Blackwell goes enterprise
- Nvidia and Google Cloud announced collaborations and infrastructure to extend Nvidia's Blackwell GPU ecosystem beyond hyperscale clouds. - Announcements included AI factories, Vera Rubin-powered instances, confidential Blackwell GPUs, and expanded GB300 access for customers. - Blackwell is being pushed into ordinary data centers and workstations to broaden Nvidia's reach from hyperscale to mainstream enterprise (nvidia.com; letsdatascience.com).
Nvidia is pushing Blackwell beyond giant cloud clusters and into the broader enterprise market through new Google Cloud infrastructure and on-premises systems. (nvidia.com) At Google Cloud Next on April 22, 2026, the companies said Google will offer A5X bare-metal instances built on Nvidia’s Vera Rubin NVL72 platform, with clusters that can scale to nearly 1 million Rubin graphics processing units, or GPUs. Google also said its A4X Max instances with Nvidia GB300 NVL72 are now shipping in production. (nvidia.com; cloud.google.com) Google and Nvidia also said they are bringing confidential virtual machines with Blackwell GPUs to Google Cloud and previewing Gemini on Google Distributed Cloud running on Blackwell and Blackwell Ultra systems. Google Distributed Cloud is Google’s on-premises and edge platform for customers that need to keep data closer to home. (nvidia.com; cloud.google.com) A GPU is the chip that does the heavy math for artificial intelligence, and hyperscalers such as Google, Microsoft and Amazon have been the biggest buyers. Nvidia’s new pitch is that the same Blackwell family can now cover giant cloud training clusters, private data centers and office workstations. (nvidia.com; nvidia.com) That expansion has been building for months. At Nvidia GTC in March 2026, Google Cloud said its G4 virtual machines powered by Nvidia RTX PRO 6000 Blackwell Server Edition were seeing strong demand, and it previewed fractional G4 instances that split those GPUs for smaller jobs. (cloud.google.com) Nvidia started this enterprise push earlier with the RTX PRO 6000 Blackwell Server Edition, which it described in March 2025 as its first Blackwell-powered data center GPU for both enterprise artificial intelligence and visual computing. Nvidia said the chip was built for “every industry,” not just the biggest cloud providers. (nvidia.com) The workstation side is moving in parallel. Nvidia’s RTX PRO 6000 Blackwell Max-Q Workstation Edition comes with 96 gigabytes of GPU memory, and Nvidia says it is designed to let developers and data scientists run larger AI and visualization workloads locally instead of relying entirely on cloud capacity. (nvidia.com) Server makers are filling in the middle of the market with denser Blackwell hardware. Nvidia partner materials say RTX PRO Servers can be configured with as many as eight RTX PRO 6000 Blackwell Server Edition GPUs in 2U, 4U and 6U systems, while Cisco said in March 2026 that it was adding the RTX PRO 4500 Blackwell Server Edition across its Unified Computing System and edge products. (tdsynnex.com; blogs.cisco.com) The result is a wider Blackwell ladder: GB300 for the largest Google Cloud training and reasoning systems, confidential Blackwell instances for regulated workloads, and RTX PRO parts for mainstream servers and deskside machines. Nvidia is no longer selling Blackwell only as hyperscale infrastructure; it is selling it as standard enterprise computing gear. (cloud.google.com; nvidia.com; nvidia.com)
