For a decade the web talked to the GPU through WebGL — a graphics API bent, awkwardly, into doing everything. WebGPU is the reset: a modern interface that exposes the GPU as what it actually is today, a massively parallel compute machine, not just a triangle painter.
Beyond WebGL
The headline feature is compute shaders — the ability to run arbitrary parallel programs on the GPU straight from JavaScript. Graphics is now just one thing you can do with all that silicon.
Local AI in the browser
This is why on-device inference suddenly works in a tab. A quantised language model that would crawl on the CPU runs interactively on the GPU — no install, no server round-trip, no data leaving the machine.
What you can build
- Private, in-browser LLMs and speech models
- Real-time 3D and simulation without a plugin
- Video and image pipelines that stay on-device
- Data tools that crunch millions of rows locally
The browser stopped being a document viewer a long time ago. WebGPU is where it becomes a runtime.
The catch is that great GPU code is still hard, and capabilities vary across devices. But the ceiling just rose dramatically — and the best web apps of the next few years will be the ones that reach for it.