Lepton was acquired by NVIDIA (now DGX Cloud Lepton). The information below may be outdated.

Lepton

GPU compute marketplace and unified AI development platform, formerly Lepton AI

Originally a serverless AI inference platform, Lepton AI was acquired by NVIDIA in April 2025 and relaunched as DGX Cloud Lepton. The platform now operates as a GPU compute marketplace connecting developers to capacity from 20+ cloud providers through a single interface. It supports interactive dev pods, batch training jobs, and inference endpoints with auto-scaling.

Pricing: Monthly subscriptions + usage based

Hosting Cloud

Pricing Usage Based

HQ 🇺🇸 United States

Founded 2023

License APACHE-2.0

GitHub 2,800 stars

Compliance SOC 2 · HIPAA

Visit website →

Pricing

GitHub

Posts

Suggest changes

Resources

Open Source Perplexity Clone with Lepton AI in 2 Min...

Yangqing Jia: Lepton AI

What is Lepton AI?

Lepton AI was a serverless inference platform that let developers run open-source AI models with a simple API. In April 2025, NVIDIA acquired Lepton AI and relaunched it as NVIDIA DGX Cloud Lepton in May 2025.

What is DGX Cloud Lepton?

DGX Cloud Lepton is a GPU compute marketplace that connects developers to capacity from 20+ cloud providers (CoreWeave, Lambda, Nebius, Crusoe, and others) through a single unified interface. NVIDIA describes it as "ridesharing for AI," aggregating available GPU resources across partners so developers can find and use compute without managing multiple cloud accounts.

Key Features

The platform supports three workload types: Dev Pods for interactive development with SSH and Jupyter access, Batch Jobs for distributed training, and Inference Endpoints for model deployment with auto-scaling via NVIDIA NIM microservices. Developers can switch between cloud providers without rearchitecting, and the platform supports data sovereignty by allowing workloads to run in specific regions.

Pricing

Pricing is marketplace-based. Individual cloud providers set their own rates for GPU capacity, with options for on-demand or reserved compute. There is no publicly documented free tier. NVIDIA offers up to $100,000 in credits to eligible VC portfolio companies through partner programs with Accel, Elaia, Partech, and Sofinnova Partners.

Who Should Use It

DGX Cloud Lepton is aimed at teams that need GPU compute across multiple providers without the overhead of managing separate cloud accounts. It is a different product from the original Lepton AI, which focused on serverless inference with a developer-friendly API and free tier.