The Short Answer
Sovereign AI is determined by who holds the source code and the data at contract end, not by where the supplier is incorporated. ibl.ai is the agentic AI platform where you own all the code and the data: you self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere, from your own cloud to a fully air-gapped network.
The UK is currently running the clearest test of that distinction anywhere in the world, and the results are genuinely mixed.
Its Sovereign AI Unit is real money on real procurement, which is more than most national AI strategies achieve. Its contract terms also leave the intellectual property with the supplier — which means the state is buying access, not ownership, even when it buys British.
What has the UK actually committed to sovereign AI?
Real budget attached to a live procurement, which distinguishes it from the strategy-document phase most governments are still in.
The Sovereign AI Unit launched in April 2026 with £500 million, run by the Department for Science, Innovation & Technology (DSIT). It is structured to behave like a venture fund inside government: it takes positions in UK AI companies rather than only writing grants.
Balderton partner James Wise chairs it, with Josephine Kant leading ventures.
Alongside the investment mandate, DSIT opened an £80 million procurement window, with individual contracts of up to £5 million running 12 to 24 months, across scientific discovery, health and social care, national security and defence, cybersecurity, transport, energy and net zero, and public service delivery.
Companies were asked to register interest by 16 May 2026, with competitions following from July.
The stated logic is that government acts as an early customer to validate capability and de-risk private investment behind it. That is a coherent industrial policy, and the UK deserves credit for executing it rather than publishing about it.
Who is actually winning UK public-sector AI contracts?
Mostly the same US firms that win everywhere else — and the numbers are not close.
According to the Tussell AI Procurement Tracker, UK public bodies awarded £1.41 billion across 453 AI-related contracts between January 2026 and 3 August 2026, already exceeding 2025's full-year total of £1.18 billion. Procurement is accelerating sharply.
The distribution is concentrated. Since 2018, the largest recipients are Microsoft at £1.03 billion, Palantir at £628 million, and KPMG at £320 million.
Cumulatively, Tussell counts 2,129 AI-related contracts worth £5 billion since 2018 — which is still only about 4% of total UK public-sector IT services spending.
The supplier-nationality picture is closer than the headline names suggest: since 2018, UK-based firms have won £2.19 billion against £2.28 billion for US-based firms, a gap narrowed to roughly £80 million by August 2026.
So the £80 million Sovereign AI procurement window is real, and it sits next to £1.41 billion of conventional AI procurement flowing largely to incumbents in a single year.
Both things are true at once, and reporting only the first one is how sovereign-AI coverage usually goes wrong.
Does buying from a domestic supplier make a deployment sovereign?
No — and the UK's own contract terms are the clearest available demonstration of why.
Under the Sovereign AI competition, successful bidders retain ownership of all background IP and all foreground IP created during the project. They may exploit it commercially or sell it to other customers.
Government retains usage rights to the foreground IP and has stated it will not seek to capture further economic value from it.
Read that as an industrial policy and it is defensible, even smart: the state validates a capability, absorbs early risk, and deliberately leaves the commercial upside with the company so that a domestic AI sector actually forms.
Traditional government contracting often claws IP back and kills the company's ability to sell the same thing twice. This is a considered departure from that.
Read it as a sovereignty claim and it does not hold. When the project ends, the government has a licence to use something someone else owns.
It does not have the source code as an asset, it cannot fork the system, it cannot hand the codebase to a different supplier, and it cannot audit what it does not possess.
That is the definitional split worth being precise about:
- Supplier sovereignty — the vendor is domestically incorporated and domestically taxed. This is what most national AI strategies actually purchase.
- Technical sovereignty — the buyer holds the source code, the model weights and the data, and can run, modify, audit and re-host the system without the vendor's participation.
A domestically-owned supplier operating a black box is a supply-chain improvement. It is not control.
And the asymmetry shows up exactly when it matters most: if the supplier is acquired, changes its pricing, deprecates the product, or simply fails, a usage right is not a continuity plan.
What would technical sovereignty actually require in a contract?
Four clauses, and none of them are exotic.
Source code under a perpetual licence, delivered. Not escrow contingent on bankruptcy — escrow protects against vendor death, not against a roadmap change or a price rise. Delivered source, runnable by the buyer's own engineers.
Model weights and the right to change models. A deployment pinned to one provider's hosted model inherits that provider's jurisdiction, retention policy and deprecation schedule regardless of who wrote the wrapper around it.
Data resident and exportable in an open format. Sovereignty over data that can only be read through the vendor's API is a subscription, not custody.
The ability to run with no outbound connectivity. This is the honest test of every other clause. A system that cannot run air-gapped has a dependency somewhere, and the procurement should surface where.
Applied to the UK programme, the first clause is the one its own terms decline. The others are procurement choices still open on every contract that follows.
Is the UK's approach wrong?
Not wrong — incomplete, and worth separating into the two goals it is trying to serve at once.
As industrial policy it is working as designed. Leaving IP with suppliers is how you get companies rather than contractors. A government that claws back every asset produces vendors who cannot scale, and the UK is right that £500 million of patient anchor-customer demand is more useful to a domestic sector than the same money as grants.
As sovereignty policy it stops short. The programme secures where the company is registered. It does not secure whether the state can operate the technology without that company. Those are different guarantees, and only one of them survives an acquisition.
The two are not in conflict, which is the useful part.
A contract can leave commercial exploitation rights entirely with the supplier — letting them sell the same capability worldwide — while still delivering the buyer a perpetual, runnable, auditable copy of what was built for public money.
Open-source-style dual arrangements do this routinely. It is a drafting choice, not an economic trade-off.
For any government reading the UK programme as a template, that is the amendment worth making: keep the industrial policy, and add the delivery clause.
Where ibl.ai fits
ibl.ai is the agentic AI platform where you own all the code and the data. The full source ships under a perpetual licence and runs inside your own authorization boundary, so the deployment survives any change in the vendor relationship — including the end of it.
It is model-agnostic across any LLM, with weights running locally where a workload requires no external egress, and carries no per-seat pricing, so cost tracks usage rather than headcount.
Deploy anywhere: your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity.
ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.
1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.
Related: Why Government AI Must Be Sovereign: EU, Kenya, Taiwan — the same ownership test applied to three other national programmes.
Related: Sovereign or Supervised: Government AI Architecture
Related: Cohere Alternative: Sovereign AI
Related: Why Government Agencies Cannot Afford to Rent Their AI Infrastructure
Related: The Government AI Deployment Gap and Sovereign Infrastructure