Sources cited by The Information indicate that NVIDIA is undergoing an internal organizational restructuring, with plans to scale back its public cloud division, DGX Cloud. Rather than being fully dismantled, the unit is set to be folded into the company’s core engineering organization under the leadership of Senior Vice President Dwight Diercks. Going forward, its primary mandate will shift from external leasing to internal research and development.
NVIDIA unveiled the DGX Cloud service with considerable fanfare in 2023. The original vision was to partner with cloud providers such as Oracle, Microsoft Azure, and Google Cloud to deploy NVIDIA-dedicated DGX server clusters within their data centers. Enterprise customers could then subscribe directly to NVIDIA for compute capacity—featuring top-tier GPUs such as the H100 and Blackwell—along with a complete AI software stack. The service was positioned as a solution to the chronic shortage of GPUs, enabling companies to rapidly access massive accelerated computing resources without stockpiling large quantities of hardware.
The latest restructuring signals NVIDIA’s intention to dial back DGX Cloud’s outward-facing commercial role. In the future, DGX Cloud will primarily serve NVIDIA’s own engineering teams, supporting internal chip development and testing, as well as the training of in-house AI models such as Isaac for robotics and Nemotron for language modeling. The prevailing view is that this shift is driven chiefly by a desire to eliminate conflicts of interest.
Although NVIDIA is often described as the “arms dealer” of the AI era, its largest customers are precisely the major cloud service providers—AWS, Microsoft, and Google. By operating DGX Cloud as a public cloud service, NVIDIA effectively placed itself in direct competition with its most important clients.
Scaling back DGX Cloud’s external operations not only helps repair and preserve relationships with these cloud providers, but also allows NVIDIA to concentrate its resources on areas where demand far exceeds supply: AI chip design, hardware platform delivery, and the continued reinforcement of its formidable CUDA software moat. In this sense, the move can be seen as a shrewd and calculated strategic retreat.
While Jensen Huang has long emphasized that NVIDIA is not merely a chipmaker but a “platform company,” the infrastructure-as-a-service market is already a fiercely contested arena dominated by entrenched players such as AWS and Azure. Having already reaped enormous profits from hardware sales, NVIDIA has little reason to raise its own cow just to drink milk—especially if doing so risks alienating the very distributors that sell that milk on its behalf.
By repurposing DGX Cloud for internal use, NVIDIA may in fact accelerate its own pace of innovation. After all, in the race to build more powerful next-generation chips and more intelligent autonomous-driving AI, NVIDIA itself is the ultimate hyperscale consumer of compute. Keeping the finest tools for internal mastery while entrusting cloud services to partners may well prove to be the most sustainable strategy for maintaining its dominance in the AI era.