AMD unveils Helios rack-scale AI compute portfolio
Fri, 24th Jul 2026 (Today)
AMD has unveiled a new AI compute portfolio led by its Helios rack-scale systems. The launch also included new data centre chips, embedded processors and software tools.
The announcements cover the company's server and accelerator lines, along with products for robotics and edge systems. AMD also highlighted adoption by customers including OpenAI, Anthropic, Meta, Microsoft, AT&T, Cerebras and Cisco.
At the centre of the launch is AMD Helios, a rack-scale AI system now in production. The platform combines 72 Instinct MI455X graphics processors with 18 sixth-generation EPYC "Venice" central processors, linked with Pensando networking and ROCm software.
Helios is designed for large AI training and inference workloads. AMD said the system will be deployed by AI companies at multi-gigawatt scale, naming OpenAI, Anthropic, Meta, Microsoft, Oracle, HUMAIN, Tensorwave, Vultr and Cirrascale among adopters.
The chipmaker also introduced its Instinct MI400 series graphics processors and sixth-generation EPYC server processors. According to AMD, the MI455X is positioned for cloud, enterprise and high-performance computing workloads, while the MI430X is aimed at scientific computing and sovereign AI.
In server processors, the sixth-generation EPYC range is intended to cover cloud, enterprise, telecoms and AI host-node workloads. AMD argued that demand for AI is expanding from model training into inference and agent-based computing, increasing the need for processors, graphics chips, networking and related software.
On software, AMD launched ROCm.ai, which it described as an AI-native development environment for its hardware platforms. The tool supports coding agents such as Claude, Codex and Cursor, and is intended to help developers build, optimise and deploy GPU software across AMD systems.
Several open-source frameworks, including PyTorch, Hugging Face, vLLM and SGLang, are already enabled on the MI455X, AMD said. The company has placed growing emphasis on software support as it tries to strengthen its position against Nvidia in AI infrastructure.
Customer push
Partners used the launch to outline how they are working with AMD on large-scale AI infrastructure. Anthropic said it is deploying up to 2 gigawatts of Instinct MI455X graphics processors in Helios systems as part of a strategic partnership and engineering collaboration.
OpenAI is also working with AMD to tune its AI stack around the company's silicon and software. The two groups are optimising GPT-class workloads using OpenAI's Triton framework with ROCm software on MI455X processors and Helios racks, according to AMD.
Meta is validating sixth-generation EPYC platforms and has started testing workloads on Helios racks, AMD said. Cerebras is collaborating with AMD on a combined offering intended to pair low-latency AI compute with rack-scale infrastructure for inference workloads.
Beyond the largest AI model builders, AT&T is using Instinct graphics processors and ROCm software for its OTel 2.0 telecoms model across cloud, on-premises and air-gapped environments, AMD said. Cisco is working with AMD on AI-enabled collaboration devices and workplace systems, using Ryzen AI hardware with Cisco networking, observability and security products.
Physical AI
AMD also broadened its push into robotics and embedded AI. It introduced Ryzen AI Embedded X100 series processors, new Kria AI system-on-modules and a Robotics Developer Platform aimed at what it called physical AI.
The company said the robotics platform combines CPU, GPU, NPU and FPGA compute in one system. This reflects a broader effort by semiconductor companies to apply AI models beyond data centres and personal computers to machines that can sense and act in the physical world.
AMD linked the product push to a larger market opportunity, saying AI demand now extends across data centres, personal computers, edge devices and embedded systems. It said that breadth could expand its total addressable market to about USD $2 trillion by 2030.
Dr Lisa Su, Chair and Chief Executive Officer, AMD, framed the launch as part of a broader industry shift. "The next phase of AI will span frontier models, agents and physical AI, creating new opportunities to bring intelligence everywhere," said Su. "Realizing that potential will take the entire industry working together. AMD is partnering across the ecosystem to deliver leadership compute and open platforms that give customers the performance, flexibility and choice to scale AI from the data center to the edge."