Andy Burnham has weeks to decide whether Palantir keeps running NHS data. The case against renewal is no longer about who Palantir is. It is about whether the software works, and whether anyone ever tested the market.
When Andy Burnham was mayor of Greater Manchester, his was the only region in England that refused Palantir access to its data.1 He is now Prime Minister, and the first significant decision of his premiership is whether the same company continues to run the data platform of the NHS.
The £330 million Federated Data Platform contract reaches the end of its first term in March 2027. To continue past that point, the Department of Health and Social Care has to actively choose to extend it. Do nothing and it lapses.2 That sounds like a decision for next spring. It is not. Three Whitehall sources told WIRED that a replacement provider or an in-house build needs lead time, and that any decision to end the contract has to be taken by the autumn or the changeover gets botched.1
It is the end of September.
There is a moral case against this contract, and it is not a weak one. It is simply not the case that has been winning. The people who have spent three years making it in Britain have worked that out, and one of them put it plainly to WIRED: "The initial campaign was focused around the company's ethical issues, but we found that it didn't have the success we were hoping for. By focusing on competence we've found a lot more success, and by targeting the contract break."1
That is not a retreat, and it should not be read as one. The ethics were never the weak part of the argument. They were the part that ministers had learned to absorb — nod, acknowledge the concerns, commission a review, sign anyway. What ministers have not learned to absorb is a permanent secretary being asked, in public, why a £330 million system cannot produce evidence that it works, and why nobody ever tested the market.
So the case here is built on competence and procurement, because those are the arguments that bite. The moral objection sits where it belongs: not as the reason to end the contract, but as a description of what renewing it endorses.
01 What Is Actually on the Table
Palantir holds roughly £670 million of UK public sector work across health, defence and policing.1 Four contracts matter, and they have had strikingly different fates.
The pressure on the NHS contract is not fringe. In June 2026 the cross-party Science, Innovation and Technology Committee singled Palantir out as the technology provider it found most concerning and called for the contract to end, warning that the UK risked becoming dependent on a single foreign firm for critical health infrastructure.3 The Health and Social Care Committee followed in July, telling ministers they could invoke the break clause if data security concerns persisted.4 The British Medical Association, representing over 200,000 doctors, had already called in February for a complete break from Palantir technologies in the NHS.5
An official petition asking for patients to be able to opt out of the platform passed 10,000 signatures; the government replied that it would consider the matter in future policy making. An unofficial petition to remove Palantir from the NHS entirely has passed 177,000.1
02 The Question Nobody in Government Wants Asked
Set aside every ethical objection for a moment. Assume Palantir is exactly the partner it says it is. The first question a government should ask before renewing a £330 million contract is whether the thing does what it was bought to do.
NHS England says it does. As of June 2026 it reported that 139 trusts were using the Federated Data Platform, and that 137 of them were seeing benefits.6
The campaign group Foxglove obtained the trust-level numbers under Freedom of Information and found something the aggregate had concealed. Of the 41 trusts using the FDP's inpatient scheduling module — the flagship product, the one meant to get more operations done — 13 were carrying out fewer operations than before they adopted it. Those 13 trusts recorded 9,073 fewer procedures after adoption than before.7
- 41 trusts using the inpatient scheduling module; 13 of them — roughly 30% — performing fewer operations than before adoption.
- 9,073 fewer procedures across those 13 trusts.
- Earlier FOI responses put active use of the platform at around 34 trusts, under 15% of the total; more than half of those who replied had not adopted it at all.
- NHS England had published only aggregate figures showing an overall increase, withholding the trust-level breakdown that would have revealed the variation.
This is not proof the software causes fewer operations. Foxglove does not claim it, and neither do we: there is no control group, no comparison against trusts that did not adopt it, and a hundred confounders in any hospital's theatre list. That is precisely the problem. Four years and £330 million in, nobody can say whether the platform works, because the evaluation that would answer the question was never built into the contract.
It is worth pausing on what 9,073 is a count of. Not records, not rows, not a dashboard metric trending the wrong way. Operations. Hip replacements, cataract removals, biopsies — procedures that people on a waiting list did not get, at trusts that had bought a system sold to them on the promise of getting more of them done. Whether the platform caused that or merely failed to prevent it, the department cannot say. It bought the system anyway, and it is being asked to buy it again.
The government appears to have noticed. The Office for Statistics Regulation is now assessing how ministers have used data to justify the platform.8 NHS England has separately paid Imperial College Projects £700,000 to evaluate whether the FDP is meeting its objectives and delivering value for money.9
Commissioning a £700,000 study to find out whether a £330 million system works is not due diligence. It is due diligence performed four years late, with the renewal decision already on the table and the answer not due in time to inform it. If the platform were delivering what was claimed, the trust-level data would have been published as a matter of pride rather than extracted by campaigners under FOI.
03 Palantir Loses Wherever Someone Runs a Proper Procurement
Here is the pattern that emerges once you line the four contracts up. It has nothing to do with ideology and everything to do with process.
Read it in one direction and it looks like a story about who has political cover. Read it the other way and it is simpler still: where a procurement authority applied ordinary market-testing discipline, Palantir lost the work. Where that discipline was bypassed, the contract stands.
That is not an argument about American spy tech or Silicon Valley ideology. It is an argument any Treasury official would recognise. A supplier that wins on merit does not need the exemption. The consistent use of non-competitive routes is itself the evidence.
Palantir's response to losing the one contract that was properly tested is instructive. It has formally notified MOPAC that it intends to challenge the decision in court.13 A company whose British pitch is partnership with the public sector is suing a police oversight body for declining to waive a competition.
04 The Company You Would Be Renewing With
An honest account has to concede something that the critics of two years ago got wrong. Palantir's business has grown into a serious share of its valuation. In Q2 2026 it reported revenue of $1.94 billion, up 93% year on year, and net income of $1.06 billion. It guides to $8.15 billion for the full year, with adjusted free cash flow of $4.5 to $4.7 billion.1415 Anyone who argued in 2024 that the revenue would never arrive has been answered.
What has not changed is the gap between that business and the price. At roughly $187 a share and a market capitalisation around $455 billion in late September 2026, the stock trades at about 50 times trailing sales — against a ten-year median for the company of 26.1617 Michael Burry — the investor who shorted the subprime mortgage market in 2007 — renewed his short position on 2 September, holding put options and arguing that the fundamental value is well under $50 a share and that the market capitalisation could eventually fall below $100 billion.18
Insiders have been steady sellers throughout. Between 2023 and 2025 Palantir insiders sold roughly $3.2 billion in shares, with Karp accounting for an estimated $2 billion-plus. In one six-month window through mid-2025, insiders made 244 market trades: 243 sells, one buy. In November 2025, Karp filed to sell 585,000 shares worth around $96 million days after record earnings and an all-time high; President Stephen Cohen filed for 405,000 shares in the same window.1920
None of this is illegal, and executives sell shares for many reasons. It matters here for one narrow, practical purpose: a department signing a decade-shaping infrastructure dependency should understand that it is contracting with a company whose own leadership is systematically reducing its exposure to it, at a valuation Michael Burry puts at close to four times what the business is worth — roughly $187 a share against his estimate of well under $50.18
05 The Part Ministers Have Learned to Absorb
Everything above is the argument that works on a permanent secretary. This is the part that ought to work on everyone else, and the reason the BMA's objection is not sentimentality but a clinical one: a health service runs on patients telling the truth to clinicians, and trust is the mechanism by which that happens.
None of this makes the scheduling module work better or worse. That is the point. These two arguments have been running in parallel for three years and only one of them has ever moved a minister, which tells you something uncomfortable about how procurement decisions actually get made in Britain — and precisely nothing about which argument is true.
A government that renews on the evidence currently available is not saying the ethics do not matter. It is saying something narrower and worse: that it did not check whether the system worked, did not test the market, and considered neither omission sufficient reason to pause.
06 The Channel That Keeps Working
If the procurement channel has started to fail Palantir, the political one has not. The reinforcement has been conspicuous.
On 17 September 2026, nine days before Labour's conference and with the NHS decision pending, Bloomberg reported that Palantir had appointed Tom Watson — former deputy leader of the Labour Party, and an adviser to the company since 2024 — as its UK Senior Vice President.24 At that conference, the Conference Arrangements Committee refused to accept an emergency motion calling on DHSC to trigger the break clause. Members did not get to debate it.25
Watson is the most recent hire, not the first. In November 2025, the month before the MoD contract was signed, Palantir hired Barnaby Kistruck — the MoD's outgoing director of industrial strategy, who had helped draft the UK's Strategic Defence Review. openDemocracy reported that he joined nine days after leaving public office, the company's fourth hire from the UK defence establishment that year. Nine days is not a cooling-off period. It is a handover.26
Inside government, WIRED reports that Varun Chandra — the official chiefly responsible for the relationship between Westminster and Big Tech — has been kept in post by Burnham. Chandra came from Hakluyt, the corporate intelligence firm turned venture investor, and, according to the same reporting, dined with Palantir co-founder Joe Lonsdale in December 2024.1 The Financial Times has separately reported that the Department of Health's permanent secretary had worked for Carnall Farrar, a member of Palantir's bidding consortium, while it competed for the NHS platform contract, and had to recuse herself.1
Palantir's UK chief executive, Louis Mosley, has been blunt about the stakes, telling critics — in remarks reported by WIRED — that they put politics before performance, and offering the line that "If NATO went to war, it would go to war using Palantir."1 It is worth taking seriously rather than dismissing. It is also, precisely, an argument for why a single supplier holding defence, health and policing data simultaneously is a concentration risk rather than a convenience.
That concern is not confined to campaigners. Germany's Federal Constitutional Court ruled in 2023 that Palantir-based automated police data analysis was unconstitutional, nullifying the authorising law.27 The Swiss Army recommended its forces consider alternatives following an investigation into vendor lock-in and data sovereignty. And in December 2025, the incoming chief of MI6 used her first public speech to warn that national security is being reshaped by corporations acquiring reach previously reserved for states.28 She named no company. The MoD signed the £240 million no-bid contract the same month.
07 What Leaving Actually Costs
The standard objection to ending the NHS contract is that there is no way out: the data is in there, the integrations are built, and unpicking it would cost more than carrying on. It is the argument every incumbent makes, and in this case there is a British counter-example that settles it.
Homes for Ukraine was exactly the same shape as the NHS deal in miniature. Palantir arrived free, moved fast — a working platform in nine days — and became the system of record. The price then rose past £10 million. MHCLG decided the flexibility and the cost were both wrong, rebuilt the service with its own staff, and reports saving millions a year in running costs.11
Exit is not free, and anyone who tells you otherwise is selling something too. It costs a rebuild. What Homes for Ukraine demonstrates is that the rebuild is affordable, that a department can own the code and the data afterwards, and that the running costs fall rather than rise. The lock-in is real but it is not absolute — it is a bill, and the bill has been paid once already by a UK department with far less engineering capacity than NHS England.
Which leaves the decision in front of the Prime Minister looking less like a moral referendum on an American company and more like an ordinary question of public administration, with an unusually well-documented answer.
There is no evidence the platform is delivering what was claimed for it, and the government has commissioned a study that will not report in time to say otherwise. There is clear evidence that when the contract was market-tested, at the Met, it did not survive the test. There is a worked example of a department leaving and saving money. And there is a supplier response — a lawsuit against a police oversight body, and a senior political hire nine days before conference — that tells you which channel the company believes actually decides these things.
What makes this worth being angry about is not that an American company sells software to governments. It is that the ordinary disciplines — test the market, measure the outcome, publish the numbers, keep the option to leave — were suspended one at a time, by people who each had a defensible reason, until a single foreign supplier held defence, health and policing data simultaneously and the only remaining question was whether anyone still had the authority to say no. None of that required a conspiracy. It required a series of officials finding it easier to sign than to compete.
There is a version of this argument that overreaches, and it is worth naming in order to avoid it. The problem is not that Palantir is American, and the answer is not to buy British on principle. The answer is duller and more durable than that: a department that can specify what it needs, evaluate what it is offered, and build the thing itself if the offers are poor will get better terms from every supplier it ever deals with — Palantir included. MHCLG has that capability now, and it acquired it by leaving. What the Federated Data Platform has cost, beyond the £330 million, is four years in which NHS England did not.
We build software for a living, so discount that accordingly. But note which way the interest runs: the departments that ended up least dependent are the ones that stopped buying capability and started holding it.
Burnham refused Palantir access to Greater Manchester's data when he had the power to do so and the decision cost him nothing. He now has the same decision at national scale, a select committee mandate, an autumn deadline, and a break clause that requires him to do nothing at all in order to take effect.
The interesting question is no longer whether Palantir should hold this contract. It is whether a British government can still decline to renew one.