At first glance, it seems pretty simple: NVIDIA’s recently launched DGX Cloud Lepton is a marketplace that rents GPUs to developers, chips sourced from NVIDIA itself along with its network of cloud vendors. As NVIDIA CEO Jensen Huang explained in mid-May when he announced the new cloud venture, “we’re building a planetary-scale AI factory.”
Underneath that grand but straightforward sales pitch is a competitive strategy that challenges, at some level, a large segment of the cloud AI vendor community. In past years, NVIDIA simply sold AI chips and left the cloud to other vendors. Now with the new DGX Cloud Lepton adding to DGX Cloud that NVIDIA launched two years ago, the company offers products very similar to both the top hyperscalers and smaller cloud-based players like Coreweave and Nebius.
Coreweave, for instance, is known as an “AI hyperscaler,” so obtaining and offering access to NVIDIA GPUs is core to its offering. With NVIDIA’s DGX Cloud Lepton renting access to its GPUs, buyers have yet another source in addition to Coreweave. Adding a layer of complication, NVIDIA has invested in Coreweave, so it is both a supporter and a competitor of sorts.
NVIDIA, for its part, has said that it does not compete with its partners, and that DGX Cloud Lepton enables developers to access GPUs from many providers. Still, in a marketplace driven by supply and demand, a greater supply of the sometimes scarce NVIDIA chips could drive down prices.
None of this is nefarious, of course. Examples of businesses that both compete and cooperate are numerous. AWS rents its extensive cloud network to Netflix for video streaming, yet the company also offers Amazon Prime Video in direct competition with Netflix.
But NVIDIA’s cloud-based initiatives that both compete and cooperate are particularly complex. The cloud service that the chip giant debuted in 2023, DGX Cloud, rents GPU servers to large enterprise customers. It built this cloud service on top of the cloud platforms of the top hyperscalers, AWS, Google and Microsoft, all of which supply access to NVIDIA chips as part of their offering.
These three leading cloud vendors are NVIDIA’s top customers; AWS reportedly spent more than $20 billion on NVIDIA chips in 2024; Microsoft was NVIDIA’s biggest customer, spending approximately $30 billion last year. So when NVIDIA debuted DGX Cloud, it was renting a platform from these three top cloud vendors, then using that platform to launch a product that competed with them. AWS initially was, understandably, reluctant to support the DGX Cloud, but later decided to offer cloud hosting.
But the competition goes both ways. The top hyperscalers are developing their own chips to compete with NVIDIA’s. AWS offers its Trainium series chips and Microsoft has introduced the Maia 100 Accelerator and the Cobalt 100 CPU to serve AI workloads. Google’s Tensor Processing Units (TPUs) are designed to support machine learning, and the company touts them as a competitor to NVIDIA’s GPUs, particularly in large-scale AI training.
“NVIDIA has been working to create competition for their large customers Microsoft, Amazon and Google so they are not reliant on them in the future,” Gil Luria, Head of Technology Research for D.A. Davidson, told Techstrong.ai. “At the same time those companies are also working on alternatives to NVIDIA, especially by developing and expanding the use of their own custom chips.”
A key question posed by NVIDIA’s move to broaden its competitive efforts: will the company’s cloud initiatives grow enough to actually threaten the top cloud vendors? This seemed to be less of an issue for the hyperscalers when DGX cloud launched in 2023. The gold rush of AI had started but they had yet to build up their AI tools as they have now.
Since NVIDIA doesn’t disclose DGX Cloud’s revenue, the size of the platform is unknown. In the company’s recent fiscal year, it announced $10.9 billion in cloud service contracts, a dramatic improvement from the $3.5 billion the previous year. But that’s still small compared with Amazon’s $107 billion in cloud business in 2024.
The more recently launched DGX Cloud Lepton rents directly to developer customers, instead of offering services via the large hyperscalers. That certainly sounds like an aggressive move to build closer relationships with customers of all sizes, including developers at large enterprise. One element that the company’s cloud has in its favor: NVIDIA provides not just AI software and chips to its cloud customers, but also AI experts, and such scarce professionals have become today’s hottest commodity, even rarer than NVIDIA semiconductors.
