Ada: British AI, answered on British soil.
Ada.
British AI, answered on British soil.
The problem
Where does British AI actually run?
Most AI used in the UK is trained and served on infrastructure outside UK jurisdiction.
List hubs and companies
- Seattle · Microsoft, Amazon
- Bay Area · OpenAI, Anthropic, Google, Meta, Nvidia
- Northern Virginia data centres · AWS, Microsoft Azure, Google Cloud
- Dublin data centres · Microsoft, Google, Meta, AWS
- Paris · Mistral AI, Scaleway, OVHcloud
- Frankfurt data centres · AWS, Microsoft Azure, Google Cloud
- Hangzhou · DeepSeek, Alibaba Cloud
Showing today: requests from the UK travel to overseas AI and cloud hubs and back.
A prompt typed in the UK can be processed thousands of miles away — on hardware, and under laws, the UK does not control.
Locations are approximate.
What Ada is
Ada is an AI built, run and governed in Britain. Ask it anything and your prompt never leaves the UK: the models, the hardware and the rules are all British.
Runs in Britain
Every layer, from the grid to the model, on UK soil.
Open and accountable
Open-weight models, open standards and independent audit.
Built for Britain
Adapted on British data, for public services, research and business.
The stack
Sovereignty is a stack, not a slogan.
Every layer of Ada — from energy to policy — is British.
Showing Policy
Policy
UK law, open standards and clear rules for who may take part.
- Accreditation criteria
- Audit and rollback
- Open standards
Follow a request
One prompt. Never leaves the UK.
- 01You
You send a prompt
Everyone uses one public endpoint. It looks like any AI service — but it's hosted in the UK.
- Endpoint
- One UK address
- Feels like
- Any AI service
- 02Gateway
It lands at a UK gateway
The gateway checks the request and keeps it onshore. There is no fallback to an overseas provider.
- Hosted
- In the UK
- Checks
- Identity, limits, safety
- Overseas fallback
- None
- 03Router
A router picks the model
A small, fast classifier reads the prompt's length and complexity and chooses the right-sized model — no bigger than it needs.
- Reads
- Length, complexity, reasoning
- Chooses
- The right-sized model
- 04Registry
The registry picks a node
A live registry of accredited nodes finds the nearest one with that model loaded and room to run it.
- Tracks
- Status, capacity, location
- Only
- Accredited nodes
- Picks
- Nearest with room
- 05UK node
A British node answers
The model runs on UK hardware. Training carries on underneath, so spare capacity is never wasted.
- Hardware
- UK-hosted GPUs
- Model
- Latest verified version
- Meanwhile
- Training continues
- 06You
The answer comes home
The response returns to you and the request is recorded for audit. At no point did it leave the UK.
- Endpoint
- One UK address
- Feels like
- Any AI service
Roadmap
From framework to national capacity.
We start in Westminster and build outward — until sovereign compute covers the whole country.
Map showing Framework: London.
Framework
Define what 'sovereign' means for AI compute, end to end.
- Sovereignty criteria
- Reference architecture
- Open consultation
Swipe the map or tap a phase.
Define what 'sovereign' means for AI compute, end to end.
- Sovereignty criteria
- Reference architecture
- Open consultation
Get in touch
Be among the first to use Ada.
Register your interest and we'll be in touch when you can try it. Investors, researchers and compute partners are welcome too.