NVIDIA lands Singapore testbed

Singapore is putting Nvidia at the center of a new robotics push, pairing a local research hub with a public testbed meant to move “physical AI” from lab work into field deployment. The announcements were made on May 20 at ATxSummit, where the city-state said Nvidia will open its first Singapore research center and its second such research presence in Asia Pacific. Singapore also said it will launch its first physical AI testbed later this year to let companies co-design, test and validate commercially viable robotics services. The Nvidia lab will focus on embodied AI and on improving the efficiency of AI infrastructure, working with university researchers, industry partners and government agencies, CNBC reported. In parallel, Nvidia and Google Cloud said this week they are expanding developer access to Nvidia’s software stack and tools, including resources tied to physical AI and production deployment on Google Cloud. (cnbc.com) ### Where will the Singapore testbed actually sit? Punggol Digital District is the site Singapore has chosen for the new testbed, according to reporting on the ATxSummit announcements. The Infocomm Media Development Authority, JTC and the Singapore Institute of Technology are developing the site as a mixed-use public area where robots and AI systems can be tested outside controlled environments. (cnbc.com) Singapore has described it as its first testbed for multi-use-case and multi-operator deployments at scale. That matters because the setup is aimed at shared, real-world trials rather than one company running a closed pilot on its own premises. ### Which companies are first in line to use it? Certis, DHL, Grab and QuikBot are expected to be among the first companies to use the testbed, according to CNBC and local reporting on the summit. (techedt.com) The stated goal is to help them co-design, deploy, test and validate robotics services that could be used commercially in public spaces. Slamtec, Unitree and QuikBot are also part of a separate set of trials through a new Center for Intelligent Robotics, with early use cases including food and parcel delivery, cleaning and security patrolling to complement human operations. Those use cases give a clearer picture of the kinds of service-sector deployments Singapore wants to evaluate first. (cnbc.com) ### What does Nvidia gain beyond another research office? Nvidia is adding a research base in a country that is openly pitching itself as a regional hub for AI development, testing and deployment. CNBC reported that the Singapore lab will be Nvidia’s first research hub in the city-state and its second in Asia Pacific. Nvidia has spent the past year building out a broader physical AI stack around robotics and autonomous machines. (cnbc.com) The company said Cosmos, launched in January 2025, is a platform of world foundation models, tokenizers, guardrails and video-processing tools built for robots and autonomous vehicles, and said later releases added controllable world generation, reasoning and synthetic-data tools for physical AI developers. ### Where does Google Cloud fit into this? Nvidia said on May 19 that more than 100,000 developers had joined its joint developer community with Google Cloud. The companies said the community offers learning paths, hands-on labs and events built around the full-stack Nvidia AI platform on Google Cloud. Google Cloud and Nvidia have separately said their partnership is aimed at pushing “agentic and physical AI” into production, from robots to digital twins on factory floors. (nvidianews.nvidia.com) Google Cloud executive Mark Lohmeyer said the companies were combining Google Cloud infrastructure and managed services with Nvidia platforms, systems and software so customers could train, tune and serve physical AI workloads. (blogs.nvidia.com) ### Why does Cosmos keep coming up in this story? Cosmos is Nvidia’s world-model platform for physical AI, and the company says it is designed to generate photoreal, physics-based synthetic data and support training and evaluation for robots and autonomous vehicles. Nvidia has also said newer Cosmos releases support robot policy evaluation in simulation and synthetic data generation for perception systems. (hpcwire.com) Nvidia’s own materials say Cosmos is available through build.nvidia.com and related open-model resources, while the company’s Google Cloud announcements tie that model layer to broader cloud tooling and developer workflows. Later in 2026, the next visible milestone will be the opening of the Punggol testbed and the first trials involving Certis, DHL, Grab, QuikBot and other robotics partners. (nvidia.com) (nvidianews.nvidia.com)