The Fenon Thesis
Infrastructure for the physical era of intelligence
We think the next era of intelligence is physical. And it'll run on shared infrastructure, not the same perception pipeline rebuilt a thousand times. Fenon is Vercel for Physical AI.
The physical era of intelligence has begun
For the last decade, intelligence lived on the internet. Models learned from text and pixels scraped off the web, and everything they powered lived on a screen. That era is maturing. The next one is physical. Robots that touch the world, cars that move through it, headsets that render it, systems that have to read space and motion and bodies the way language models read text.
The physical era runs on different fuel. Not tokens of text, but frames of the real world. Depth, pose, geometry, motion. And the thing holding it back isn't model architecture anymore. It's the infrastructure to turn raw sensor data into structured understanding, cheaply and at scale.
Everyone is rebuilding the same infrastructure
Look inside almost any physical AI team and you'll find the same thing getting built from scratch. Body pose estimation. Depth perception. 3D mesh reconstruction. Segmentation. These are the perception primitives of the physical world, and right now every team re-implements them, wires up its own GPU orchestration, and burns months of runway before shipping anything that touches the actual product.
It's the same duplicated work the web era eventually abstracted away. Nobody racks bare metal to serve a website anymore. Nobody should have to stand up a perception pipeline and a GPU fleet just to measure a runner's gait or reconstruct a room.
Vercel for Physical AI
Vercel took the messy parts of servers, builds, and scaling and turned it into one thing. Push your code, we handle the rest. We're doing the same for perception. Call an endpoint for pose, depth, or mesh, and the models, the GPU orchestration, the scaling all disappear behind the API.
Underneath, jobs run on fractional GPUs. A slice sized to the workload, spun up on demand and released the second the work is done. You get billed per second, for exactly the compute you used. No hourly reservations. No fleet sitting idle between bursts. The economics finally match how these workloads actually behave, which is spiky, uneven, and hungry for data.
Everyone becomes a data entrepreneur
The demand for diverse, high-quality physical data is exploding, and a handful of big labs aren't going to meet it. It gets met by a long tail of people and small teams who spot a niche - a sport, a workflow, a factory floor, a species - and build the dataset the world needs.
What's held them back is infrastructure. Perception at scale used to mean a research team and a GPU budget. We take that off the table. Give anyone the same pipelines and the same pay-per-second compute the labs have, and you get a new kind of builder. The data entrepreneur. That's who we're building for.
What we're building
Fenon is the infrastructure layer for the physical era of intelligence. Out-of-the-box APIs for the perception pipelines every team needs. Fractional GPUs so compute is sized to the job, not the contract. Per-second billing so you pay for exactly what you use. And the tooling to string all of it into data operations that scale with demand.
The physical world is about to get measured, modeled, and understood at a scale we've never seen. We're building the infrastructure that makes it affordable to everyone building it.