Where Are They Now? NVIDIA DGX Cloud Lepton

TBPN，科技每日秀 Substack 专栏。关注

Happy Thursday.

We’re live on YouTube and 𝕏.

The current thing in tech and business is the drama on the timeline between Factory, Cognition, and Khosla Ventures.

Today’s Lineup

Varda Space Co-Founder & CEO Will Bruey at 11:45 AM
Ecolab Chairman, President, & CEO Christophe Beck at 12:00 PM
Tavus Co-Founder & CEO Hassaan Raza at 12:15 PM
Brodie Rec League Founder & CEO Connor Renton at 12:25 PM

Run of Show

Where Are They Now? NVIDIA DGX Cloud Lepton

Yesterday, our first two guests were from Nvidia and CoreWeave, the topics barely overlapped but the companies work very closely together. So closely in fact that at various points they have shown us glimpses of potentially becoming competitors, at least trying to renegotiate their economic relationship. Mike Intrator, the CEO, launched CoreWeave Forge, combining tools for serving models, observing production workloads, curating data, evaluation and redeployment. Reframing as a full AI cloud vs just rent-a-GPU.

It reminded me that just a few years ago, Nvidia was planning on scaling the opposite direction and go from just buy-a-GPU into a full AI cloud, abstracted on top of a wide range of data center companies and neoclouds. Nvidia DGX Cloud Lepton was essentially an abstraction layer over other GPU clouds. Lepton didn’t own the GPUs. Instead, you could rent capacity through Lepton from another provider and pay a small incremental software fee, or separately rent GPUs from a provider such as Nebius and then put Lepton’s management platform over them.

Nvidia acquired Lepton in April 2025 after spending a few years pitching a lighter version of the model under just “DGX Cloud”. SemiAnalysis reported the purchase price was rumored in the $300m to $900m range. Jensen was bold, he said: “NVIDIA DGX Cloud Lepton connects our network of global GPU cloud providers with AI developers. Together with our NCPs (Nvidia Cloud Partners), we’re building a planetary-scale AI factory.”

Some cloud providers were uneasy. The Information reported that neocloud execs viewed Lepton as a potential competitive threat. But refusing to participate created the risk that Nvidia could direct workloads toward competitors. It was a bit of a tense time, but the end result was more balanced. Instead of layering their customers and trying to control aggregated demand for compute, the product evolved into a “unified AI platform” (their words). It’s one level up from a set of standards and one level down from “Expedia for GPUs” - lots of proprietary software to handle GPU node groups, interactive dev pods, distributed batch jobs, inference endpoints, storage, observability, reservations, and more.

So today, a few years after Jensen declared “the iPhone moment of AI”, you can’t exactly go to an Nvidia website, enter a credit card and buy 64 B200s from the cheapest available provider, but the product is still evolving and shows that Nvidia is not going to take lightly to even the slightest whisper that the CUDA moat will dissolve anytime soon.

Clip Spotlight: MongoDB interim CEO Dev Ittycheria on Meta poaching CJ Desai:

"CJ made a decision, and I'll let you pass judgment."

"What I will say is I got a note from a good friend who says, 'At the end of the day, no one's going to remember how much money you make. They're going to remember what kind of leader you were and what kind of person you were.'"

"I'll just leave it at that."