AIANALYSISPODCAST
Jensen Huang spent years pitching NVIDIA's DGX Cloud Lepton platform as a potential 'Expedia for GPUs.' After acquiring Lepton AI and facing pushback from cloud partners, the product evolved into something more modest — but no less strategic.
Back in 2023, Jensen Huang stood on stage at a keynote and declared that NVIDIA would aggregate underlying GPU compute into a single front door — a one-stop shop that would compete directly with AWS, Azure, and GCP. The vision was essentially a GPU version of the App Store: NVIDIA, not the individual cloud providers, would sit between developers and the hardware.
At the time, the pitch landed with impact. NVIDIA was the hottest company in tech, and the neo-cloud providers that depended on its chips faced an uncomfortable choice. As the TBPN podcast recounts, if they joined DGX Cloud Lepton, they risked losing direct relationships with their customers. If they refused, NVIDIA could simply direct workloads toward competitors who had signed up.
"It really did feel like NVIDIA had some leverage," the podcast host noted. "Biggest company in the world very shortly after. Very aggressive business."
NVIDIA acquired Lepton AI in April 2025. SemiAnalysis reported the purchase price was rumored in the range of $300 million to $900 million. Jensen Huang was enthusiastic. At the announcement, he said: "NVIDIA DGX Cloud Lepton connects our network of global GPU cloud providers with AI developers. Together with our NCPs, we're building a planetary-scale AI factory."
The "planetary-scale AI factory" framing reflected the original ambition: a platform that would route customer demand, not just facilitate it.
Cloud providers noticed. According to The Information, neo-cloud executives viewed Lepton as a potential competitive threat. The concern was straightforward: if NVIDIA's platform became the entry point for AI developers, NVIDIA — not the individual cloud — would own the customer relationship.
Refusing to participate carried its own risk. Executives worried that NVIDIA could direct workloads toward cloud providers who had signed on, penalizing holdouts.
The tension did not escalate into a full rupture. Instead, the product evolved. Rather than layering NVIDIA between cloud providers and their customers, DGX Cloud Lepton became what NVIDIA calls a "unified AI platform" — essentially a marketplace and set of standards, one level above a pure Expedia-for-GPUs aggregator but still proprietary software handling GPU node groups, dev pods, batch jobs, inference endpoints, storage, observability, and reservations.
The product description now emphasizes that developers can bring their own compute capacity or purchase through the marketplace — not just go through NVIDIA's abstraction layer. The platform integrates NVIDIA NIM microservices, NeMo, and soon Blueprints and Cloud Functions, but it does not require routing through NVIDIA's own infrastructure.
NVIDIA Cloud Partners — including CoreWeave, Crusoe, Lambda, Nebius, Nscale, and others — appear on the platform as providers rather than as intermediaries.
The TBPN host called it a "quiet" pivot, noting that "NVIDIA is not going to take lightly to even the slightest whisper that the CUDA moat will dissolve anytime soon." DGX Cloud Lepton remains an active product, still evolving, with early access available and more provider integrations in progress.
The original vision — NVIDIA as the front door to AI compute, aggregating demand across a fragmented cloud market — did not materialize as described. The "planetary-scale AI factory" is real, but it operates more like a mall with multiple storefronts than a single NVIDIA-owned megastore.
What remains unclear is whether the current marketplace model represents a stable equilibrium or whether NVIDIA still harbors ambitions to increase its intermediation over time.