Residual

UK data. UK-owned compute. UK jurisdiction.

AI inference for organisations whose data can’t leave UK control. The company is UK-owned, the hardware will be in the UK, and the models are open-weight, so you know exactly what runs on your data.

Your data, the UK-owned site, the open-weight model and the results all sit inside UK jurisdiction. The US CLOUD Act applies to providers under US jurisdiction, so it has no route in through ownership. UK JURISDICTION YOUR DATA NHS, council, legal, financial UK-OWNED SITE HEAT → LOCAL NETWORK OPEN-WEIGHT MODEL RESULTS stay with you, in the UK US CLOUD ACT Reaches providers under US jurisdiction. No US parent here, so no route in. Your data, the UK-owned site, the open-weight model and the results all sit inside UK jurisdiction. The US CLOUD Act applies to providers under US jurisdiction, so it has no route in through ownership. US CLOUD ACT no US parent, no route in UK JURISDICTION YOUR DATA UK-OWNED SITE HEAT OPEN-WEIGHT MODEL RESULTS stay with you
Schematic. It shows ownership and jurisdiction, not network architecture.

Who it’s for

Organisations that need solid, capable models on sensitive data. None of them needs frontier-scale training.

  • NHS bodies

    Summarising clinical documents, coding and triage support on patient data that has to stay under UK control.

  • Councils

    Casework, social care records and resident correspondence.

  • Law firms

    Privileged documents, disclosure review and drafting support.

  • Financial services

    Client data, compliance review and internal knowledge search.

  • Defence suppliers

    Controlled technical data that can’t sit with a foreign-owned provider.

The CLOUD Act distinction

The US CLOUD Act (2018) lets US authorities require providers under US jurisdiction to hand over data they control, wherever it’s stored. A UK region of a US-owned cloud is still run by a US company.

We’re a UK-owned company, and our sites will be in the UK, so that route won’t apply to us.

This is how we see the position, not legal advice. Your own counsel should confirm it for your data and your contracts.

Refurbished, not second-rate

A100 and H100 GPUs coming off hyperscaler contracts are well-proven hardware. For inference on open-weight models, they are more than enough. This is how we plan to run them.

  • Burn-in before service

    Every card is stress-tested before it carries a customer workload. Cards that fail are replaced, not put into service.

  • Spare capacity

    Sites are sized with headroom, so a failed card doesn’t take a service down.

  • Monitoring and replacement

    We track the health of every card and replace ageing cards before they fail, from a stock of spares.

  • Right-sized workloads

    Open-weight models for inference and fine-tuning, not frontier-scale training runs that need the newest silicon.

Assurance, honestly

We’re early, and we don’t hold certifications yet. We’re designing sites to meet what public-sector and regulated buyers ask for, such as ISO/IEC 27001 and Cyber Essentials Plus. We’ll tell you exactly where we are at each stage.

If your sector has specific requirements, tell us now. They shape how we build the first sites.

Tell us what you need to run.

Describe the workload and how sensitive the data is. We'll tell you whether our first sites fit and when.

I need sovereign compute