Nvidia releases five new software and hardware platforms targeting enterprise customers building proprietary AI infrastructure.

At CES 2026, Nvidia Chief Executive Jensen Huang positioned the company as an end-to-end infrastructure provider, unveiling five interconnected platform announcements aimed at reducing operational costs and expanding market reach. The launches target distinct segments of the AI value chain, including data center compute economics, autonomous vehicle commercialization, industrial manufacturing workflows, desktop development environments and consumer gaming performance. These announcements reflect a calculated shift from discrete chip sales to integrated systems that address enterprise infrastructure bottlenecks, with organizations now evaluating deployment timelines, integration complexity and total cost of ownership as Nvidia accelerates its product roadmap ahead of original schedules.

Nvidia announced that its Rubin platform has entered full production with partner deployments expected in the second half of 2026. The platform integrates six distinct silicon components: Rubin GPU, Vera CPU, NVLink 6 interconnect, ConnectX-9 network interface, BlueField-4 data processing unit and Spectrum-6 ethernet switch. The Rubin GPU delivers 50 petaflops of NVFP4 inference compute through sixth-generation tensor cores that implement adaptive precision scaling, using 4-bit floating point operations with dynamic per-block scaling factors that reduce memory requirements by 3.5 times compared to 16-bit precision while maintaining model accuracy within 1 percent degradation. The architecture triples memory bandwidth to 22 terabytes per second through HBM4 integration and increases capacity to 288 gigabytes per GPU. The Vera CPU features 88 custom ARM-compatible cores, 1.2 terabytes per second of LPDDR5X memory bandwidth, and 1.8 terabytes per second of NVLink-C2C connectivity, handling data movement and agentic reasoning workflows while supporting full confidential computing. Nvidia claims the platform reduces inference token costs by up to 10x compared to Blackwell while requiring 4x fewer GPUs for training mixture-of-experts models, with CoreWeave confirming plans to be among the first cloud providers deploying Rubin-based systems.

Nvidia introduced Alpamayo, an open portfolio of vision-language-action models designed for level 4 autonomous vehicle development, with the first model being a 10-billion parameter architecture built on the Cosmos vision-language foundation that outputs both driving trajectories and reasoning traces explaining decision logic. The approach differs from perception-only systems by enabling vehicles to verbalize why they take specific actions, addressing regulatory requirements for safety validation and transparency as automakers work toward commercial deployment. Mercedes-Benz will launch the first passenger vehicle featuring Alpamayo in the all-new CLA, with United States production of level 2 point-to-point driver assistance expected by year end. Nvidia expanded its DRIVE Hyperion ecosystem to include tier 1 suppliers and sensor partners including Bosch, Magna, ZF Group, Aeva, Hesai and Sony, with the platform combining two DRIVE AGX Thor systems-on-chip delivering over 2,000 FP4 teraflops for real-time sensor fusion and transformer-based perception. Companies including International Motors and PlusAI adopted the platform for factory-built autonomous trucks.

Nvidia and Siemens announced an expanded partnership to build what they describe as an industrial AI operating system, aiming to integrate AI across product design, electronic design automation, manufacturing execution and supply chain operations. Siemens will deploy the first fully AI-driven adaptive manufacturing site in 2026 at its electronics factory in Erlangen, Germany. The partnership includes GPU acceleration across Siemens' simulation portfolio with CUDA-X libraries and PhysicsNeMo integration targeting 2-10 times speedups in verification, layout and process optimization workflows, with the companies applying industrial AI principles to electronic design automation to address bottlenecks in chip development cycles. Nvidia will provide AI infrastructure, simulation libraries, models and frameworks while Siemens commits hundreds of industrial AI specialists and hardware and software platforms, validating technologies on their own systems before scaling to customers.