AMD Sees AI Inference Growing 80% Annually
In This Article:
AMD (NASDAQ:AMD) CEO Lisa Su says AI inferencing demand is set to grow more than 80% annually over the next few years, overtaking training as the primary driver of data-center compute.
Speaking at the Advancing AI 2025 conference in San Jose, Su highlighted an explosion of models tailored to coding, healthcare and finance, and forecast that hundreds of thousands, and eventually millions of purpose-built models will drive massive GPU usage.
She reminded the audience that last year AMD projected the data-center AI accelerator total addressable market would expand over 60% annually to $500 billion by 2028and now sees that number trending even higher as inference takes off.
Su used the stage to unveil the MI350 GPU series, including the flagship MI355, which delivers a 35 performance leap over its predecessor and 40% more tokens per dollar versus Nvidia's (NASDAQ:NVDA) B200 when running LLMs like DeepSeek-R1 and Llama 3.3. Customers such as Meta (NASDAQ:META), xAI and Microsoft are already preparing upgrades from MI300 to MI350, highlighting growing commercial traction.
Su also touted AMD's record 40% share of the AI-accelerator market, powering the world's two fastest supercomputers and serving seven of the top ten model providers with its Instinct chips. Eric Boyd, Microsoft's AI-platform VP, reinforced the momentum, praising AMD's high-capacity memory per chip as a significant advantage for LLM workloads.
Why It Matters: With inferencing now the largest driver of AI compute, AMD's optimistic growth outlook and new hardware could translate into accelerated revenue and margin expansion as enterprises race to deploy ever more specialized AI models.
This article first appeared on GuruFocus.