AMD CEO Lisa Su said AI infrastructure demand continues to accelerate and the market opportunity remains in its early stages [1].

This growth signals a shift in how companies deploy artificial intelligence, moving from the initial training of models to the active deployment of AI agents that reshape computing economics.

During an interview with Yahoo Finance on March 19, 2024, Su said inference is the biggest growth driver for the industry [1]. She said the company is positioning itself to capture a larger share of the market by focusing its roadmap on high-performance compute for both training and inference [3].

The company's outlook coincides with a significant financial surge. Following the release of its first-quarter earnings, AMD doubled its long-term revenue forecast [4]. Market reaction was immediate, with the company's stock price jumping between 17% [2] and 19% [4] after the announcement.

Su said market speculation regarding a potential slowdown in the AI boom is incorrect, stating that the demand for infrastructure is not waning but rather speeding up [1]. This acceleration is driven by the transition toward inference, where AI models apply learned data to real-world tasks in real time.

AMD is competing for dominance in the data center market against established rivals. The company's strategy relies on delivering hardware capable of handling the massive computational loads required for modern AI agents [3].

"AI infrastructure demand continues to accelerate, and we see the opportunity still in its early stages," Su said [1].

“Inference has become the industry's biggest growth driver.”

The shift from AI training to inference marks a critical transition in the technology lifecycle. While training requires massive one-time bursts of power to create a model, inference is the ongoing process of using that model to generate answers. If inference becomes the primary driver of growth, it suggests that AI is moving out of the laboratory and into widespread commercial application, increasing the long-term demand for specialized hardware over short-term experimental bursts.