AMD has launched Helios, its first rack-scale AI system and its most direct attempt yet to compete with Nvidia's Grace Blackwell and Vera Rubin racks — and it arrived with a marquee customer. Microsoft said it will deploy Helios across Azure, positioning AMD as a genuine second source for the compute powering large-model inference.
What a Helios rack holds
A single Helios rack combines 72 Instinct MI455X GPUs with 18 EPYC "Venice" CPUs (over 4,600 cores), 31TB of HBM4, 1.4 PB/s of aggregate memory bandwidth and 2.9 FP4 exaflops of inference compute, tied together with AMD's Pensando networking and its open ROCm software stack. Shipments begin in the second half of 2026.
Microsoft brings the demand
Microsoft will run Helios for frontier-model inference and is adding two EPYC "Venice"-based Azure VM series — HDv2 for data-heavy and agentic workloads and HXv2 for semiconductor design — plus an ND MI455X v7 instance. Satya Nadella said Microsoft is "expanding the Azure infrastructure portfolio with AMD Helios to give customers the performance, scale and choice they need." AMD joins Meta, OpenAI, Oracle and Tata Consultancy Services on the buyer list.
The economics AMD is selling
Analysts at Futurum peg Helios at roughly $5 million to $5.5 million per rack, above the estimated $3.5 million to $4 million for Nvidia's Vera Rubin, but AMD is pitching total cost of ownership rather than sticker price. Data-center chief Forrest Norrod said the company is "very focused on providing the best total cost of ownership, the lowest cost per token." AMD told CNBC it expects to book tens of billions in data-center AI revenue starting in 2027.
Still an underdog
The gap remains vast: Nvidia controls more than 95% of the data-center GPU market against AMD's roughly 4.5%. But a named Azure commitment, arriving two days before AMD's Advancing AI conference on July 22, gives the challenger something it has lacked — a hyperscaler willing to stand up its rack-scale system at scale.