NVIDIA Vera Rubin Hits Full Production, Delivers Spectrum-6 Networking for Gigascale AI Factories

NVIDIA's Vera Rubin platform is now ramping into full production, the company announced at Computex in Taipei. Taiwan's top server makers and global supply chain leaders are manufacturing Vera Rubin-based systems at scale, fueling AI labs, cloud providers and hyperscalers building the next generation of intelligence infrastructure. The platform unifies Vera CPUs, Rubin GPUs, BlueField-4 DPUs, Groq 3 LPX and Spectrum-6 Ethernet switches into a single rack-scale design.

According to NVIDIA's official newsroom, seven new chips are entering full production simultaneously, a first for the company. The Vera Rubin NVL72 GPU racks are already running at early partners including CoreWeave, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure. Spanning more than 350 factory sites across 30 countries, NVIDIA claims this is the largest, most mature rack-scale supply chain ever assembled to meet customer compute demand.

Vera Rubin delivers up to 10 times the performance per watt compared to its Grace Blackwell predecessor, CNBC reports from an exclusive first look at the system. While a single Vera Rubin rack draws approximately 600kW and individual chips hit thermal design points around 2,000 to 2,300 watts, the efficiency gains come from massive architectural improvements rather than raw power scaling. Tom's Hardware notes that NVIDIA reportedly boosted clock speeds and memory bandwidth, pushing power demand up by 500 watts per chip to stay ahead of AMD's Instinct accelerators.

The platform harnesses extreme codesign across hardware and software, slashing both training time and inference token generation cost. According to the NVIDIA Technical Blog, the Rubin platform combines the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet Switch into a single supercomputer architecture. The Vera Rubin platform is built on TSMC's N2 process with approximately 500 billion transistors and HBM4 memory providing 13 terabytes per second of bandwidth, as detailed by Introl's analysis.

Networking becomes the critical computing power multiplier at gigascale, and NVIDIA's Spectrum-6 Ethernet switch is purpose-built for Vera Rubin deployments. The next-generation Spectrum-X Ethernet connects hundreds of thousands of GPUs with the performance, resilience and efficiency required for frontier model training and agentic AI inference. NVIDIA's blog on the Spectrum-6 launch emphasizes that AI infrastructure leaders are already adopting the technology to drive token generation across massive GPU clusters.

Spectrum-X Ethernet Photonics, now in production, combines co-packaged optics with Spectrum-X switching to enable million-GPU AI factories, DQIndia reports. This optical networking approach addresses the bandwidth bottlenecks that emerge when scaling beyond tens of thousands of accelerators. The technology is part of NVIDIA's broader push to own every chip inside AI data centers, a strategy Wired highlighted this week as the company revealed new performance benchmarks ahead of rival AMD's annual product event.

NVIDIA is positioning Vera Rubin as the foundation for agentic AI factories, a concept introduced at GTC 2026 that reframes data centers as continuous intelligence production lines. Global Leaders Today reports that these AI factories represent a shift where businesses build entire operations around AI systems rather than treating AI as a tool they occasionally use. The Vera Rubin platform is designed to handle every phase of AI development, from pretraining and post-training to test-time scaling and agentic inference.

Stocktitan highlights that Vera Rubin delivers a tenfold increase in agent throughput specifically for agentic AI workloads. This matters because agentic AI, where models autonomously plan and execute multi-step tasks, requires fundamentally different compute patterns than traditional inference. NVIDIA's configurable AI infrastructure approach means the same rack-scale system can be optimized for different phases of the AI lifecycle, giving cloud providers flexibility in how they allocate compute resources.

NVIDIA's aggressive Vera Rubin production ramp is partly designed to ward off competition from AMD's Instinct AI accelerators. Tom's Hardware reports that NVIDIA increased boost clocks and memory bandwidth specifically to maintain a performance lead over AMD's upcoming offerings. The Wired analysis notes that NVIDIA's ambition extends beyond GPUs, the company wants to own every chip inside AI data centers, from CPUs and DPUs to networking switches and SuperNICs.

Cloud providers are already lining up. CoreWeave, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure have Vera Rubin NVL72 racks running in their environments. The platform reportedly delivers 8 exaflops of combined performance, equivalent to the entire TOP500 supercomputer list, according to Introl's specifications. With Rubin Ultra slated for the second half of 2027, featuring HBM4e and 365 terabytes of memory across NVL576 configurations, NVIDIA's roadmap extends well beyond the current launch.

Vera Rubin systems are scheduled to ship in the second half of 2026, with the production ramp already underway at the 350-plus factory sites NVIDIA has mobilized. The platform's power delivery requirements are substantial, Introl notes that 48 volt direct-to-chip power delivery is necessary for the 600 kilowatt per rack consumption. This infrastructure requirement will shape data center design decisions for years to come.

The competitive landscape is intensifying. AMD's annual product event in San Francisco will reveal its response to Vera Rubin, while hyperscalers continue developing custom silicon alternatives. But NVIDIA's scale advantage, with the largest rack-scale supply chain ever assembled, creates a meaningful barrier. The company's ability to ship seven new chips simultaneously across hundreds of factory sites in 30 countries sets a new standard for AI hardware deployment velocity, one that competitors will struggle to match in the near term.