Joint Initiatives Span Infrastructure and Open Model Optimizations, Offer Major Strides in Robotics, Drug Discovery and More
In a significant expansion of their longstanding collaboration, Google, Alphabet, and NVIDIA unveiled a series of new initiatives aimed at advancing the development of AI, expanding access to cutting-edge tools, and driving transformation across key industries including healthcare, manufacturing, and energy.
Teams of engineers and researchers across Google and Alphabet are closely collaborating with NVIDIA’s technical experts to harness the power of AI and simulation technologies. These joint efforts span projects ranging from robotic dexterity and drug discovery to intelligent energy grid optimization. Leveraging platforms like NVIDIA Omniverse™, NVIDIA Cosmos™, and NVIDIA Isaac™, teams from Google DeepMind, Isomorphic Labs, Intrinsic, and X’s moonshot initiative, Tapestry, are set to present key collaboration milestones at the GTC global AI conference.
To support these ambitious efforts, Google Cloud will be one of the first to deploy the newly announced NVIDIA GB300 and NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, delivering next-gen infrastructure to power AI research and production workloads.
In another milestone, NVIDIA becomes the first industry partner to adopt SynthID, an innovative watermarking technology from Google DeepMind that embeds secure, tamper-resistant identifiers into AI-generated content across images, audio, text, and video.
“I’m proud of our ongoing and deep partnership with NVIDIA, which spans the early days of Android and our cutting-edge AI collaborations across Alphabet,” said Sundar Pichai, CEO, Google and Alphabet. “I’m really excited about the next phase of our partnership as we work together on agentic AI, robotics, and bringing the benefits of AI to more people around the world.”
“Alphabet and NVIDIA have a longstanding partnership that extends from building AI infrastructure and software to advancing the use of AI in the largest industries,” said Jensen Huang, founder and CEO of NVIDIA. “It’s a great joy to see Alphabet and NVIDIA researchers and engineers collaborate to solve incredible challenges, from drug discovery to robotics.”
Advancing Responsible AI and Open Model Development
A shared commitment to responsible AI development lies at the heart of the Google DeepMind–NVIDIA collaboration. NVIDIA is now the first external partner to use SynthID, reinforcing trust and transparency by watermarking outputs from NVIDIA Cosmos world foundation models. Available via build.nvidia.com, this integration enhances content integrity without compromising visual or audio quality.
Further strengthening open innovation, the teams optimized Gemma—Google’s lightweight family of open models—for superior performance on NVIDIA GPUs. The latest release, Gemma 3, benefits from acceleration via NVIDIA’s AI platform and is deployed as a high-performance NVIDIA NIM microservice. With TensorRT-LLM under the hood, developers now enjoy enhanced inference speed and scalability.
This collaboration also extends to optimizing Gemini-based workloads on NVIDIA GPUs using Vertex AI, while NVIDIA Cosmos models are now featured in the Vertex AI Model Garden, enabling easier deployment and experimentation.\
Also Read: aiOla launches Jargonic, the most accurate speech AI
Ushering in the Era of Intelligent Robotics
Intrinsic, an Alphabet company, is on a mission to make adaptive AI-powered robotics more accessible to manufacturers. Currently, most industrial robots are programmed manually—a costly and time-intensive process.
Intrinsic Flowstate, a web-based digital twin and developer environment, now integrates with NVIDIA Isaac Manipulator foundation models, allowing developers to seamlessly move between simulation and real-world robotic workcells. This innovation dramatically reduces development cycles while increasing the adaptability of robotic systems.
At GTC, Intrinsic will also showcase an early USD streaming integration between Flowstate and NVIDIA Omniverse, delivering real-time cross-platform visualization of robotic environments.
Furthering this momentum, NVIDIA and Google DeepMind, in collaboration with Disney Research, are introducing Newton—a new open-source physics engine built on NVIDIA Warp and compatible with MuJoCo. Now powered by MuJoCo Warp, Newton can accelerate robotics machine learning workloads by over 70x, compared to MuJoCo’s current GPU simulator, MJX.
Real-World AI Applications for Health and Energy
Isomorphic Labs, led by Google DeepMind CEO Demis Hassabis, is redefining drug discovery by building a next-generation drug design engine. Hosted on Google Cloud and powered by NVIDIA GPUs, the platform delivers the performance and scalability required for advanced AI-driven biomedical research.
Meanwhile, Tapestry, X’s ambitious project for modernizing the electric grid, is leveraging AI to create smarter, more sustainable energy systems. Together with NVIDIA, Tapestry is exploring ways to accelerate and refine electric grid simulations, addressing challenges such as incorporating new energy sources and expanding capacity for AI and data center demands.
The joint research initiative will evaluate solutions like AI-powered interconnection optimization to enhance planning, ensure stability, and support infrastructure modernization in pursuit of a more sustainable energy future.
Redefining AI Infrastructure for the Future
Demonstrating its leadership in cloud-based AI infrastructure, Google Cloud is set to be among the first providers to offer Blackwell-powered instances—including the NVIDIA RTX PRO 6000 Blackwell and GB300—to its customers.
Following the recent preview of A4 and A4X virtual machines, Google Cloud became the first to support both NVIDIA B200 and GB200 instances. A4 is now generally available, with A4X launching soon, enabling customers to unlock Blackwell’s full potential via Google Cloud’s AI Hypercomputer.
Ongoing joint engineering efforts between Google and NVIDIA have also optimized essential frameworks like JAX and MaxText to run efficiently at scale on NVIDIA GPUs. MaxText, in particular, was co-developed to support training across tens of thousands of GPUs, enabling breakthrough performance for large model training.
These infrastructure advances are further bolstered by improvements to Google’s XLA compiler and OpenXLA, now deeply integrated with NVIDIA AI tools—including its newly launched reasoning model family. These models offer enhanced system utilization and are deployable across Google Cloud and Google Kubernetes Engine (GKE) environments.
