How Britain's housing ministry escaped Palantir, and why the NHS probably can't
An obscure refugee-matching system shows what exit from an embedded vendor looks like. The conditions that made it possible are largely absent from the £670 million of contracts that remain.
In September 2025, a team inside the Ministry of Housing, Communities and Local Government switched off a piece of software built on Palantir's Foundry platform and switched on something they had written themselves. The transition went live without ceremony. Local authority caseworkers signed into a new interface, the data flowed, the Ukrainian families being matched with British hosts kept being matched, and the contract with the US analytics firm was allowed to expire.
Seven months later, Coco Chan, a senior digital leader on the project, wrote a blog post describing what her team had done. It was careful, civil-service prose. It did not name Palantir. It noted that moving to the in-house model was "already saving MHCLG millions of pounds a year in running costs" and that the replacement, called Share, gave the department "control of the system data and code". The user feedback she quoted was the gentlest possible rebuke of the predecessor system. "It's easier to navigate than the previous system."
The previous system cost the British taxpayer roughly £10 million across two twelve-month contracts after Palantir's initial six months of free work. The replacement, according to MHCLG, costs a fraction of that to run. The replacement's source code sits on GitHub, open by default, where any other government or local authority can use it. The Homes for Ukraine scheme has resettled more than 157,000 refugees since March 2022, and continues to do so on different infrastructure.
This is what a working procurement exit looks like. It is also, when set against the broader picture of Palantir's position in the British state, the exception rather than the new rule.
The exit nobody quite mentioned
The story would have stayed obscure had Terence Eden, a former government technology advisor, not done the deduction in a blog post the BBC then picked up. As Eden noted, the British state publishes its contracts. The 2022 award to Palantir was on the public record. So was its 2023 extension. The cancellation, by contrast, was visible only by absence, in the kind of careful Whitehall language that announces a departure without naming the person who has left the room.
This caution is not accidental. Contracts end for many reasons, and government departments need to be able to work with the same suppliers again. The convention is that disagreements stay quiet. But the National Audit Office report on the scheme, published in October 2023, was less restrained. It recorded that the Government's chief commercial officer had written to Palantir to note "concern about the practice of offering services to public sector customers for a zero or nominal cost to gain a commercial foothold, contrary to the principles of public procurement which usually require open competition." That sentence, buried in an audit report, is the closest thing the British state has produced to a formal warning about Palantir's go-to-market strategy.
The House of Commons Public Accounts Committee returned to the same concern in early 2024, recommending that the department set out its commercial options for when the Palantir contract ended. The department did so. The result is the system MHCLG turned on in September 2025.
What makes the Homes for Ukraine case unusual is not that the civil service can build software. The Government Digital Service, founded in 2011 by Martha Lane Fox and Tom Loosemore, established that point years ago and was for a period treated as a model by governments from Canada to Singapore. What is unusual is that the conditions for exit were all present at once. The emergency that justified the original direct award had genuinely ended. The contract was small enough that transition costs did not overwhelm the savings. The replacement could be built by a modest multidisciplinary team using three years of accumulated user research. And the department, crucially, was both the system's customer and its commissioning authority, which meant the decision to switch did not require coordination across dozens of independent trusts or operational units.
Subtract any of those conditions and the picture changes.
Why emergencies become permanent
The legal mechanism that brought Palantir into the Homes for Ukraine scheme was Regulation 32(2)© of the Public Contracts Regulations 2015, which permits direct award without competition where "extreme urgency brought about by events unforeseeable" makes ordinary timelines impossible. It is a sensible provision. Refugees fleeing artillery do not wait for a competitive tender to conclude. The provision exists in similar form across most European procurement law and is closely related to the doctrine of necessity that underwrites emergency authorisations in other regulatory domains.
The closest structural parallel is not in technology at all. It is in pharmaceutical regulation. When a new drug receives an emergency use authorisation during a pandemic, regulators are doing exactly what procurement officials do under Regulation 32(2)©: suspending ordinary scrutiny because the cost of waiting exceeds the cost of imperfect information. The pharmaceutical regime, however, has spent decades developing the mechanisms that procurement law lacks. Emergency authorisations carry explicit sunset clauses. They include forced conversion to full authorisation or withdrawal once the emergency abates. The default state, in other words, is transition.
Public procurement has no equivalent default. A contract awarded under emergency provisions can be extended indefinitely so long as both parties wish to continue. The original justification (extreme urgency) need not be re-examined when the renewal is signed. The chief commercial officer's 2023 letter to Palantir was a complaint, not a constraint. The 2024 Procurement Act, which came into force in February 2025, addresses several adjacent issues, but does not impose mandatory transition timelines on contracts that originated under emergency provisions.
This is a regulatory design choice with predictable consequences. The New York City Comptroller's office reviewed its own emergency contracts and found that many had remained in place for years past the emergencies that produced them, often without competitive replacement, because nobody had built the obligation to plan for transition into the original award. The Comptroller's recommendation was to require, at the moment of emergency award, a written assessment of how and when the contract would be migrated to ordinary procurement. Britain has no equivalent rule.
Without that obligation, the economics of staying outweigh the economics of leaving. Switching costs (rebuilding the system, retraining users, managing the cutover) fall on the procuring department within a single budgetary cycle. The savings from a cheaper in-house alternative accrue to the same department over many years, often beyond the tenure of the officials making the decision. For a senior civil servant facing a choice between a known supplier whose system works and the risk of a build that might fail, continuing the contract is almost always the safer career move. Vendor capture is not, in most cases, the result of corruption. It is the result of incentives that point one way.
The intellectual property question
Emma Logan, deputy president of the BCS, the Chartered Institute for IT, told the BBC that "external specialists can bring experience, specialist skills, and the ability to put large teams in place quickly, which can be particularly important for urgent national programmes." She is right. Logan's defence of vendor use is the strongest available version of the case, and the Homes for Ukraine story does not contradict it. The original Palantir engagement put a working system in front of caseworkers in nine days. No in-house team could have done that, because no in-house team existed.
Palantir itself made a sharper point in its statement to the BBC. The MHCLG exit demonstrates, the company argued, that there is no real vendor lock-in. Customers can leave whenever they decide it is worth the trouble. As a logical claim this is unfalsifiable, but as a practical claim it depends on a condition that the Homes for Ukraine case happened to meet and the larger Palantir contracts do not. MHCLG's team could rebuild because they were rebuilding a relatively small case-management system. The intellectual property required to do so resided mostly in the team's own accumulated knowledge of the scheme and the users, plus the GOV.UK Design System that already existed as a public asset.
Compare the NHS Federated Data Platform, the much larger Palantir engagement awarded in November 2023 for £330 million over seven years. In an April 2026 Westminster Hall debate, Martin Wrigley, the Liberal Democrat MP for Newton Abbot, summarised what the contract actually delivers. "The current contract delivers a subscription service that leaves no deliverables after the subscription, no software, no improvements and no intellectual property after spending more than £330 million." All specially written software and intellectual property rights, he noted, belong to the supplier under the contract terms. All rights to any know-how are explicitly retained by Palantir and are not transferred on termination.
Triggering the break clause in 2027 would therefore not produce the conditions MHCLG used to exit. It would produce, instead, an empty room. The NHS would own its data, but no code, no integration work, no operational tooling. Whatever replaced the FDP would have to be built from scratch by whoever the NHS could persuade to take the contract, on a timeline measured in years rather than months, while the existing system was kept running under whatever transitional arrangement could be negotiated. Greater Manchester's Integrated Care Board, which serves 2.8 million people and is the only ICB in England to have refused to join the FDP, can avoid this trap because they never entered it. The other 41 ICBs cannot.
The MoD's £240 million Palantir agreement signed in December 2025, awarded without competitive tender for data analytics across strategic and live operational decision making, has the same structural feature. So does the Financial Conduct Authority's pilot, signed weeks after the government announced it wanted to reduce dependency on large technology vendors. Across 34 separate UK public sector contracts, by a count cited in Hansard, Palantir holds the code. The state holds, at most, the data the code reads.
A small victory in a much bigger contest
It is tempting to read the MHCLG story as the start of something. Eden's blog post, which captured the most attention, called it a "triumph for sovereign technology" and expressed hope that "where MHCLG leads, others will follow." That hope is reasonable. It is also, on the available evidence, premature.
Rob Miller of Public Digital, the consultancy founded by former GDS staff including Tom Loosemore, made the more careful observation in his comments to the BBC. The question is not just whether to reduce reliance on big technology vendors but "how quickly the government is willing to invest in the capability to do so". Building MHCLG's Share system required a multidisciplinary team holding scarce skills (product, delivery, user-centred design, business analysis, development) inside the civil service for long enough to deliver a substantial system. The market rate for those skills is significantly above civil service pay bands, and the wider trend in Whitehall remains the one that produced the original problem. Hire fewer permanent specialists, contract out for the rest, accept the vendor terms that come with the contracts.
This is the bind. The civil service can demonstrably build excellent software when adequately resourced and given political cover. It is rarely both at once. MHCLG had a small, specific brief, an existing user-research base, and a contract whose end was visible enough to plan around. The NHS, the MoD and the FCA have none of these. They have very large contracts with intellectual-property terms designed for a different relationship to the supplier than the one the British state would now prefer to have. The break clauses, where they exist, do not unlock the exit. They unlock the decision to start a rebuild that the contracting authority is not currently staffed to perform.
The honest reading of Coco Chan's blog post is therefore narrower than the headlines have suggested. MHCLG did not show that Palantir can be removed from the British state. They showed what the preconditions for removal look like when they are all present at once. Code ownership. Programme ownership. A team capable of inheriting both. An emergency that has ended cleanly enough to permit planning. The £10 million Homes for Ukraine contract had all four. The £670 million of remaining Palantir contracts, by the published tally, have none.
Whether the next four years see the lessons of MHCLG generalised across Whitehall, or stay confined to a single departmental success story, depends less on what Palantir does than on whether the British state is willing to fund and retain the kind of teams that built Share. The original GDS experiment showed it was possible. The years since have shown how easily that capability can be allowed to wither. The team that turned off the Foundry system in September 2025 is, in Chan's words, "sticking together to keep improving Share". Whether the system that produced them is sticking together for long enough to produce more of them is the question this story has not yet answered.