The operation of Vera Rubin racks marks a major shift in the race to develop the computational capabilities required to train the world’s largest advanced intelligent models.
Article Index:
- Deploying the first batch of the Vera Rubin system
- Pricing levels and infrastructure profit margins
- Future expansion prospects and Nvidia earnings results
- Frequently Asked Questions
Deploying the first batch of the Vera Rubin system
Uday Rudaraju, head of infrastructure at OpenAI, announced this week the deployment and successful activation of the first batch of Nvidia Vera Rubin supercomputer racks within the company’s technical training environment. Rudaraju wrote on the X platform: «This represents a prominent and tangible milestone in expanding our massive computing capabilities, which will provide full support for the initial training phases of OpenAI’s next generation of frontier AI models.» Meanwhile, Nvidia’s official account reposted the announcement, emphasizing that both sides are working together jointly to build a high-speed, accelerated computing infrastructure to drive the development of leading intelligent models to unprecedented heights.
This announcement serves as a clear signal of the expanding deployment and operation of Nvidia’s next-generation AI platforms in global markets. Vera Rubin system racks are now operational and entering active service across a wide range of major cloud companies and infrastructure platforms, including CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. Microsoft announced in early August that it was the first hyperscale cloud provider to operate the Vera Rubin NVL72 system, with Microsoft CEO Satya Nadella stating on X that this step represents «a new qualitative leap toward building the next generation of AI infrastructure in collaboration with Nvidia».
Pricing levels and infrastructure profit margins
The price of a single Vera Rubin NVL72 system rack ranges between $5 million and $7 million, according to a report by Tom’s Hardware cited by Eugene Securities, surpassing the previous cost of Blackwell GB series racks, which were estimated at about $5 million per unit. CoreWeave executives confirmed that Vera Rubin pricing levels and resulting profit margins are reaching record highs, with profit margins on newly signed contracts exceeding previous rates by 5 to 10 percent.
Nvidia’s global supply chain network dedicated to this advanced system spans more than 350 manufacturing nodes across 30 countries, forming what the company describes as the largest rack-level AI infrastructure supply chain in the world to date. CoreWeave was the first cloud provider to successfully complete full deployment and performance validation of the Vera Rubin NVL72 system at the full-system level during the second quarter of 2026.
Future expansion prospects and Nvidia earnings results
Nvidia is increasingly moving toward delivering and supplying fully integrated systems directly to customers, as complete rack-level integration concentrates value-added and bargaining power with the company at the completed systems level. Looking further ahead, the company is currently studying multi-rack expansion configurations, including dual-rack, quad-rack, and eight-rack setups, which would expand unified GPU pools to 144, 288, or 576 unified GPUs, with expectations pointing to readiness for rollout in the second half of 2027.
These deployments come ahead of Nvidia’s quarterly financial earnings announcement scheduled for August 26, as investors and analysts await second-quarter results against the company’s guidance targeting $91 billion in revenue, as well as searching for precise indicators regarding the pace and acceleration of Vera Rubin platform production in global factories.
These developments highlight the fierce race among tech giants to build the world’s largest computing complexes, where supercomputing capabilities have become the decisive factor in determining the speed of innovation and access to more efficient and secure general artificial intelligence models.
Frequently Asked Questions
Question: What is the Nvidia Vera Rubin system adopted by OpenAI?
Answer: It is the new generation of supercomputer racks and accelerated graphics processing units designed to train the largest artificial intelligence models.
Question: How much does a single Vera Rubin NVL72 rack cost?
Answer: The integrated rack price ranges between $5 million and $7 million according to industry reports and financial analysis.
Question: What other cloud companies have started operating this system?
Answer: The list includes CoreWeave, Microsoft Azure, Google Cloud, and Oracle Cloud Infrastructure.
