NVIDIA AI Updates: November 4, 2026

1. NVIDIA Reportedly Tells Cloud Customers to Take Less Business From Chinese Companies and Disclose Who Uses the Chips

NVIDIA has told some cloud customers, including neoclouds it has backed, to take on less business from Chinese companies, The Information reported citing three people familiar with the matter, as summarized by The Decoder. The guidance is described as a push toward a more diverse customer mix rather than a requirement: it is not mandatory and applies to future contracts, not existing deals, and cloud customers are also being asked to disclose the identity of the end users accessing the chips. According to the Financial Times, the instruction was passed on verbally through sales reps and regional teams. The move targets a gap in U.S. export controls, which restrict chip shipments but generally do not cover remote access through cloud services, so Chinese AI developers can rent advanced NVIDIA capacity in data centers abroad without importing any hardware; U.S. authorities had been looking into such access after allegations involving Moonshot AI, the developer of the Kimi chatbot. NVIDIA has not announced the guidance publicly, and The Decoder’s summary includes no company comment. Source

2. NVIDIA Rounds Up City Deployments of Nemotron, Cosmos and Metropolis Agents Ahead of Smart City Expo

NVIDIA published a roundup of municipal and infrastructure deployments built on its agentic and physical AI stack, timed to Smart City Expo World Congress in Barcelona (November 3 to 5), where NVIDIA’s EMEA vice president Paolo Guglielmini is giving a keynote. In Spain, 1MillionBot combines the open ALIA models trained at the Barcelona Supercomputing Center with Nemotron and other models in resident-service agents for Madrid Digital (Comunidad de Madrid), the Administració Oberta de Catalunya and more than 150 deployments across the country, handling municipal procedures, appointments and urban-planning incidents. In Malaysia, ITMAX’s SamurAI V2 vision agents, built with reference to Cosmos, Nemotron and the Metropolis video search and summarization (VSS) blueprint, are scaling on RTX PRO 6000 GPUs toward more than 17,000 cameras by 2027, and ITMAX uses Laplace Intelligence’s open-source TeraSim traffic simulator to test signal timing, which NVIDIA says cut peak-hour corridor travel times by up to 26%. K2K in Wrocław, Poland, runs Cosmos reasoning models and VSS on one shared infrastructure for mobility, safety and event response; Honeywell’s Forge Plant Cognition uses Nemotron-powered agents on operational data at Dangote’s Lekki refinery in Nigeria, on Dell PowerEdge servers with RTX PRO GPUs, with an Omniverse digital twin planned; and Linker Vision and the architecture firm MVRDV in Taiwan use Omniverse and Nemotron agents to simulate 15 years of urban forest canopy growth in 10 minutes, cutting a design iteration from two weeks to under an hour. NVIDIA says some city workloads can start on RTX workstations or a small GPU system rather than a large infrastructure buildout. Source

3. NVIDIA Lists New Fab Partner Results: 17x Lithography Speedups in Siemens Calibre and Millisecond Plasma Predictions at Lam Research

NVIDIA laid out what it calls the “AI-defined fab,” citing partner results across accelerated simulation, physics-trained AI models, agents and digital twins. Siemens is integrating cuLitho into its Calibre computational lithography software for up to 17x acceleration on key simulation workloads, Applied Materials’ ACE+ multiphysics platform runs topography simulations up to 35x faster with PhysicsNeMo, and Lam Research showed PhysicsNeMo predicting plasma behavior for etch processes in milliseconds, where a full simulation takes days. On the test floor, Advantest uses NeMo Switchyard agents to automate the manual, sequential process of choosing which AI model to apply to a task, cutting response times from hours to minutes, and Applied Materials builds digital twins of entire fabs on Omniverse libraries to simulate material flow, find bottlenecks and evaluate layouts before changes reach the line. The post’s cuDSS figures (up to 10x for COMSOL multiphysics and for Applied’s Ginestra platform) repeat partner results NVIDIA reported in October. Source