NVIDIA Cloud Functions · Omniverse Storage APIs

Complex worlds. A browser away.

One environment for streaming Omniverse apps: NVIDIA Cloud Functions, elastic GPUs and USD scenes read from your own bucket. Run it in your account on AWS, Azure, Google Cloud or on premises.

See the offer

The offer

From the control plane to the pixel in the browser.

We install and run NVIDIA's self-managed Cloud Functions stack with the Omniverse Storage APIs, so your team streams interactive USD scenes to any browser. Scenes, GPUs and streams stay in your account.

01 · FAST

A picture, not a download

Kit apps stream to the browser over WebRTC. No client install, and mouse and keyboard input go straight to the scene.

1920×1080 at about 60 fps in our lab
USD scenes read straight from your bucket
Relay over TLS for viewers behind corporate firewalls

02 · CONTROLLABLE

GPUs warm when people work

…and released when they leave. An operator activates a stream, the Console shows every session, and you set the GPU ceiling.

Activate a stream in one click, with an automatic stop
The GPU pool scales to zero every night
Scoped keys, single-use share links and an audit trail

03 · ANY CLOUD

The same stack, your account

One design on managed Kubernetes or on your own GPUs. Terraform and Helm set it up, with NVIDIA versions pinned.

AWS EKS and Azure AKS
Google Cloud GKE and on-prem k3s
Your bucket: S3, Blob, GCS or on-prem object storage

How it works

One stack. Four ways to run it.

Omniverse apps run as NVIDIA Cloud Functions on Kubernetes. The Control Console starts and stops the GPU pool, the Storage APIs serve USD from your bucket, and the stream reaches the browser.

streamcontrolstorageDrawn from our running lab design.

1080p60

Browser stream; mouse input reaches the Kit app.

Lab · AWS · 5 Nov 2026

0GPUs

The GPU pool scales to zero every night and starts again on demand.

Lab · stops on record

NoNucleus

A USD scene read from a private bucket through the Omniverse Storage APIs.

Lab · 6 Nov 2026

TLSrelay

The stream still reaches viewers behind a corporate firewall.

Lab · 6 Nov 2026

What you get

Every lever moves another. We know which.

Streaming, storage, GPUs and cost are tuned as one system, on a stack we built and run ourselves.

Control Console

Activate a function, watch live sessions, end a stream and read the audit trail, in one place for every environment.

Elastic GPU pool

Zero GPUs when nobody watches. Capacity follows sessions within a ceiling you set.

Omniverse Storage APIs

USD scenes are data in your own bucket. Kit opens them through NVIDIA's Storage and Discovery services, with no Nucleus server.

Secure by default

Allow-listed operator access, scoped deploy and stream keys, nothing secret in the browser.

Share links

Show a live scene to a guest with no account. Each link is for one viewer, lasts 5 to 60 minutes, works once and can be revoked.

Infrastructure as code

Terraform foundations per cloud, Helm for the NVIDIA stack, image digests locked. Every change is reviewed as a plan first.

Clouds

Fast and controllable wherever you run.

The same Console, the same functions and the same stream on each target. Pick the cloud you already use, or run on your own GPUs.

AWS

EKS · L40S GPUs

Streaming end to end in our lab, with nightly scale-to-zero.

Microsoft Azure

AKS · NVIDIA GPUs

Streaming end to end in our lab, with scenes from Blob storage.

Google Cloud

GKE · NVIDIA GPUs

Being brought up now on the same design.

On-prem

k3s · your GPUs

Being brought up now: k3s on a single L40S GPU server.

Start

See your scene in a browser.

We start with a two-week assessment and get from an empty account to a first streamed scene early in the pilot.