Advanced Micro Devices CEO Lisa Su said artificial intelligence has reached an inflection point as inference workloads now outpace training [1].
This shift in AI activity signals a transition from developing models to deploying them in real-world applications. For hardware providers like AMD, this transition dictates where engineering resources and capital are allocated to maintain a competitive edge in the semiconductor market.
Su said the company is committed to AI-focused products during an interview with Yahoo Finance [1]. She said the demand for inference, the process of using a trained AI model to make predictions, is now the primary driver of the industry's evolution [1].
Financial data reflects this strategic pivot. AMD's revenue increased by 50% year-over-year [2]. The growth is largely attributed to the company's data-center business, which Su said is driving growth and is up more than 100% year-over-year [3]. Specifically, data-center sales saw a 107% increase, effectively doubling the segment's output [2].
Despite these gains, the company's stock performance has remained volatile. The stock price fell following the release of earnings reports, even as revenue climbed [2]. This disconnect between fundamental growth and market price suggests a gap between internal performance and investor expectations.
Market sentiment remains divided on how to value the company during this transition. While some reports emphasize that the stock is down despite strong growth [2], other analysts maintain a "buy" rating for AMD [4]. These analysts said the company's current AI positioning is a primary reason for long-term optimism [4].
Su said the company will continue to double down on AI chips to capitalize on the shift toward inference [1]. The strategy aims to position AMD as a primary winner in the expanding AI infrastructure market [4].
“AI has hit an inflection point as inference surpasses training.”
The transition from training to inference marks the 'deployment phase' of the AI boom. While training requires massive compute power to create a model, inference is the ongoing process of running that model for users. If inference becomes the dominant workload, the market for AI chips will shift from a few massive training clusters to a broader, more distributed demand across various data centers and devices, potentially diversifying the revenue streams for chipmakers like AMD.
