AMD CEO Predicts AI Needs 10 Yottaflops

AMD CEO Lisa Su has projected that the future of artificial intelligence will require massive computing infrastructure. During her keynote address at CES 2026 in Las Vegas, Su introduced the concept of yottaflops to quantify this demand. She stated that the world will need more than 10 yottaflops of compute over the next five years to keep up with AI's growth.

This projection represents a significant leap in technological requirements. Su noted that this scale of computing power has never been built before in the history of the industry. The announcement highlights the rapid acceleration of AI development and the hardware needed to support it.

During the presentation, Su engaged the audience regarding the technical term. "How many of you know what a yottaflop is?" she asked, requesting attendees to raise their hands. When no one responded, she provided a definition herself.

A yottaflop is defined as a one followed by 24 zeros. In computing terms, a flop represents a single basic math calculation. To provide context for the scale:

• A computer performing 1 billion calculations per second is a gigaflop
• A yottaflop equals one septillion calculations per second
• 10 yottaflops is 10,000 times more compute than was available in 2022

Scientists theorize that 10 yottaflops would provide enough power to run complex, atom-level simulations for entire planets.

The growth in AI compute requirements has been exponential. According to Su, global AI compute stood at approximately one zettaflop in 2022. A zettaflop is a one followed by 21 zeros. By 2025, that figure had surged to more than 100 zettaflops.

Su emphasized the unprecedented nature of this expansion. "There's just never, ever been anything like this in the history of computing," she said at the Las Vegas conference.

To illustrate the magnitude of her prediction, Su compared the 10 yottaflop goal to current hardware capabilities. Her prediction is approximately 5.6 million times faster than El Capitan, the most powerful supercomputer currently in existence. El Capitan is operated by the US Department of Energy.

Scaling AI compute to the predicted levels presents significant challenges. Su noted that powering current AI compute is already putting a strain on the US power grid. Consequently, the build-out of energy infrastructure represents a major bottleneck for scaling up AI capabilities.

Despite these challenges, AMD is moving forward with new hardware releases. During the keynote, Su unveiled the company's next generation of AI chips. This lineup includes the MI455 GPU. These announcements signal AMD's continued push to supply data-center hardware for major customers, including OpenAI.