AMD and Marvell Unveil New Hardware for AI Data Centers
At the OCP Global Summit 2026, AMD introduced Helion, an open rack-scale design for agentic AI that can combine EPYC processors and Instinct GPUs. Marvell launched second-generation Structera X memory expansion controllers intended to increase AI server memory capacity and support larger models. Both companies' shares fell in Monday morning trading amid broader worries about AI spending and semiconductor valuations.
AMD's Helion is a reference rack design that lets operators mix sixth-generation EPYC chips with Instinct MI350P accelerators, or run a CPU-only setup with as many as 32,768 cores. It supports air and liquid cooling, includes a token router for directing model requests, and is slated for partner configurations in November 2026.
Marvell's second-generation Structera X controllers aim to give AI servers more memory, helping them handle larger models and access data more efficiently. Initial samples are expected in early 2027. Despite the announcements, both firms' shares dropped Monday morning as investors weighed wider worries about how much companies are spending on AI and how chipmakers are valued.
If adopted, these technologies may shape how AI services are built and priced. Cloud providers and enterprises could gain flexibility and memory capacity, potentially enabling more complex AI agents. Workers and consumers may eventually see effects through AI tools' capabilities, availability, and costs. Investors may remain sensitive to whether such hardware translates into demand.