Argyll launches UK sovereign AI inference cloud with SambaNova
TL;DR:
- Argyll Data Development has launched a sovereign AI inference cloud in the UK, built with SambaNova, for organisations that want AI workloads kept under UK jurisdiction.
- The platform is aimed at defence, healthcare and finance customers and is positioned around sovereignty rather than headline compute scale, with hardware drawing about 10kW per rack so it can be deployed in existing UK data centres.
- Resultsense view: this is the first UK-headquartered launch this year that explicitly defines sovereignty as control over infrastructure, models and operations — not just data residency — which is the definition the financial regulators have been moving towards in their own communications.
The launch comes as more UK organisations move AI projects out of pilot stage and into production, and as scrutiny grows over where data sits, who controls the infrastructure and what it costs to run inference at scale. Argyll’s offer is a UK-resident environment for inference — the stage at which trained models process live requests — rather than for model training.
What the platform does
The service uses SambaNova’s Reconfigurable Data Unit architecture with SambaManaged software. The companies said the system runs at about 10kW per rack, which is well below the power density of dense GPU clusters and means it can be installed in existing UK data centre estates without major cooling upgrades. Argyll said the service hosts open-source models including Minimax and can run at up to 400 tokens per second in a UK-resident environment.
The platform is built as a disaggregated system, with compute, storage and networking able to operate across multiple UK sites while functioning as a single inference layer. Argyll said that design is intended to give regulated customers resilience without locking them into a single physical location.
The sovereignty argument
Peter Griffiths, chairman of Argyll Data Development, framed sovereignty as something that has to be demonstrated rather than claimed. “Sovereignty in AI is not a label you can apply to a contract or a colocation agreement,” he said. “It is a condition that has to be demonstrated: who is accountable, where the infrastructure sits, who controls the intelligence layer, and whether all of that aligns with the expectations of the society being served.”
That is a notably stricter definition than the “data stays in the UK” framing many hyperscaler regional offerings use. SambaNova’s EMEA managing director Jude Sheeran positioned the partnership as an alternative to default GPU-based AI infrastructure, citing long-term cost, energy and operational complexity.
UK relevance
For UK-regulated buyers — FCA and PRA-supervised firms in particular — the operational question is whether they can audit the platform against their own regulatory expectations and demonstrate that the legal jurisdiction applying to the infrastructure aligns with their home supervisor. Argyll’s pitch is that its model satisfies those conditions because infrastructure, models and operations all sit in UK jurisdiction.
The launch lands the same week opinion writers and policy commentators have been arguing publicly that UK AI ambitions are constrained by infrastructure the UK does not control. Argyll has previously outlined plans for renewable-powered digital infrastructure in Scotland, including its Killellan AI Growth Zone in Argyll.
Looking forward
Procurement teams in UK financial services, defence and the NHS now have a concrete domestic option to test against hyperscaler regional offers. The deciding factor will be whether Argyll’s price-performance, model availability and audit posture stack up against an Azure UK South or AWS London setup once real RFPs land.