NVIDIA x Modal (Vera CPUs for Sandboxes)

World leader in accelerated computing and AI (NASDAQ: NVDA). In physical AI, NVIDIA develops the Isaac robotics platform: the GR00T open foundation model, Jetson edge compute (AGX Thor), Isaac Lab and Isaac Sim simulation environments, and Omniverse. NVIDIA positions itself as the platform layer for the humanoid industry. Developer of [GR00T N1](/brains/gr00t-n1), the first open humanoid foundation model.

Modal Labs is a serverless GPU cloud built for AI and data workloads, letting developers spin up containerized functions on Nvidia hardware billed by the CPU cycle rather than by the reserved instance. Founded in 2021 by Erik Bernhardsson (formerly of Spotify and Better.com) and Akshat Bubna, the company sits in the mid-market inference and fine-tuning slice of the neocloud landscape, competing less with the hyperscale training clouds like CoreWeave and Crusoe and more with Replicate, Baseten, and RunPod for the developer-facing long tail.\n\nRather than owning campuses, Modal operates as a software layer over rented GPU capacity, with a proprietary container stack that advertises sub-second cold starts and per-second billing. Its published on-demand rate card runs from $0.59/hr on a T4 up to $6.25/hr on a B200 and $7.10/hr on a B300, and its customer roster leans toward the current generation of AI-native startups: Suno, Runway, Physical Intelligence, Cognition, Lovable, Decagon, Ramp, and Substack all appear as named users, alongside Meta and Scale.\n\nAfter a $16M Series A led by Redpoint in October 2023, Modal was reported by The Information in February 2026 to be in talks to raise at a $2.5B valuation, a step-change that would put it in the mid-tier of the AI-infrastructure funding board without the balance-sheet weight of the campus-owning neoclouds.