NVIDIA Advances Physical AI with Project GR00T for Humanoid Robots

NVIDIA is accelerating the development of physical AI with the announcement of Project GR00T and the Cosmos world foundation model at its GTC 2026 conference. Project GR00T (Generalist Robot 00 Technology) is a general-purpose foundation model designed to empower humanoid robots to understand natural language and learn complex skills by observing human actions. To support this initiative, NVIDIA is releasing new simulation frameworks within Isaac Sim and new AI models. The company is deepening its partnerships with leading robotics firms like ABB Robotics, FANUC, KUKA, and YASKAWA, integrating its Omniverse and Isaac simulation platforms to develop and test sophisticated robot applications via digital twins. Furthermore, NVIDIA is collaborating with companies such as Skild AI and Workr to deploy AI-driven robotics for intricate manufacturing tasks like electronics assembly.

Sources:

Rakuten Launches Japan’s Largest Language Model: Rakuten AI 3.0

Rakuten Group has officially released Rakuten AI 3.0, its latest and most powerful Japanese large language model (LLM). This model was developed under the Generative AI Accelerator Challenge (GENIAC) project, a government initiative promoted by Japan’s Ministry of Economy, Trade and Industry (METI) and NEDO. Originally unveiled in December 2025, the newly fine-tuned Rakuten AI 3.0 is being hailed as Japan’s largest high-performance AI model. It is specifically designed to empower companies and developers building advanced AI applications. Rakuten has rigorously evaluated the model’s performance across multiple Japanese benchmarks, confirming its deep understanding of Japan-specific culture and history, alongside strong reasoning and instruction-following capabilities.

Sources:

Orange Business Introduces Trusted AI and Sovereign Cloud Solutions

At the Orange Business Summit 2026, Orange Business unveiled a new portfolio of technological innovations focused on Trusted AI, cloud, and secure connectivity. The company launched four new solutions aimed at helping enterprises enhance adaptability and ensure business continuity. A key offering is the expansion of its ‘Live Intelligence’ generative AI platform, which now provides customers with trusted AI agents through the ‘Live Intelligence Studio’. This studio enables clients to develop, deploy, and manage intelligent AI agents for task automation and data analysis within a secure, private infrastructure. Another significant launch is ‘Live Collaboration’, a suite of modular sovereign collaboration tools that integrates messaging, calendars, and video conferencing into a unified, secure platform.

Sources:

Kubernetes 1.36 Focuses on AI Workloads with Dynamic Resource Allocation

The upcoming Kubernetes version 1.36 has entered its code and test freeze period as of March 18, 2026, targeting a release date of April 22, 2026. This version will graduate 36 enhancements, with 18 becoming stable. A major theme of this release is improved support for AI and machine learning workloads. This is achieved through significant enhancements to Dynamic Resource Allocation (DRA), including the introduction of taints and tolerations for more granular management of specialized AI hardware. Other key alpha features include the ability to report the last usage time of a PersistentVolumeClaim (PVC), helping DevOps teams optimize costs by identifying and removing unused storage resources. The release also continues the platform’s evolution by deprecating the gitRepo volume driver.

Sources:

Breakthrough in Quantum Chip Design: 7,000 GPUs Used for Detailed Simulation

In a major step forward for quantum computing, researchers have successfully simulated a quantum chip in extreme physical detail before fabrication by leveraging a supercomputer with nearly 7,000 GPUs. This high-fidelity modeling, which accounts for how signals travel and interact within the chip, enables scientists to identify potential design flaws early and ensure optimal performance. This method provides a powerful new pathway to build better quantum hardware more rapidly by capturing the behavior of real materials, layouts, and qubits. The team from Berkeley Lab’s Quantum Systems Accelerator (QSA) utilized an exascale modeling tool named ARTEMIS for the simulation. The unprecedented precision, which involved discretizing the chip into 11 billion grid cells, was only possible through the massive parallel processing power of GPUs.

Sources: