Nebius and NVIDIA Open 2026 Physical AI Awards With Five $150,000 Compute Prizes

The program is aimed at startups whose physical AI product is already in the field and generating fleet data faster than teams can label it, and whose sim-to-real loop has become a release process rather than a research exercise. Nebius says it is looking for a demonstrable product, meaningful traction and a plan to scale. Along with the compute credits, each category winner receives joint promotion from Nebius and NVIDIA across both companies' channels, working sessions with a Nebius C-level executive on scaling, a Winners Circle feature published by Nebius, and two seats at an executive dinner next year with leaders from the physical AI ecosystem. There is no entry fee and no sponsored categories.

Because Nebius publishes its GPU-hour rates, the prize converts to a stated amount of compute. At on-demand prices, $150,000 buys roughly 38,960 GPU-hours on H100 at $3.85 per GPU-hour, about 33,330 hours on H200 at $4.50, about 20,980 hours on B200 at $7.15, and about 19,100 hours on B300 at $7.85, equivalent to roughly 203, 174, 109 and 99 days on an eight-GPU node respectively. For a team already running multi-node training, Nebius says $150,000 on H200s is about 521 hours on a 64-GPU cluster, roughly three weeks of continuous work; at preemptible H200 rates of $2.45 per GPU-hour it stretches to about 61,200 GPU-hours, a profile suited to checkpointable reinforcement learning and data-generation jobs. vCPU and RAM are bundled into the GPU rate, so the headline figure is not diluted by separate instance charges.

The five categories are Physical AI Models, covering vision-language-action models, vision-language models, world models and reinforcement learning; Perception and Spatial Intelligence, covering computer vision, SLAM, sensor fusion and scene understanding; Simulation and Synthetic Data, covering physics simulation, digital twins, sim-to-real transfer and synthetic data; Systems and Deployment, covering humanoids, industrial robots, autonomous mobile robots and warehouse automation; and Software, Tooling and Orchestration, covering middleware, MLOps, benchmarking, fleet management and deployment tooling. Each application covers one product. Companies with several products may enter additional categories, but Nebius advises leading with the strongest product in the single most relevant category. Eligibility requires a clear physical AI use case, an MVP in active use or testing, a registered legal entity and a live public website.

Entries are first screened in-house for quality and completeness, after which a panel selected by Nebius and NVIDIA scores each on depth of innovation, breadth of implementation, market potential and real-world impact. Finalists then go to a jury vote that decides the category winners.

The 2026 judging committee listed on the awards page comprises Evan Helda, head of Physical AI at Nebius; Anu Maheshwari, head of venture ecosystem at Nebius; Sumay Parikh, robotics and physical AI startup lead at NVIDIA; Amber Yang, a partner at Lightspeed; Adrian Macneil, chief executive of Foxglove; Brian Moore, co-founder and chief executive of Voxel51; Ulrik Stig Hansen, co-founder and co-CEO of Encord; Harry Mellsop, co-founder of Antioch; and Calvin Zhou, co-founder and vice president of engineering at RoboForce. Nebius expects to add more judges before the deadline. Three of the listed judges run companies that sell data and tooling infrastructure to physical AI teams, and Zhou's company won the industrial robotics category in 2025.

The first edition, run in 2025 as the Nebius Robotics and Physical AI Awards, drew 254 applications. Judges picked 55 finalists who shared $1.5 million in Nebius compute credits, with a panel that included executives and investors from Nebius, NVIDIA, OpenAI, Physical Intelligence, ABB, Khosla Ventures, Accel, Radical Ventures, Next47 and Shanda Grab Ventures. Several winners have since raised money: RoboForce, the industrial robotics winner, closed a $52 million round led by YZi Labs in March 2026 and reports more than 11,000 robot orders through letters of intent, while Gather AI, the vision AI and analytics winner, closed a $40 million Series B led by Smith Point Capital in February 2026.

Editor's Summary

Nebius and NVIDIA have opened the 2026 Physical AI Awards to global startups, offering $150,000 in compute credits in each of five categories, with entries due October 25, 2026. The prizes are pitched at teams moving from fielded products to scaled retraining, and the listed judging panel draws heavily on data and tooling operators plus one 2025 winner. The 2025 edition drew 254 applications and 55 finalists, and several past winners have since raised venture funding.