AMD downplays risk that DeepSeek's AI efficiency innovations could reduce demand for Instinct GPUs.

AMD's chief executive Lisa Su has predicted the chip designer's Instinct accelerators will drive tens of billions of dollars in annual revenue in coming years, dismissing concerns that DeepSeek-inspired speculation about more efficient ML architectures could undermine demand. On Tuesday's Q4 earnings call, Su acknowledged that the surprisingly capable made-in-China LLM family demonstrated training and inference with fewer GPUs than expected, prompting questions about whether cutting-edge AI truly requires the billions in capex Silicon Valley demands. However, Su stated, "The fact that there are new ways to bring about training and inference capabilities with less infrastructure is actually good, because it allows us to continue to deploy AI compute in the broader application space and [with] more adoption."

Su expects Instinct revenues to increase significantly compared to the $5 billion-plus they generated during the 2024 fiscal year, though she resisted committing to a specific number. She indicated AMD's strongest growth would come in the second half of fiscal 2025, which ends December 28, driven by shipments of next-generation MI355X accelerators. Originally expected to ship toward the end of the second half, the company now anticipates ramping production toward the middle of the year based on early silicon performance and customer demand. The MI355X will employ the CDNA 4 architecture with floating-point performance roughly on par with Nvidia's B200 accelerators while delivering 50 percent more memory at 288GB.

Su also previewed AMD's next-generation MI400-series accelerators, which will be offered in a rack-scale architecture. She stated that the CDNA-Next architecture would enable "powerful rack-scale solutions that tightly integrate networking, CPU and GPU capabilities at the silicon level to support Instinct solutions at data center scale." The systems are expected to use Ultra Ethernet and UA Link rather than the proprietary standards employed by competitors like Nvidia.

On the threat of custom AI ASICs, Su expressed confidence that GPUs would remain essential alongside specialized silicon. "I have always been a believer in you need the right compute for the right workload," she said. "Given the diversity of workloads, large models, medium models, small models, inference... broad foundation models or very-specific models, you're going to need all types of compute and that includes CPUs, GPUs, ASICs, and FPGAs." Su expects accelerators to eventually contribute to a total addressable market exceeding $500 billion, with ASICs remaining a smaller portion given ongoing AI algorithm evolution.

AMD's financial results underscore the importance of its accelerator business. For fiscal 2024, profits surged 92 percent to $1.6 billion on revenues up 14 percent to $25.8 billion. The datacenter division generated $12.6 billion in annual revenues, up 94 percent, with Instinct GPUs accounting for $5 billion. In the fourth quarter alone, datacenter revenues reached $3.9 billion, up 69 percent year-on-year, while the division recorded $482 million in net income and $7.75 billion in revenues overall. AMD's client computing division achieved $2.3 billion in Q4 revenues, up 58 percent, though gaming revenues declined 59 percent year-over-year to $563 million.