CES 2026 saw AMD take the stage with a bold, all‑in‑one message: “AI everywhere.” CEO Lisa Su introduced the Helios rack‑scale platform as the backbone of that vision, promising to bring massive AI compute out of the cloud and into every data centre, edge node and even the laptop on your desk. Helios packs up to three exaflops of AI performance into a single, liquid‑cooled rack, leveraging the latest Instinct MI455X GPUs paired with AMD’s next‑gen EPYC “Venice” CPUs. The design is fully open, aligning with the Open Compute Project, and ships with the ROCm 7 software stack so developers can tap into the same tools they use on Nvidia or Google clouds. By exposing a unified API set and supporting industry‑standard frameworks such as TensorFlow, PyTorch and ONNX, AMD aims to lower the barrier for enterprises that want to run custom models without rewriting code for each accelerator. Moreover, the platform includes built‑in safety and security features—hardware‑rooted attestation, encrypted memory channels and a hardened software supply chain—that are becoming table‑stakes for hyperscale customers handling sensitive workloads.

Beyond raw horsepower, AMD highlighted a series of partnerships that give the platform real‑world footing. HPE, Oracle and a host of hyperscale providers have already signed on to deploy Helios in their facilities, while a 6‑gigawatt infrastructure deal with OpenAI underscores the company’s commitment to feeding the next wave of large‑scale model training. The rack’s 72 Instinct accelerators, each sporting 320 billion transistors and 432 GB of HBM4 memory, deliver a ten‑fold jump over the previous MI355X generation. Networking is handled by Pensando “Vulcano” NICs, providing ultra‑low latency that makes near‑real‑time inference possible even at the edge. In addition, AMD has integrated its Infinity Fabric interconnect to enable seamless, cache‑coherent scaling across multiple racks, a feature that will be crucial for workloads that span hundreds of nodes. The system also supports dynamic resource partitioning, allowing a single Helios rack to be sliced into isolated AI training, inference and HPC slices that can be re‑allocated on the fly based on demand.

AMD didn’t stop at the data centre. The keynote also unveiled a suite of AI‑enabled client products, including the Ryzen AI 400 series that brings up to 60 TOPS of NPU performance to laptops, and the Ryzen AI Halo developer kit for on‑device experimentation. Embedded chips for robotics and autonomous systems were showcased, delivering 60 TOPS while staying power‑efficient. A humanoid robot demo highlighted how AMD’s silicon can handle tactile sensing and real‑time decision‑making, reinforcing the “AI everywhere” narrative. To nurture the next generation, AMD announced free access to its AI development tools for universities worldwide, along with a curriculum that maps hardware capabilities to real‑world AI challenges such as drug discovery, climate modelling and edge vision. The company also launched a cloud‑based sandbox—AMD AI Foundry—that lets students and researchers spin up pre‑configured environments on Helios hardware, complete with Jupyter notebooks, model libraries and performance profiling dashboards.

The market reaction was immediate: AMD’s stock rallied as analysts pointed to a potential 10 % share of the AI compute market by 2030. With a roadmap that includes the MI500 series promising a 1,000× performance boost, the company is positioning itself as a flexible, open alternative to more proprietary ecosystems. Energy efficiency and scalability are core to the Helios story, aligning with global sustainability goals while meeting the surging demand for AI workloads. If AMD can deliver on its promises, the “AI everywhere” slogan may soon become a reality across clouds, data centres, edge devices and everyday laptops, reshaping how enterprises and developers think about deploying artificial intelligence at scale.