AMD · Earnings call · 2026Q2T · Full report

AI Chip Competition

ADVANCED MICRO DEVICES INC · 2026-08-04 · Importance 34 · Surprise 42

AMD positioned Helios against competing AI infrastructure with up to 15% more throughput at the same rack power and up to 30% more tokens per dollar across a broad range of inference workloads. The platform's memory bandwidth and capacity are intended to improve performance and total cost of ownership for larger models, although AMD can modify memory configurations for workloads that do not justify larger footprints. AMD expects MI500 to deliver the largest generational leap in Instinct history and increase inference performance more than 2,000 times in four years. Management said the open ecosystem and workload-specific optimization approach, including the Cerebras partnership, will support differentiated inference solutions.