Nvidia unveils new AI chips
At GTC 2026 Nvidia revealed a fresh lineup of AI chips and an expanded software strategy geared to power everything from in‑car autonomy to large‑scale AI services — the keynote framed these chips as central to both automotive self‑driving systems and broader AI deployments Nvidia Unveils New AI Chips and Software Strategy at GTC 2026 - CXO Digitalpulse NVIDIA GTC 2026: Live Updates on What’s Next in AI | NVIDIA Blog.
The Vera Rubin platform announced) at GTC 2026 ships with seven new chips now in full production and configurable rack designs (NVL72 GPU racks, Vera CPU racks and related DSX reference systems). NVIDIA integrated) the Groq 3 LPU into a liquid‑cooled LPX rack containing 256 LPUs and large on‑chip SRAM to accelerate low‑latency inference, and the underlying Groq asset deal was reported at about $20 billion in December 2025. (cnbc.com) Dynamo 1.0 entered) production as an open‑source “inference operating system” that NVIDIA says integrates with TensorRT‑LLM and frameworks like LangChain to orchestrate GPUs at data‑center scale. (github.com) NVIDIA reported) recent benchmarks showing Dynamo can boost inference performance on Blackwell GPUs by up to 7×, and the company lists major cloud providers including AWS, Microsoft Azure, Google Cloud and OCI as early integrators. (nvidianews.nvidia.com) On agents and models, Nvidia unveiled) the NemoClaw stack to secure OpenClaw agents and launched an Agent Toolkit that VentureBeat says counts Adobe, Salesforce and SAP among 17 early adopters. (venturebeat.com) Automotive moves were specific: NVIDIA’s GTC newsroom listed) BYD, Geely, Isuzu and Nissan as DRIVE Hyperion adopters for Level‑4 projects, while Hyundai and Kia expanded strategic partnerships with the company. (nvidianews.nvidia.com) Jensen Huang projected) roughly $1 trillion in orders for Blackwell and Vera Rubin systems through 2027 during the keynote, a figure repeated across multiple financial and industry reports. (cnbc.com)
