Advanced Micro Devices (AMD) has launched its Helios AI rack-scale system, a move directly challenging Nvidia's leading position in the artificial intelligence hardware market. The company announced the system on July 23, 2026, at its Advancing AI 2026 conference. Helios is AMD's first fully integrated rack-scale AI infrastructure solution and is slated to begin shipping to customers in the second half of 2026.
Nvidia currently holds an estimated 75% to 81% of the AI accelerator market revenue for 2026, according to various analyst reports. AMD's new system aims to offer a competitive alternative by integrating 72 AMD Instinct MI455X GPUs with AMD EPYC "Venice" server CPUs and AMD Pensando networking technology. This co-designed architecture is optimized for compute, data movement, and system efficiency, providing what AMD claims is industry-leading memory capacity and bandwidth.
AMD states that Helios offers up to 15% more AI compute, 50% more HBM capacity, and 50% more scale-out bandwidth compared to Nvidia's Vera Rubin NVL72 rack. The Helios system provides 2.9 exaflops of FP4 compute and 1.4 exaflops of FP8 compute, with 31 terabytes of HBM4 memory. The system also features 260 terabytes per second of scale-up bandwidth and 43 terabytes per second of scale-out bandwidth.
The launch of Helios is supported by significant growth in AMD's Data Center segment, which generated $5.8 billion in revenue in the first quarter of 2026, a 57% increase year-over-year. The company has also secured major customer commitments, including plans to deploy AMD's AI infrastructure at scale with companies like OpenAI, Anthropic, Meta, and Microsoft.
AMD is also partnering with Cerebras Systems to develop a platform that combines AMD's EPYC processors in Helios rack-scale infrastructure with Cerebras' Wafer-Scale Engine (WSE) solutions. This collaboration aims to deliver low latency and high throughput for AI inference workloads. The joint solution is expected to be available through Cerebras Cloud in the second half of 2026.
AMD's strategy with Helios extends beyond hardware. The company is emphasizing an open platform and software ecosystem, including its ROCm software. This approach contrasts with Nvidia's proprietary CUDA software, which has been a significant factor in its market dominance and developer lock-in. By offering an open architecture, AMD seeks to provide customers with greater flexibility and choice in their AI infrastructure deployments.
The Helios system is built on the Open Compute Project's Open Rack Wide standard, featuring wider rack frames to accommodate the increased density and power requirements of AI hardware. Each Helios rack integrates 18 open rack wide-aligned 4-GPU compute trays, totaling 72 GPUs. The system's power density can reach up to 246 kW per rack.
While AMD's new offerings present a significant challenge to Nvidia, the long-term success of Helios will depend on independent performance validation and the continued maturity of AMD's software stack. Nvidia's established CUDA ecosystem and its ongoing hardware advancements, such as the Rubin platform, continue to present formidable competition.