Advanced Micro Devices rolled out a suite of new hardware and software options during its Advancing AI 2026 conference, aiming to challenge rivals in the data center and robotics sectors. The Santa Clara, Calif.-based company introduced the Helios rackscale platform, the Robotics Partner Network, and updated roadmaps for its Instinct and Epyc product lines. The announcements signal a push into both high-volume AI inference and specialized physical AI applications.
A New Ecosystem For Robotics
AMD launched a Robotics Partner Network designed to help system integrators and hardware vendors build autonomous systems. The program includes members like World Wide Technology, Bosch’s Rexroth division, MulticoreWare, and Makarena. Companies in the network will use AMD’s open software stack, including ROCm, Vitis AI, and the ROS 2 Perception Node, to validate systems before deployment. This approach avoids proprietary runtimes and promises to reduce the time needed to move from prototype to production. The network operates on a tiered system with a free entry level for qualified companies, offering higher tiers that include co-marketing opportunities and early platform access.
AMD also revealed the Kria AI Solutions portfolio, featuring System-on-Modules powered by the Ryzen AI Embedded X100 Series processors. These chips are slated for release in the fourth quarter of 2026. The portfolio aims to provide deterministic real-time control for robotics, supporting more than 8,000 control decisions per second. It also promises to help users migrate code from Nvidia’s CUDA to AMD’s ROCm stack, though the company notes this migration will preserve only about 75 percent of existing code. The Kria AI Robotics Developer Platform is currently in sampling and will become generally available later this year.
Related: NetApp acquires DataPelago to ease AI storage strain
Helios Platform Targets Frontier AI
The centerpiece of the hardware announcements is Helios, an open AI infrastructure platform. Helios unifies AMD’s Instinct MI455X GPUs, sixth-generation Epyc server CPUs, Pensando networking, and ROCm software. The system is designed to support frontier AI, large-scale inference, and foundation-model training. AMD CEO Lisa Su stated that Helios shipments are expected to begin in the third quarter and ramp up during the fourth quarter. Major OEMs, including Bull, HPE, Lenovo, and Supermicro, are expected to offer systems based on the platform.
Helios spans 72 MI455X GPUs and 18 Epyc Venice CPUs. The vendor claims the platform offers better peak 4-bit floating point performance, high-bandwidth memory capacity, and higher token throughput compared to similar offerings from Nvidia. The platform scales from a single rack to gigawatt-scale AI clusters, though this level of expansion will likely depend on the specific OEM implementation. The Venice CPU is already in full production, with major cloud providers set to roll it out in the fourth quarter.
Next-Generation Processors And Roadmap
AMD detailed the specifications for its upcoming Ryzen AI Embedded X100 Series processors. These chips integrate up to 16 Zen 5 CPU cores and a neural processing unit. They promise 2.1 times higher multithread CPU performance and 3.5 times higher token generation throughput compared to previous generations. The processors are rated to operate in harsh environments, functioning between -40 degrees Celsius and 105 degrees Celsius for up to 10 years. The company also introduced the Ryzen AI Max Pro 400 series, which will feature 192GB of unified memory and support for models up to 300 billion parameters.
Related: KeyBanc questions Salesforce Agentforce maturity amid weak traction
The company provided a timeline for future releases through 2030. The roadmap includes next-generation Epyc server CPUs based on the Zen 7 architecture in 2028, followed by the Ravenna series based on Zen 8 in 2030. On the GPU side, AMD plans to launch the Instinct MI500 Series in 2027 and the MI600 Series in 2028. These will feed into next-generation Helios rackscale products, such as the Helios 500 and Helios 600, which will utilize updated Epyc and Pensando networking components.
AMD is also simplifying its software development environment with the upcoming launch of ROCm.ai. The platform, available in August, will offer AI-assisted development and intelligent deployment tools. The vendor claims the latest version of ROCm will provide an average 3.3 times improvement in inference and 2.4 times improvement in training compared to its predecessor. This move aims to lower the barrier for developers working with AMD hardware.
