AMD unveiled the Instinct MI400X graphics processor and EPYC 9006 server CPU at its Advancing AI 2026 conference on July 22-23, claiming both are the fastest in their class and forming the core of its new Helios rack-scale AI system; the company also launched ROCm.ai software and announced major customer deployment plans, positioning the full stack as a direct challenge to Nvidia's installed base
The US chip designer competing for second place in AI accelerators, with a US$34.6bn 2025 revenue base and a 6GW GPU supply pact with OpenAI.
full dossier ›Hyperscalers plan US$725 billion in 2026 capex for GPU campuses, chip contracts, and power deals that determine who can train and serve frontier AI models.
A 560-page spec gives Ethernet InfiniBand-class RDMA and congestion control; the 2026 work item is Programmable Congestion Management
The 115-member consortium ratifies In-Network Compute, a UCIe 3.0-compliant Chiplet spec, Manageability and a split 200G physical layer, scale-up fabric as an open standard
First 1GW of Instinct MI450 deploys 2H 2026; the Helios rack targets ~3 AI exaflops with 72 MI455X accelerators, AMD's bid to break Nvidia's data-center grip