In May 2020, the first patient walked into Mann-Grandstaff VA Medical Center in Spokane and became part of an experiment that would go on to cost American taxpayers close to ten billion pounds and counting. The new electronic health record system, built on Cerner's platform after Oracle's 2022 acquisition, was meant to unify care across the Department of Veterans Affairs. Instead it produced clinician workarounds, patient safety incidents serious enough to trigger a formal pause, and a three-year "reset period" during which barely a handful of sites went live. This week, six years on, the department quietly filed a notice on the federal procurement site seeking to extend its contract with Oracle Health through 2031, with the actual scale of the increase redacted from public view. On the same day, the VA's own inspector general reported that a separate benefits-processing modernisation, contracted to Booz Allen Hamilton in 2021, had slipped from a planned 2026 completion to the second quarter of 2034, undermined by the absence of a credible programme schedule or lifecycle cost estimate.

Neither story is really about America. Both are about what happens when a state health system makes an enormous, multi-year bet on a single technology partner without building the governance architecture to catch failure before it compounds. That is a question the NHS cannot afford to treat as foreign.

The parallels are not abstract. Trusts across England are deep into their own Epic-dependent electronic patient record rollouts, with the workforce credentialing risk and vendor concentration this creates now a live concern for anyone managing digital transformation budgets. The Federated Data Platform, built on Palantir's infrastructure, represents a similarly consequential wager on a single vendor relationship persisting across a decade of political and financial pressure. And NHS England's absorption into the Department of Health and Social Care has reopened a genuine question about whether the oversight function that should catch slippage early is being strengthened or diluted in the process.

What the VA's experience demonstrates is how these programmes fail in practice. It is rarely a single catastrophic decision. It is a slow accumulation of deferred reckonings: a go-live that produces adverse outcomes gets treated as a local implementation problem rather than a systemic warning; a contract renegotiation in 2023 that changes the terms from one long commitment to five annual ones looks like accountability but functions as a way of avoiding the harder conversation about whether the underlying approach is sound; and a redacted budget line becomes the visible symptom of a deeper absence, the lack of any external mechanism forcing the department to justify continued spending against delivered outcomes rather than sunk cost.

The NHS has its own version of sunk cost logic already embedded in its digital strategy. Billions have been committed to platforms and vendor relationships that would be extraordinarily disruptive to unwind, which means the incentive structure for anyone overseeing these programmes tilts naturally towards extension and renegotiation rather than confrontation. Wes Streeting's productivity agenda depends on these systems delivering, which makes it politically as well as operationally difficult for anyone inside government to say clearly that a given rollout is not working.

None of this argues against ambition in health technology, and the case for interoperable data and modern clinical systems remains sound. What it argues for is a governance mechanism, independent of the department delivering the programme, with the authority to assess cost and outcome data against original commitments and to say so publicly before extension becomes the default response to failure. The VA's inspector general could name the absence of a reliable programme schedule only after years of drift. The NHS has the chance, while its own EPR and data platform commitments are still being built out rather than a decade deep, to build that scrutiny in from the start rather than discover its absence the way Washington just did.