NVIDIA Opens Vera Rubin Architecture to Accelerate the Global AI Race

NVIDIA has announced a major expansion of its open AI infrastructure strategy, contributing key designs for its next-generation Vera Rubin computing platform to the Open Compute Project (OCP). The announcement was made on October 12, 2026, at the OCP Global Summit in San Jose, California. Through its modular NVIDIA MGX architecture, the company is sharing rack and compute-tray designs intended to help manufacturers and data-centre operators develop compatible infrastructure for advanced AI workloads. The initiative comes as technology companies invest heavily in computing capacity to support increasingly sophisticated AI models and autonomous agents. NVIDIA says standardised infrastructure designs can simplify deployment, improve interoperability and help businesses scale their AI systems more efficiently.

The Vera Rubin platform represents NVIDIA’s next major step in accelerated computing. Named after astronomer Vera Rubin, the platform combines multiple computing and networking components into an integrated system designed for demanding AI training and inference workloads. NVIDIA says its Vera Rubin NVL72 configuration brings together 72 Rubin GPUs, 36 Vera CPUs, networking equipment and data-processing units within a rack-scale system. This integrated design aims to improve communication between processors and reduce the cost of running increasingly complex AI models. As businesses move towards AI agents that can perform multi-step tasks, analyse large volumes of information and interact with software tools, the demand for high-performance computing systems capable of handling these workloads is expected to remain substantial.

A central element of the announcement is the expansion of NVIDIA’s MGX architecture, which allows manufacturers to develop modular data-centre systems using standardised components. According to NVIDIA, the updated MGX 1.2 specification introduces improvements for high-density racks, liquid cooling, power distribution and maintenance. The design supports deployments of up to 250 kilowatts per rack and includes standardised interfaces intended to simplify integration between computing equipment and supporting infrastructure. These developments are increasingly important because advanced AI processors generate substantial heat and require significant amounts of electricity. By providing common reference designs, NVIDIA hopes to help system builders reduce engineering work and accelerate the construction of large-scale AI facilities.

NVIDIA is also expanding its partnerships with major technology companies. Meta is adopting NVIDIA’s Spectrum-6 Ethernet switches, which provide networking capacity of up to 102.4 terabits per second, to support the development of larger AI infrastructure systems. NVIDIA also announced work with Cisco on an ultra-dense networking module standard and collaboration with Google on transport technology for future networking products. These partnerships demonstrate that building AI infrastructure involves much more than manufacturing powerful processors. High-speed networking, cooling systems, electrical distribution and software must work together to keep large computing clusters operating efficiently. Greater standardisation could help suppliers collaborate more easily and enable data-centre operators to expand capacity without designing every component independently.

The announcement also comes as competition in AI hardware intensifies. Microsoft revealed its Maia 300 custom AI chip on October 12, saying it plans large-scale deployments beginning in 2027 to support its own AI services and reduce computing costs. Other technology companies are also developing specialised processors and infrastructure to meet their growing requirements. NVIDIA’s open-design strategy could help strengthen its position by encouraging more manufacturers to build systems around its architecture, even as customers explore alternatives to its GPUs. However, the long-term impact will depend on factors including performance, energy efficiency, component availability and the total cost of operating AI facilities. Opening reference designs does not mean that every part of NVIDIA’s technology is freely available, nor does it eliminate the cost of the underlying hardware.

The broader significance of NVIDIA’s announcement is that the AI industry is increasingly focused on building complete computing infrastructure rather than competing solely on individual chips or AI models. As organisations deploy autonomous agents and more advanced generative AI applications, demand is growing for systems that can support continuous processing, rapid data exchange and complex workloads at scale. By contributing Vera Rubin rack designs to an open industry initiative, NVIDIA is seeking to make its infrastructure easier for partners to adopt while helping establish common standards for future AI data centres. Whether this strategy produces faster deployment and lower operating costs will become clearer as manufacturers adopt the designs and new systems enter service. For now, the announcement highlights the growing importance of collaboration between chipmakers, cloud providers and infrastructure manufacturers in shaping the next phase of AI development.