NVIDIA’s next-generation Vera Rubin AI chip platform has moved from production lines into active shipments. On July 21, NVIDIA confirmed the platform had reached full production, with Ian Buck, the company’s vice president of accelerated computing, telling reporters at NVIDIA headquarters that systems are now shipping to customers including OpenAI, CoreWeave, Google Cloud, Microsoft Azure, Meta, and Dell. NVIDIA had announced at GTC Taipei on May 31 that the Vera Rubin platform was ramping into full production, with Taiwan’s top server makers and global supply chain leaders manufacturing Vera Rubin-based systems at scale.
What Vera Rubin actually is
The Rubin platform comprises six chips in total, the NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch, in addition to the Vera CPU and Rubin GPU.
On March 16, 2026, at GTC in San Jose, NVIDIA announced the platform had expanded to seven chips in full production, with the newly integrated NVIDIA Groq 3 LPU joining the existing six to scale the world’s largest AI factories.
Extreme co-design is the foundation of the Vera Rubin platform, the result of what the company calls “extreme co-design” across six types of chips: the Vera CPU, the Rubin GPU, the NVLink 6 Switch, the ConnectX-9 SuperNIC, the BlueField-4 DPU, and the Spectrum-6 Ethernet Switch. With extreme co-design, NVIDIA takes rack-scale integration to the next level, making the rack system a unit of compute, a single distributed accelerator.
Vera is the successor to NVIDIA’s Grace CPU, while Rubin succeeds Blackwell, with NVIDIA claiming Rubin will be capable of achieving 5x and 3.5x of inference and training performance, respectively, compared to Blackwell.
The Rubin platform harnesses extreme co-design across hardware and software to deliver up to a 10x reduction in inference token cost compared with the NVIDIA Blackwell platform.
Shipments begin, partners confirmed
NVIDIA Rubin is in full production, with Rubin-based products available from partners in the second half of 2026. Among the first cloud providers to deploy Vera Rubin-based instances will be AWS, Google Cloud, Microsoft and OCI, as well as NVIDIA Cloud Partners CoreWeave, Lambda, Nebius and Nscale.
Microsoft will deploy NVIDIA Vera Rubin NVL72 rack-scale systems as part of next-generation AI data centers, including future Fairwater AI superfactory sites. The Rubin platform will provide the foundation for Microsoft’s next-generation cloud AI capabilities.
On the systems side, Dell Technologies, HPE, Lenovo and Supermicro, together with ASUS, Foxconn, GIGABYTE, Pegatron, Quanta Cloud Technology, Wistron and Wiwynn, are adopting NVIDIA DSX to accelerate AI factory ramp with Vera Rubin. According to the official NVIDIA newsroom announcement from May 31, production shipments of Vera Rubin are set to begin starting this fall.
The rack-scale design behind the numbers
Vera Rubin is NVIDIA’s most extensive POD-scale platform, five purpose-built racks operating as one massive AI supercomputer for agentic workloads. The flagship NVL72 rack integrates 72 Rubin GPUs and 36 Vera CPUs connected by NVLink 6, along with ConnectX-9 SuperNICs and BlueField-4 DPUs, delivering up to 10x higher inference throughput per watt at one-tenth the cost per token versus Blackwell.
Each Rubin GPU carries 288 GB of HBM4 memory with up to 22 TB/s of bandwidth, compared with the Blackwell Ultra B300’s 8 TB/s memory bandwidth.
The Vera Rubin NVL72 rack will also be 100 percent liquid cooled and feature cable-free modular tray designs, which NVIDIA claims will allow installation times to be reduced from two hours to five minutes, in comparison to its Blackwell equivalent.
CoreWeave, one of the first cloud providers to receive the hardware, told Bloomberg that its NVL72 racks are delivering ten times the token output of the previous generation. NVIDIA’s vice president of accelerated computing, Ian Buck, told reporters at NVIDIA headquarters that systems are now shipping to customers.
Why the timing matters for the chip market
NVIDIA executed an aggressive 18-month development cycle from Blackwell’s launch to Rubin production, compressing what typically spans 24 to 30 months in semiconductor development.
That pace arrives against a broader surge in chip demand: the Semiconductor Industry Association reported global semiconductor sales of $120.6 billion during the month of May 2026, an increase of 9.2% compared to the April 2026 total of $110.5 billion and 104.1% more than the May 2025 total of $59.1 billion.
The Vera Rubin buildout is already reshaping how hyperscalers plan their infrastructure spending, a trend that is also driving competition in adjacent semiconductor markets. The same wave of AI inference demand that is filling Rubin-equipped racks has pushed enterprise AI inference providers, such as Fireworks AI, to raise $1.505 billion at a $17.5 billion valuation, while the speculative decoding techniques demonstrated on Blackwell have set the performance baseline that Rubin is now expected to surpass.
With AWS, Google Cloud, Microsoft and OCI all confirmed as early deployers, alongside NVIDIA Cloud Partners CoreWeave, Lambda, Nebius and Nscale, the Vera Rubin ramp is positioned as the hardware foundation for the next phase of large-scale AI model training and inference, moving the industry’s compute baseline well beyond what the current installed Blackwell fleet can deliver.
