WebGPU
WebGPU is a new programming interface for browsers that gives web pages direct access to a computer's graphics unit — enabling computationally intensive tasks such as 3D graphics or AI models to run natively in the browser for the first time. It is considered the successor to WebGL and has been available in the first browsers since 2023.
A modern computer has two very different computing units. The CPU, the main processor, handles general tasks one after another — fast, but sequential. The GPU, the graphics unit, is built differently: it has thousands of small computing cores and can process many tasks simultaneously. For a long time, web pages could barely make use of this GPU performance. WebGPU is an interface — that is, a clearly defined connection between software and hardware — that for the first time gives browsers real, direct access to the GPU. It was developed by the World Wide Web Consortium (W3C), the web’s most important standards body, and was first enabled in Chrome in 2023.
Why WebGL, its predecessor, reached its limits
Until now, browser graphics relied mainly on WebGL, an interface dating back to 2011. WebGL was originally designed for 3D graphics and is based on an older standard called OpenGL. By modern standards, it has become too limited. It gives developers little control over how the GPU works internally — and that control is exactly what’s needed for efficient AI computations.
WebGPU instead takes its cues from newer, more powerful interfaces such as Vulkan, Metal, and Direct3D 12, which have been available to native applications — that is, ordinary programs outside the browser — for years. WebGPU brings these capabilities to the web for the first time. This is not a minor update, but a fundamental rebuild.
How WebGPU communicates with the graphics unit
When a web page uses WebGPU, the developer writes small programs called shaders. Shaders don’t run on the CPU, but directly on the GPU’s thousands of cores. WebGPU provides its own lean programming language for this purpose: WGSL (WebGPU Shading Language). These programs are translated by the browser and passed on to the GPU — as efficiently as in a native desktop application.
The concept of pipelines is also important. A pipeline is a fixed sequence of computation steps that the GPU works through. The developer defines this pipeline once, and the GPU then executes it very quickly — without the detour via the main processor. This direct path is the crucial difference from WebGL, which requires much more communication between CPU and GPU and is therefore slower.
A common misconception: WebGPU is not a pure graphics tool. It is equally well suited for general-purpose computations on the GPU — referred to in English as “General Purpose GPU Computing” or GPGPU. This makes it possible, for example, to run AI models without the user having to install anything.
WebGPU in browsers, AI tools, and games
The best-known application is running AI models directly in the browser. Google's MediaPipe project uses WebGPU to run face recognition and hand tracking in real time in the browser — without any data needing to be sent to a server. The Transformers.js project, a JavaScript library, even makes it possible to run language models directly in the browser once WebGPU is available.
WebGPU is also a major step forward for 3D graphics in games and simulations. Engines like Babylon.js and Three.js have already built in WebGPU support. Complex 3D scenes that used to be possible only in native games can increasingly be rendered directly in the browser window.
In financial and tech reporting, WebGPU comes up mainly in connection with “on-device AI” — that is, AI that runs on the user’s own device rather than in the cloud. This conserves server capacity, saves costs, and improves privacy. For businesses, this means that AI features can be integrated into web applications without operating expensive computing infrastructure.