Hadron.

The ARM of AI compute.

A patent-pending runtime that routes every AI workload to the cheapest hardware that can actually run it, across cloud, edge, on-prem, mobile and your user's own browser. Operators cut their compute bills. We take a fee on every workload routed. No data leaves their systems.

Only filed IP in cross-platform dispatch 5 UK patents 447 tests, zero failures Up to 45x faster 5B devices addressable

The problem

Companies have no alternative but to overpay for compute.

To run AI and heavy compute, operators pay the AWS or Google Cloud bill, or buy more servers, and watch it crush their margins. Hyperscalers spent over $450B on AI data centres in 2025, while roughly 5 billion capable GPUs sit idle in people's own phones and browsers, already paid for. The hardware to run things cheaply already exists. The intelligence to route to it did not.

How it works

Look. Pick. Run.

Hadron is a drop-in library inside your software. On every workload it reads the shape of the job (size, type, precision needs, never the data), scores the options, and runs it on the cheapest hardware that can actually do it, falling back safely if a target is unavailable.

Cloud and edge

Hyperscaler GPU, plus CDN and serverless edge.

Client browser

WebGPU, on five billion devices already in your users' hands.

Mobile and on-prem

iOS Metal, Android Vulkan, and enterprise CUDA / DirectCompute.

Specialised silicon

NPU, TPU and embedded accelerators. The cheapest capable target wins.

Proof

A working runtime, not a deck.

The engine is live in your browser today. The numbers below are measured, not projected.

447production tests, zero failures
45xpeak sort speedup vs standard JS
3.7xpeak search speedup vs CPU
£10M+modelled annual saving per Tier 1 customer

Speedups are peak figures and routing is configurable for cost or speed. The £10M+ saving is modelled from measured benchmarks, not yet realised in production.

Why now

The window is open, and it is timed.

WebGPU shipped across every major browser in Q4 2025, an open W3C standard, making five billion client devices programmable for GPU compute overnight. No commercial dispatch runtime has shipped, we are the only filed IP in the category, and the international patent priority window closes in March 2027. The technical unlock, the economic pain and the IP window line up now.

Why we win

Everyone else optimises one workload. We route across all of them.

ONNX Runtime Web

Static. Decided ahead of time, not at runtime.

TVM / WebLLM

Compile-time. Optimises a model, does not route a workload.

gpu.js and friends

Primitives. The building blocks, not the decision.

Hadron

The runtime routing layer across all of them, with deterministic scoring, GPU inhibition and safe CPU fallback.

The moat

Five patents on the dispatch layer.

Five UK patent applications, all assigned to Ayoob AI Ltd, filed with the UKIPO in 2026. The only filed IP in cross-platform compute dispatch.

GB2607734.7Platform GPU inhibition, the cross-platform routing core
GB2607044.1Precision-aware numerical dispatch
GB2607047.4Per-operator SQL dispatch
GB2607740.4Two-phase pattern matching
GB2606693.6Seven-factor sorting dispatch

Team

Built solo, advised by the best in the field.

Husain Ayoob, founder and sole inventor on all five patent families, built the entire runtime and live demo, so the core technical risk is already retired. He is advised by Prof. Graham Morgan (Newcastle University, 20-plus years in GPU compute, close to directors at NVIDIA) and Ian Wilkin (ex-ADIA, private equity).

See it run. Then let's talk.

Hadron is live and working today. Try the demo, or get in touch.

Ayoob AI Ltd · Newcastle upon Tyne · Companies House 15931832 · ayoob.ai