Nvidia steps back from DGX Cloud — stops trying to compete with AWS and Azure

News
By published

DGX Cloud fades into internal use as Nvidia positions Lepton to direct AI workloads across partner infrastructure.

Nvidia logo
(Image credit: Getty / NurPhoto)

Nvidia's DGX Cloud, once positioned as a direct-to-enterprise AI cloud, has quietly taken a backseat in the company's external strategy. According to an insider speaking to The Information, the GPU behemoth now uses most of DGX Cloud's capacity for internal research rather than promoting it as a customer-facing product.

It's a subtle change, but in Nvidia's financial results for the second quarter of its 2026 financial year, the company no longer attributes its multibillion-dollar cloud spend commitments to DGX Cloud, a disclosure it had included in prior quarters. This service is still listed in revenue categories, but its role has clearly shifted to in-house infrastructure. In other words, DGX is still alive and kicking, but it's no longer meant to compete head-on with the likes of Microsoft Azure or AWS.

Instead, Nvidia has turned its focus to Lepton, a GPU rental marketplace launched earlier this year. Unlike DGX Cloud — which involved Nvidia renting GPUs from neocloud players like CoreWeave and then subleasing them to customers — Lepton acts as a traffic controller. It routes workloads to partner providers, including AWS and Azure, which are set to join the marketplace despite their own GPU offerings. This makes Nvidia less of a rival and more of an aggregator in the cloud AI economy.

TOPICS
Luke James
Luke James
Contributor

Luke James is a freelance writer and journalist.  Although his background is in legal, he has a personal interest in all things tech, especially hardware and microelectronics, and anything regulatory. 

5 Comments Comment from the forums
  • Stomx
    "DGX Cloud launched in 2023 with a premium price tag of $36,999 per H100 instance per month"

    Wow! The overpriced H100 itself were at this price tag and they additionally increased it by an order of magnitude via lease if you use it for a year.

    Indeed, NVIDIA never deprives American customers in order to serve the rest of the world.
    Reply
  • bit_user
    Stomx said:
    "DGX Cloud launched in 2023 with a premium price tag of $36,999 per H100 instance per month"
    No, Nvidia said that's per-instance, not per-H100. Here's how they define an instance:
    "Each instance of DGX Cloud features eight NVIDIA H100 or A100 80GB Tensor Core GPUs for a total of 640GB of GPU memory per node."
    Stomx said:
    Wow! The overpriced H100 itself were at this price tag and they additionally increased it by an order of magnitude via lease if you use it for a year.
    A machine like the one above would've cost about $500k, IIRC. Plus, you obviously haven't looked at pricing for AWS GPU instances, which runs into the $M/year, if you used like a 4-GPU instance 24/7!
    Reply
  • Stomx
    But then after one year what -- the programmer goes to retirement? In 3-5 years this lease would cost you >10x the amount you would spent looking around for better prices or even better building yourself your own system. This is why nobody probably buying their service and they pushed it belly up
    Reply
  • bit_user
    Stomx said:
    in 2023 I was able to find A100 for $12k and H100 for 20k,
    You should read up on those DGX machines, because they're a lot more than simply 8x GPUs packed in a server chassis. The launch pricing was actually $480k:
    https://en.wikipedia.org/wiki/Nvidia_DGX#cite_ref-33
    However, there was also a long waiting list that quickly developed for them, and you know the saying: "time is money!" You can add a bit more for storage, power, cooling, infrastructure, etc. This is why AWS can get away with charging so much for their GPU instances!
    Reply
  • Stomx
    bit_user said:
    You should read up on those DGX machines, because they're a lot more than simply 8x GPUs packed in a server chassis. The launch pricing was actually $480k:
    We are still looking at them when they will crash 10-20x in price (used) which can be pretty soon.
    Reply