Nvidia welds Groq LPUs to Vera Rubin racks

Nvidia is turning its AI stack into something that looks a lot like an operating system for agents, pairing new “Claw”-branded software with Groq-style inference chips and Vera Rubin rack systems that lock in vast amounts of capital — and customers — at once. The move builds on a roughly $20 billion licensing and talent deal that gives Nvidia access to Groq’s ultra‑low‑latency Language Processing Unit (LPU) technology for inference, a partnership analysts already frame as a strategic win for extending CUDA into Groq’s ecosystem, according to coverage of the agreement by Investing.com and public filings summarized on Wikipedia.

On the hardware side, Nvidia is seeding customers with samples of its next‑generation Vera Rubin platform — modular, cable‑free trays that bundle Vera CPUs, Rubin GPUs with stacked HBM4 memory and NVLink 6.0 switching into pre‑configured NVL72 rack systems aimed squarely at hyperscale AI data centers, as detailed by Tom’s Hardware. Partners from Foxconn to Supermicro are being handed near‑finished compute trays, limiting their design freedom but accelerating Nvidia’s ability to dictate the physical shape of AI factories. That follows a broader push to blueprint “AI infrastructure” with utilities and data‑center operators, including national‑scale build‑outs with partners like YTL Power’s Malaysian campus, noted on YTL Power’s profile, and massive U.S. projects highlighted in a recent Nvidia infrastructure release.

The same centralization is creeping into orbit. Nvidia‑backed startup Starcloud is already preparing H100‑powered satellites to run generative models in space, using GPUs comparable to terrestrial data centers for training and inference, according to an Nvidia blog post on Starcloud and broader reporting on the space‑based data‑center race compiled on Wikipedia. For governments and enterprises, this emerging “agent OS plus superhardware” stack promises deterministic, low‑latency agents that live across earthbound and orbital racks — but at the cost of deeper dependency on a handful of U.S. vendors, new supply‑chain chokepoints and fresh geopolitical flashpoints as rivals move to restrict or replicate the same capabilities.