While Nvidia’s splashy move into PCs grabbed headlines, the company’s real profit engine — the data center — just got its next chapter. Nvidia’s Vera Rubin platform is rolling out to the world’s largest cloud providers, and the list of early adopters reads like a who’s-who of the industry.
Who’s deploying it
Among the first cloud operators to stand up Vera Rubin-based instances in 2026 are Amazon Web Services, Google Cloud, Microsoft and Oracle Cloud Infrastructure. Microsoft says it will deploy Nvidia’s Vera Rubin NVL72 rack-scale systems as part of its next generation of AI data centers, built to squeeze more efficiency and performance out of both training and inference workloads.
The NVL72 designation matters: these are not single accelerators but full racks engineered to act as one giant unit, the kind of dense, liquid-cooled hardware that modern frontier-model training now demands.
Why it signals acceleration, not a slowdown
For months, skeptics have argued that AI infrastructure spending must eventually cool. The Vera Rubin commitments suggest the opposite — that the hyperscalers are doubling down. When AWS, Google, Microsoft and Oracle all queue up for the same next-generation systems in the same year, it tells you demand for compute is still outrunning supply.
Nvidia frames Rubin as the successor generation to the Blackwell systems that powered the last training wave. Each cycle has delivered large jumps in performance-per-watt, and efficiency is now the constraint that matters: data centers are bumping against power and cooling limits, so getting more useful compute from each megawatt is the whole game.
The competitive backdrop
Nvidia’s dominance is not going unchallenged. AMD continues to push its Instinct accelerators, and the hyperscalers are all designing in-house silicon — Google’s TPUs, Amazon’s Trainium, Microsoft’s Maia — to reduce dependence on a single supplier. Yet the fact that those same companies are simultaneously buying Vera Rubin at scale shows how hard Nvidia’s combination of hardware and its CUDA software stack remains to displace.
The bottom line
The PC announcement may define how consumers eventually meet Nvidia, but Vera Rubin is where the money is. With every major cloud lining up to deploy it, the 2026 AI build-out looks less like a bubble deflating and more like an industry pouring the next layer of concrete. For Nvidia’s rivals, the uncomfortable truth is that even as they build alternatives, they are still writing Nvidia very large checks.
Photo: Takuya Oikawa / BY-SA via flickr