The future of Artificial Intelligence (AI) isn’t just about powerful chips; it’s about seamless, scalable, and secure ecosystems. As agentic AI drives ever-higher demands across industries, AMD is stepping forward, delivering a groundbreaking, open rack-scale AI infrastructure designed to power this transformative wave.
From its formidable current offerings to a tantalizing glimpse into the future with “Helios,” the company is cementing its position at the forefront of the artificial intelligence revolution.
Agentic AI, the next frontier of intelligent automation, empowers systems to not only automate tasks but to orchestrate complex workflows and drive critical decision-making. These sophisticated operations demand more than just raw GPU horsepower.
They require a holistic stack where high-performance CPUs manage intricate tasks, secure data flows, and handle enterprise-grade workloads with unparalleled efficiency. The chipset maker, with its comprehensive portfolio spanning Instinct GPUs, EPYC CPUs, and Pensando DPUs, is uniquely positioned to deliver this full-stack solution, all built upon an open, flexible, and fully programmable architecture.
Built for today: A foundation of open standards, leadership performance
AMD recognizes that the era of agentic AI is not a distant vision — it’s here, and enterprises are demanding robust, scalable solutions now. To meet this immediate need, the company and its partners are actively deploying an open standard AI Rack infrastructure, currently supporting up to 128 GPUs.
This powerhouse combines AMD Instinct MI350 series GPUs, 5th Gen EPYC™ CPUs, and Pensando Pollara 400 NICs, all integrated into an industry-standard Open Compute Project (OCP) and Ultra Ethernet Consortium (UEC) compliant design.
This rack-scale solution is defined by four guiding principles:
- Leadership compute performance: The AMD Instinct MI350 Series GPUs offer exceptional inference and training performance, boasting high memory capacity — up to 36TB of HBM3E in a 128-GPU rack — and up to 3.58x more FP6 performance than competitors.
- Enterprise-grade CPUs: 5th Gen AMD EPYC™ processors provide a robust, industry-standard x86 host for seamless integration with existing enterprise applications.
- Advanced networking: AMD Pensando Pollara NICs stand out as the industry’s first UEC-ready AI NICs. They feature programmable transport, path-aware congestion control, and intelligent packet spraying, ensuring leading performance for even the largest AI clusters.
- Open standards and design: By embracing open standards like the Ultra Ethernet Consortium and the holistic design principles of the Open Compute Project, AMD ensures interoperability and seamless integration, providing partners and customers with crucial choice and readily available solutions.
These integrated rack solutions, available in both liquid and air-cooled configurations, are already enabling a new class of differentiated, high-performance artificial intelligence infrastructure in the market.
A glimpse into the future: Introducing AMD “Helios,” the next frontier
AMD CEO Lisa Su at the launch of AMD “Helios”
Looking ahead, AMD is pushing the boundaries of AI infrastructure even further with the preview of “Helios.” This purpose-built, unified solution, expected in 2026, represents the next major leap in the company’s AI roadmap, engineered from the ground up to support the most demanding AI workloads with unprecedented integration and compute density.
“Helios” will redefine rack-level capabilities by seamlessly blending the company’s silicon, software, and systems expertise. It’s designed for both scale-up and scale-out scenarios, ensuring leadership performance in large-scale training and distributed inference alike. This ready-to-deploy solution is set to accelerate time to market for critical AI applications, from training frontier models to fine-tuning enterprise-specific AI on proprietary data.
The “Helios” platform will integrate a powerful trio of next-generation AMD technology:
- Next-Gen AMD Instinct MI400 Series GPUs: These GPUs are anticipated to offer up to 432 GB of HBM4 memory, deliver 40 petaflops of FP4 performance, and boast 300 gigabytes per second of scale-out bandwidth. This will bring rack-scale AI performance leadership for training massive models and running distributed inference.
- Open Scale-Up with UALink: Performance will scale effortlessly across 72 GPUs within Helios, thanks to UALink. This new open standard enables customer choice and interoperability in scale-up fabrics, ensuring every GPU in the rack can communicate as a single, unified system for breakthrough performance.
- 6th Gen AMD EPYC “Venice” CPUs: Powered by the groundbreaking “Zen 6” architecture, these CPUs are expected to offer up to 256 cores and 1.6 TB/s of memory bandwidth, sustaining maximum performance across the entire “Helios” rack.
- AMD Pensando “Vulcano” AI NICs: This next-generation NIC for AI scale-out will be UEC 1.0 compliant, supporting both PCIe® and UALink interfaces for direct connectivity to CPUs and GPUs. With 800G network throughput and an expected 8x the scale-out bandwidth per GPU compared to the previous generation, “Vulcano” will virtually eliminate communication bottlenecks in large-scale AI deployments.
Industry validation: Oracle Cloud infrastructure leads the way
AMD’s innovative approach is already gaining significant traction with industry leaders.
Oracle Cloud Infrastructure (OCI) will be among the first to adopt the AMD Instinct MI355X-powered rack-scale solution. This significant partnership underscores OCI’s commitment to offering one of the broadest AI infrastructure portfolios and highlights the tangible value AMD solutions bring to enterprise-level generative and agentic AI applications globally.
“Oracle Cloud Infrastructure continues to benefit from its strategic collaboration with AMD,” stated Mahesh Thiagarajan, Executive Vice President at Oracle Cloud Infrastructure. “We will be one of the first to provide the MI355X rack-scale infrastructure using the combined power of EPYC, Instinct, and Pensando. We’ve seen impressive customer adoption for AMD-powered bare metal instances, underscoring how easily customers can adopt and scale their AI workloads with OCI AI infrastructure.”
By extending its solution engineering from the node to the rack and cluster level, AMD is directly addressing the massive scale required for autonomous agents. Its unwavering commitment to open-standards innovation, a robust partner ecosystem, and aggressive investment in capabilities position AMD at the heart of integrated AI solutions.
With its ready-to-deploy open rack-scale AI infrastructure today, and the visionary “Helios” on the horizon, AMD is not just participating in the AI era — it’s actively shaping its unprecedented future.
