The UK government says it is already saving millions by replacing Palantir software used in the Homes for Ukraine scheme with an internal platform. The bigger story is what happens when emergency technology stops being temporary.
Palantir helped the UK move fast when speed mattered most. Now the Ministry of Housing, Communities and Local Government says it has moved the refugee system in-house, cut running costs by millions of pounds a year and gained more control over the data, code, security and future shape of the service.
That is a serious moment for governments and the technology companies that sell to them. The Homes for Ukraine system was not a side project or a vanity dashboard. It supported a national refugee programme created after Russia's full-scale invasion in February 2022, connecting people fleeing war with UK hosts who had space to offer. It needed to combine visa data, sponsor details, local authority checks and safeguarding records at speed.
According to the BBC, Palantir says it stood up the original solution in just nine days and helped enable the safe resettlement of more than 157,000 refugees. That matters. In an emergency, the perfect procurement process is not always available, and a specialist vendor that can deliver under pressure may be the difference between chaos and a working public service.
But the second half of the story is just as important. Palantir initially built the system for free, using its Foundry platform. After that, the Department for Levelling Up, Housing and Communities, MHCLG's predecessor department, awarded 12-month contracts worth £4.5 million and £5.5 million. The National Audit Office later noted that officials had considered alternatives but worried that moving away could be costly, complicated and risky for a live safeguarding system.
Free pilots are attractive because they solve a real problem at the exact moment an organisation is under pressure. They also create a commercial foothold. Once a system is live, staff are trained, data is flowing and operational risk is attached to switching, the buyer's bargaining position changes.
This is not only a Palantir issue. It is the familiar enterprise software playbook, sharpened for public services and now increasingly wrapped in the language of AI and data infrastructure. A vendor offers speed, expertise and a ready-made platform. The buyer gets an immediate result. The longer the system remains central to operations, the harder it becomes to ask whether the same job could be done more cheaply another way.
The government was warned about that risk. The NAO reported in 2023 that the government's chief commercial officer had raised concerns with Palantir about offering zero-cost or nominal-cost services to public sector customers in order to gain a commercial foothold. Palantir has argued that government guidance itself supports pilots and trials, and there is a fair point there. A pilot is not automatically a trap. But a pilot without a credible exit plan can become one very quickly.
The new Share Homes for Ukraine Data system shows what an exit can look like once the crisis phase has passed. GOV.UK guidance was updated on 15 September 2025 to reflect the new Share system, which councils use to access sponsor and guest data, record accommodation and safeguarding checks, and keep information updated for central government. In other words, the replacement is not theoretical. It is already in the machinery of the service.
Sovereign technology has to earn its name
The phrase sovereign technology can sound abstract, but here it has a practical meaning. MHCLG wanted more control over the system's data and code, lower support costs and more flexibility to adapt the platform as the Homes for Ukraine scheme matured. Those are not ideological points. They are operating requirements.
There is also a bigger question for AI infrastructure. Palantir is often discussed through politics, defence work and public concern over surveillance, but its business case rests on something simpler: organisations have messy data, urgent decisions and limited internal capacity. Foundry promises to make that complexity usable. In emergency government work, that promise is powerful.
The lesson from Homes for Ukraine is not that governments should build everything themselves. That would be naive. External specialists can bring scale, experience and speed that a department may not have ready on day one, especially in a national crisis. The lesson is that public bodies need to know which systems should remain vendor-led and which ones should be rebuilt internally once the emergency stabilises.
That distinction matters for startups as much as for governments. Many infrastructure companies are selling into healthcare, defence, local government and financial regulation with the same pitch: let us solve the hard data problem quickly, then expand from there. The model works because the first deployment often creates proof, trust and dependency at the same time.
For buyers, the practical takeaway is simple. If a free or discounted pilot is accepted, the exit route should be designed before the first dataset is loaded. That means portability, documentation, skills transfer, clear ownership of code and a realistic view of what it would cost to move away. Otherwise the cheapest offer can become the most expensive path.
Palantir can still point to the speed and scale of what it delivered for Ukrainians arriving in the UK. MHCLG can now point to a cheaper internal system that it controls. Both things can be true. What comes next is the part every government and every AI vendor should watch: whether public institutions treat this as an exception, or as a template for taking back control once the emergency is over.
Also read: Mira Murati is making AI collaboration the product • The U.S. and China are moving AI safety into power politics • Meta's Louisiana AI campus puts a price on public incentives
