NVIDIA GTC to Emphasize Edge Inference
NVIDIA’s GTC (March 16–19) is set to focus on running large models at the edge for robotics, AVs, and factory automation — with expectations of new inference chips, stacks, and partner announcements from Jensen Huang’s keynote reported reported.
Jensen Huang will deliver the GTC [keynote announced]nvidia.com on March 16 from the SAP Center in San Jose, and NVIDIA’s event page confirms the conference runs March 16–19 with a livestream option. nvidia.com Organizers expect more than 30,000 in-person [attendees reported]biztechmagazine.com, and the public schedule lists over 1,000 sessions plus roughly 240 NVIDIA Inception startups on the show floor. tech.yahoo.com NVIDIA has teased a “world‑surprising” chip ahead of the [keynote teased]biztechmagazine.com, and industry previews name the Vera Rubin platform as the likely Blackwell successor with claims of up to 10x lower inference token cost and first products targeted for H2 [2026 reported]abhs.in. Speculation around a dedicated low‑precision inference unit (LPU) and a possible “Feynman” reveal has circulated in analyst [coverage noted]seekingalpha.com, framing GTC as a possible launchpad for chips optimized specifically for on‑device and edge inference workloads. NVIDIA’s conference program explicitly calls out “Physical AI” and robotics tracks with speakers from Waabi, LangChain and [others listed]blogs.nvidia.com, while partners such as Smart Spatial plan enterprise digital‑twin demos built on Unreal Engine and NVIDIA Omniverse at GTC. marketwatch.com Preview coverage from outlets including Forbes and Yahoo highlights expectations for new inference stacks, enterprise migration paths (Blackwell → Vera Rubin), and partner announcements tying chips to software and AI‑factory [pipelines outlined]forbes.com.
