For NVIDIA, from 🤫 hussh
The silicon of the owner-operated AI factory.
You built the AI factory. Now it is moving to the edge, to inference, and to sovereign AI. We are building the version that lives in America's garages and warehouses: Jetson Thor at the edge, GB300 and Vera Rubin in the warehouse, NIM and Nemotron on every node, DGX Spark on the desk. Each one owned by the person who runs it. You make the compute people can own. We put it to work, owner by owner, by consent.
Your AI factory, owned by the people it serves.
AI factories are the largest infrastructure buildout in history, and you are taking them to the edge and to sovereign AI.
We run owner-operated AI factories in American garages and warehouses, on NVIDIA silicon, connected by Starlink.
A new deployment surface for edge and sovereign AI, growing ZIP by ZIP.
Jetson Thor, NIM and DGX Spark are your stack for physical, agentic and personal AI.
They are the bill of materials for 🤫 Puppy One and Agent One, running on hardware the person owns.
A channel that turns your edge products into a living grid.
Inference is the next volume driver, and Rubin CPX is built for it. Every watt counts.
Millions of private agents, on-device and at the edge, are inference demand you never have to host.
Distributed inference at the lowest cost per watt, on your silicon.
CUDA, the full-stack rack, and a yearly rhythm.
NVIDIA is a full-stack AI infrastructure company: silicon, systems, networking and software, sold as the whole AI factory. The numbers tell the story.
Up 65% year over year, with about $120B of net income. One of the most profitable franchises in the world.
$193.7B of FY2026 revenue came from Data Center. NVIDIA is the AI factory company.
Low-to-mid 70s. The mark of CUDA and the full stack, not a commodity chip.
Up 85% year over year. A record, with Data Center about 92% of the total.
The new Edge Computing platform: PCs, workstations, robotics, automotive, AI-RAN, agentic and physical AI.
Jensen Huang's view of the Blackwell and Vera Rubin revenue opportunity from 2025 through 2027.
Software, systems and pace. This is the ground we want to build on.
CUDA and the software moat
Two decades of libraries and the largest community of trained developers in the world. The reason the platform is the platform.
The full-stack rack
GPUs, NVLink for scale-up, Spectrum-X and InfiniBand for scale-out, BlueField DPUs. NVIDIA sells the whole AI factory.
A yearly cadence
Blackwell, Blackwell Ultra, Vera Rubin, Rubin Ultra. A new architecture every year, and the rhythm is itself an advantage.
The edge stack
Jetson Thor and IGX Thor bring Blackwell-class compute to the edge. RTX PRO brings it to the workstation. Physical AI, on site.
NIM and Nemotron
Self-hostable inference microservices and open models. An agent runs locally, on the person's own node, without renting a cloud.
Omniverse, Isaac and GR00T
Digital twins, robotics and world models. The physical AI toolchain for fleets that people own and operate.
Where NVIDIA is going.
Jensen Huang has been clear about the road ahead. Every stop on it is a place we want to meet you.
AI factories
Huang: "the buildout of AI factories, the largest infrastructure expansion in human history, is accelerating." The central idea.
The agentic inflection
Huang: "the agentic AI inflection point has arrived." Agents that use tools and drive compute demand with every task.
Inference at scale
Rubin CPX says it plainly: large-context inference, at the right cost, is the next wave of volume.
Physical AI and robotics
Huang: "every industrial company will become a robotics company." Jetson Thor, Isaac and GR00T, at the edge.
Sovereign AI
Nations building their own AI infrastructure as critical infrastructure. A named growth driver.
The edge
The Edge Computing platform takes the silicon that powers hyperscale down to the device and the site.
Vera Rubin, Rubin CPX and Jetson Thor.
The newest NVIDIA products, and the ones our grid is designed around.
Blackwell Ultra, GB300
Ramping in 2026. 288GB of HBM3e in a dual-reticle design. Today's warehouse-class rack for training and inference.
Vera Rubin NVL72
Shipping in the second half of 2026 to named clouds and OEMs. 72 Rubin GPUs and 36 Vera Arm CPUs in the next AI factory rack.
Rubin CPX
A GPU built for massive-context inference, expected at the end of 2026. The clearest sign that inference is the next volume driver.
Jetson Thor and DGX Spark
Jetson Thor puts Blackwell at the edge: shipping since November 2025 at $3,499, with 2,070 FP4 TFLOPS. DGX Spark and Station put a supercomputer on the desk.
Buy together. Build together. Sell together. Take risk together.
Every move starts from an NVIDIA product that is shipping or announced. Together they make the bill of materials for AI factories people own.
Build together: the grid's bill of materials
Jetson Thor and IGX Thor at the edge. GB300 and Vera Rubin NVL72 in the warehouse. NIM and Nemotron for inference.
We standardize 🤫 Puppy, Garage, Warehouse and Factory One on this stack, with NIM and Nemotron as the inference layer on every node.
NVIDIA gains a deployment surface run by owners. People run frontier AI on hardware they own.
Build together: the personal supercomputer
DGX Spark and Station put NVIDIA compute on the desk.
We make them the personal supercomputer at the heart of the grid: a capital asset, not a rental.
"Own your compute. Turn your personal supercomputer into a business asset." On NVIDIA.
Sell together: the edge and sovereign channel
NVIDIA is pushing hard into edge and sovereign AI.
Garage and warehouse owners become a distribution surface for Jetson, RTX and DGX, mapped ZIP by ZIP.
NVIDIA gains edge and sovereign volume. We gain a compute spine we can market together.
Buy together: volume and financing
NVIDIA runs the Inception program and DGX Cloud and financing rails.
Volume purchasing for the Fund A and Factory One rollout, with an NVIDIA-inside Puppy One as the one-tap default.
Committed volume for NVIDIA. Lower capex and a turnkey node for each owner.
Take risk together: pilot AI factories
Huang points to about $1T of Blackwell and Rubin opportunity through 2027.
We co-fund pilot AI factories on Vera Rubin in regions with cheap power and abundant sun. NVIDIA shares demand visibility; we carry the operator risk.
A new category of community-owned AI factory, built together.
Four steps, starting now.
Each step stands on its own and leads to the next.
- 01 · Build
Standardize the grid on NVIDIA
Jetson Thor edge nodes, GB300 and Vera Rubin racks, NIM and Nemotron inference. One bill of materials for every owner.
- 02 · Sell
Open the edge channel
Garage and warehouse owners become a real-world home for NVIDIA's edge and sovereign AI.
- 03 · Buy
A turnkey Puppy One
An NVIDIA-inside Puppy One as the default node, financed for the owner.
- 04 · Take risk
Pilot AI factories
Solar-powered, naturally cooled pilots on Vera Rubin, where cost per watt is lowest. Shared demand, shared upside.
The people building NVIDIA.
Jensen Huang
Founder, President and CEO
Sets the AI factory thesis and the yearly cadence. Took NVIDIA from GPU maker to full-stack platform, and now to the edge and physical AI.
Colette Kress
EVP and CFO
Allocates capital across the largest infrastructure buildout in history, turning demand into disciplined growth.
Michael Kagan
CTO (formerly Mellanox)
The networking and systems technology that makes the full-stack rack hard to match.
Jay Puri
EVP, Worldwide Field Operations
Global go-to-market across enterprise, cloud and sovereign customers.
Ian Buck
VP and GM, Accelerated Computing
One of the originators of CUDA. Leads the accelerated computing platform and the data center business behind the AI factory.
Bryan Catanzaro
VP, Applied Deep Learning Research
The models and software, NIM and Nemotron among them, that make the silicon useful at the edge.
Inference at the lowest cost per watt, for everyone.
Your edge and inference stack is exactly the layer 🤫 Agent One needs to run on hardware people own. Millions of private agents, on-device and at the edge, served on NVIDIA silicon across American communities, without ever touching a hyperscale data center. Own your AI, own your data, own your compute. And 🤫 Agent One is free to every American.