and industry spread as the levers.
“I've said this for a long time that I think as T goes to infinity, right, more CapEx and OpEx dollars in basically every industry goes into high-performance computing, AI is a significant inflection of that,” he said.
He dubbed AI “the biggest inflection in computing since the invention of the Internet” and floated that AMD’s TAM could run past the $500 billion figure chief executive Lisa Su flashed earlier this year.
AMD’s own slides have talked up inference growing at more than 80 per cent CAGR from $45 billion in 2023 to beyond $500 billion in 2028, which is an ambitious hill to climb without slipping.
