AMD has launched Helios, a rackscale AI infrastructure solution capable of delivering up to 2.9 exaFLOPS of FP4 compute. The company has also announced significant deployments with major AI players like Anthropic and OpenAI.
TL;DR
- AMD introduces Helios, a high-performance AI infrastructure solution with up to 2.9 exaFLOPS of compute, 31TB of HBM capacity, and 50% more scale-out bandwidth than competitors.
- Anthropic and OpenAI have committed to deploying up to 2 GW and 6 GW of AMD's Helios infrastructure, respectively, emphasizing co-design and early collaboration.
- The launch of Helios and strategic partnerships highlight AMD's push to capture a significant share of the rapidly growing AI accelerator market, projected to reach $1.4 trillion by 2030.
What happened
AMD has launched Helios, a rackscale AI infrastructure solution designed to meet the growing demands of AI workloads. The solution offers up to 2.9 exaFLOPS of FP4 compute, 31TB of HBM capacity, 15% more compute, 50% more memory capacity and bandwidth, and 50% more scale-out bandwidth compared to competitors. Shipments of Helios are set to begin in late Q3, with a full ramp-up in Q4.
In addition to Helios, AMD announced the Instinct MI455/MI455X accelerator, built on advanced 2nm/3nm process technology. The MI455/MI455X delivers industry-leading GPU performance, up to 34x higher inference throughput, and up to 18x lower token cost compared to previous generations. The accelerator is integrated with CPUs and networking for rack-scale AI.
AMD also introduced the 6th-gen EPYC CPU family, codenamed Venice, featuring Zen 6 cores, up to 256 cores per chip, up to 1.8x performance, 512 threads, and 1.6TB/s memory bandwidth. The Venice CPUs are designed to support agentic AI, cloud, and enterprise workloads.
Why it matters
The launch of Helios and other AI infrastructure solutions positions AMD to capture a significant share of the rapidly growing AI accelerator market, which is projected to reach $1.4 trillion by 2030. The increasing demand for AI compute, driven by the rise of agentic AI and the shift from training to inference workloads, presents a substantial opportunity for AMD and other AI infrastructure providers.
Strategic partnerships with leading AI companies like Anthropic, OpenAI, and Meta underscore the importance of collaboration and co-design in scaling AI infrastructure and innovation. These partnerships enable AMD to tailor its solutions to the specific needs of its customers and drive adoption of its AI infrastructure offerings.
For developers and startups, the availability of high-performance AI infrastructure solutions like Helios and Instinct MI455/MI455X accelerators can accelerate the development and deployment of AI models. The integration of these solutions with CPUs and networking can simplify the deployment of AI workloads and reduce the time-to-market for AI applications.
Key facts
- Helios offers up to 2.9 exaFLOPS of FP4 compute, 31TB of HBM capacity, 15% more compute, 50% more memory capacity and bandwidth, and 50% more scale-out bandwidth than competitors.
- The Instinct MI455/MI455X accelerator delivers up to 34x higher inference throughput and up to 18x lower token cost compared to previous generations.
- Venice, the 6th-gen EPYC CPU family, features Zen 6 cores, up to 256 cores per chip, up to 1.8x performance, 512 threads, and 1.6TB/s memory bandwidth.
- Anthropic will deploy up to 2 GW of Helios infrastructure, while OpenAI has committed to deploying up to 6 GW of AMD infrastructure.
- Meta is a lead partner for Venice CPUs and MI450 accelerators, emphasizing co-design and early collaboration.
- The AI accelerator market is projected to reach $1.4 trillion by 2030, with data center AI accelerator and CPU TAMs growing at >45% and >50% CAGR, respectively.
- Inference workloads have surpassed training as the primary AI workload, with agentic AI expected to account for 60% of global AI compute in 2026.
- The total AI compute TAM is projected to reach $2 trillion by 2030, driven by a 158x increase in tokens consumed monthly over two years and a 5x annual increase in training compute (FLOPS) since 2020.
Context
The rapid adoption of AI across industries, including healthcare, manufacturing, finance, education, science, software, agriculture, and media, is driving a surge in compute demand. The shift from training to inference workloads, coupled with the rise of agentic AI, is transforming the AI infrastructure landscape and creating new opportunities for AI accelerator providers like AMD.
Collaboration and open ecosystems are essential for scaling AI infrastructure and innovation. Strategic partnerships between AI infrastructure providers and leading AI companies enable the development of tailored solutions that meet the specific needs of customers and drive adoption of AI technologies.
The AI infrastructure market is highly competitive, with players like NVIDIA, Google, and Microsoft investing heavily in AI accelerator technologies. AMD's launch of Helios and strategic partnerships with leading AI companies position the company to capture a significant share of the rapidly growing AI accelerator market.
