Vera Rubin mass production

SK hynix said April 20 it has begun mass production of a 192-gigabyte memory module built for Nvidia’s Vera Rubin AI systems. (news.skhynix.com) The module is called SOCAMM2, short for Small Outline Compression Attached Memory Module 2, and it uses LPDDR5X low-power dynamic random-access memory made on SK hynix’s 1cnm process. SK hynix said the product is designed as a primary memory option for next-generation AI servers. (news.skhynix.com) Memory is the short-term workspace beside an AI chip, and bigger, faster modules let systems move more data without waiting on storage. SK hynix said its SOCAMM2 delivers more than double the bandwidth and more than 75% better power efficiency than conventional registered dual in-line memory modules, or RDIMMs. (news.skhynix.com) Nvidia said March 16 that Vera Rubin is now in full production as a platform, with partner products becoming available in the second half of 2026. The company said the system combines the Vera central processing unit, Rubin graphics processing unit, NVLink 6 switching, ConnectX-9 networking and BlueField-4 data processing units into rack-scale AI systems. (nvidianews.nvidia.com) That puts SK hynix’s announcement in the middle of a supply-chain race that starts months before customers see finished racks. Memory vendors, chipmakers and cloud providers all have to line up parts early because these systems are sold as tightly integrated clusters, not loose boxes of components. (nvidianews.nvidia.com) (news.skhynix.com) SK hynix said SOCAMM2 adapts low-power memory first used mainly in smartphones for server use, with a compression connector meant to improve signal integrity and make modules easier to replace. The company said the design targets memory bottlenecks in training and inference for large language models with hundreds of billions of parameters. (news.skhynix.com) The timing also lands while Blackwell is still scaling. Microsoft said in September 2025 that its Fairwater datacenter in Wisconsin would use a single flat network connecting hundreds of thousands of the latest Nvidia graphics processing units, and described that site as one of several identical Fairwater facilities under construction in the United States. (blogs.microsoft.com) Microsoft said Fairwater covers 315 acres and 1.2 million square feet across three buildings, built to operate as one AI supercomputer for training and inference. That means the industry is adding Blackwell-era capacity at hyperscale while suppliers prepare parts for Rubin-era systems due in the second half of 2026. (blogs.microsoft.com) (nvidianews.nvidia.com) Nvidia has framed Vera Rubin as the next step after Blackwell in a shift from individual servers to pod-scale AI factories. SK hynix’s move shows that, by April 2026, that handoff is no longer just a roadmap slide but a manufacturing schedule. (nvidianews.nvidia.com) (news.skhynix.com)