Not a straight yes. Two claims hold up on the published evidence. Two do not, and an argument resting on either would not survive review.
The proposition that holds is about capital rather than outcomes, and about structures rather than ownership class. It concedes a great deal.
Private capital — capital in the broad sense, not private equity alone — is the only material source of new capacity in UK adult social care, and will remain so under any realistic fiscal scenario. The policy question is therefore not whether private capital participates, but on what terms, and by whom.
We tested the second half of that proposition — that the terms which matter are financial structure, not the private-equity label itself — against the data. It did not hold. What follows is what the evidence supports.
Public provision has fallen from 42% to 9% of adult social care in a single generation. The figure comes from outsourcing’s most prominent academic critics, which makes it hard to contest. It describes the starting position rather than a proposal.
Four in five care homes are more than twenty years old. Nearly two in five were converted from something else. Three in ten beds have no en-suite bathroom at all, and seven in ten have no full wet room. This is the physical stock the sector actually runs on.
Over the past decade UK care bed supply grew 2.4%. The over-65 population grew 16.2%. In the most recent year, total supply rose by 136 beds. Deregistrations of small, old homes now substantially offset everything new that opens.
Knight Frank puts capacity exhaustion at 2033 and the shortfall at roughly 200,000 beds by 2050. For en-suite provision specifically, the market is already in deficit. This is the arithmetic the argument rests on.
Each is well documented. Leaving them out invites the objection instead of answering it.
US Medicare data covering more than seven million patients associates private-equity ownership with roughly 10–11% higher short-term mortality and 19% higher spending. English CQC data associates private-equity-backed chains with a 6.6 percentage-point higher probability of a Requires Improvement or Inadequate rating.
The five largest private-equity-backed providers have borrowed £35,072 against every bed. Servicing that debt consumes 16% of the average weekly residential fee. Southern Cross remains the reference case.
Across the 26 largest providers, £117m of debt service — 45% of the total — is paid to related companies. In three English regions alone, over £250m in profit was taken across three years.
Ownership chains routinely run through jurisdictions that defeat scrutiny. That is not defensible. Nor is the pattern in how sponsors talk about themselves: at least six independent PE, REIT and infrastructure sponsors publish detailed ESG reporting for the exact UK care holding in question — carbon, workforce, diversity metrics — with zero mentions of CQC ratings or safeguarding, in several cases in the same document that covers care quality in detail for a different portfolio company. One disclosure mechanism, recurring six times.
This is the most useful finding in the literature and it is rarely cited. In the largest English study, private-equity-backed chains and independent single-site for-profit homes perform the same. Both are worse than not-for-profit. Non-PE for-profit chains do best of the three.
If a small independent home and a private-equity chain score the same, the problem is not the ownership label. Regulate the financial structure, and the argument becomes empirical rather than tribal.
Two warnings attach to this. First, it does not rescue private equity: for-profit of every kind still underperforms not-for-profit. Second, it is uncomfortable for the small-provider constituency, because independent for-profit homes rate worst of all. Both halves have to be stated together.
This is a fact about balance sheets rather than about ownership, which is the point. It also sets the terms for the recommendation.
These five are named, not anonymous — and by 2026 the ownership behind them has moved. Two have been sold to the same buyer, a US-listed healthcare REIT, within twelve months of each other; a third no longer exists as an operating group. At most one of the five is confidently still private-equity-owned. The balance-sheet fact stands; the ownership label attached to it does not (R16).
The buyer has not stopped at two. The UK Competition and Markets Authority’s 7 May 2026 Phase 1 decision covers four completed Welltower acquisitions in the same window — 649 care homes across HC-One (279), Care UK (home count not disclosed), Barchester (275) and Aria/Asprey plus Danforth Care (95, one combined transaction) — and finds competition concerns in 30 local areas. On Welltower’s RIDEA-structured homes specifically, the regulator found Welltower holds approval rights over its operators’ business plans and budgets, funds their capital expenditure, and can terminate them without cause: a landlord shaping a care home’s competitive and quality offering without ever being its operator of record (R19).
The one of the five still confidently private-equity-owned, Orchard Care Homes, is also the one where real leverage data now exists: its parent, Cortina Race LLP, discloses in its own Companies House group accounts a turnover of £63.9m against interest payable of £3.0m — interest cover of roughly 0.6x on trading profit — and a related-party loan charging 10% interest. This is not the propco or the fund three steps further up, but it is real, audited, and free (R21).
Every active English care home on the CQC register, joined to Companies House on the company number CQC itself publishes for 91% of them. 13,638 rated homes, 6,812 providers, 437,514 registered beds. Logit on a poor rating (Requires Improvement or Inadequate), local-authority fixed effects, clustered errors, and case mix controlled — a control the published literature did not have.
Charge counts count charges, not amounts, and they sit on the registered-provider company. The leverage the critics describe sits in the propco and the offshore parent, which those proxies cannot see. The Companies House PSC snapshot narrows that gap by one layer — who controls the registered provider directly. A disclosed corporate or legal-person controller predicts a lower chance of a poor rating once scale, age and case mix are held constant, and that term is stable across every specification we tried. It is not separable from scale, though: corporate-PSC prevalence rises from 49% of single-site providers to 78% of chains of fifty-one or more, so it most plausibly reflects compliance capacity in larger corporate groups rather than group structure improving care.
The two non-disclosure categories are weaker than they first appear, and we have split them out rather than combining them. Homes whose provider files a PSC statement instead of naming a controller do rate better, but 76% of them are charities, which are largely exempt from PSC disclosure and rate better anyway. Combined into one “opaque” dummy the effect looked strong; decomposed, it is weaker and partly re-measures sector rather than opacity. Read it as a caution about the measure, not a finding about ownership.
Genuinely offshore corporate controllers are in the data but far too rare to test: 173 of the 6,793 rated homes whose provider has a corporate PSC, and only 35 of 3,128 such companies. Those counts exclude filings where the country field holds a misspelling of England, a postal address or “not specified”, all of which a literal match reads as offshore. The snapshot also sees only one layer up: for most homes that is a UK holding company, not the fund three steps further up.
Everything above is cross-sectional and associational, so no causal claim is available. CQC ratings are a regulatory judgement made on varying inspection dates rather than a contemporaneous measure of care.
Companies House PSC filings, walked upward past the single layer the free bulk snapshot can see, cover all 7,221 active provider companies on the register (486,020 beds). 7.7% of providers (7.7% of beds) remain unresolved — mostly small, single-home operators with no PSC trail to follow. The rest resolve to a named investor type: family or founder, charity, mutual, employee ownership trust, REIT, private equity, or pension and infrastructure fund.
The raw, unadjusted gap looks dramatic: homes under private-equity ownership have a 10.3% poor-rating rate against 19.6% for family-or-founder homes. It does not survive controlling for scale and case mix (adjusted OR 0.78, p=0.25, not significant). Private-equity-owned chains skew larger and more established, and this taxonomy's own earlier finding already established that scale, not ownership, is what predicts quality — this is that finding surviving a much sharper ownership category, not a new one.
Mutual (OR 0.62, p<0.001), charity/third-sector (OR 0.77, p=0.008), REIT (OR 0.69, p=0.003) and local authority (OR 0.60, p=0.017) are the categories that remain significant, and all four point toward better outcomes. The REIT and local-authority results are new to a 5 September 2026 pipeline revision that reclassified 873 previously-“unresolved” providers using CQC’s own registration fields (a disclosed charity number, its own Individual/Partnership ownership-type field, a local-authority name pattern) — REIT crosses from borderline (p=0.056) to significant, and local authority goes from too imprecisely estimated to interpret to a fourth significant, protective category (91 councils correctly reclassified out of “unresolved”). Every other category — pension/infrastructure funds, employee ownership trusts, and providers whose ownership still could not be resolved — shows no detectable effect either way.
This method resolves who controls the regulated care-providing entity, not who owns the underlying real estate. A REIT can be landlord to a fully independent operator without ever appearing here — confirmed directly during this build, where one of the largest 2025 REIT deals in the sector turned out to be a sale-and-leaseback of the buildings, with the operating company's own ownership unchanged. Real-estate-only financialisation is therefore undercounted by this taxonomy, not overcounted.
All 14,892 of them. The block that lights up is every home currently rated Requires Improvement or Inadequate — 2,372 homes, 16% of the estate. That is the scale of the quality problem.
5,936 care-home registrations ended between 2019 and 2025. The CQC data independently confirms Knight Frank on what leaves: the median home on exit had 21 beds against 28 across the live estate, and 47.5% of exits were under twenty beds against 38.8% of homes still operating. Small, old stock is leaving, and where it is replaced it is replaced by something larger.
A change of registered provider can deregister a home and re-register it. The bed total is gross churn; read as capacity lost it would overstate the case substantially.
The denominator can only be built from providers still registered, so a single-site operator that closed its only home vanishes from the data entirely. That deflates the single-site exit rate by an unknown amount. It needs a provider cohort fixed at a past date from the CQC archive series.
We built a 29-point quarterly panel from CQC’s own historical archive (2019–2026) to test it directly. Result: zero locations changed registered provider across the whole 6.75-year panel — not a resolution problem, a structural one. Share-purchase acquisitions, which is how this sector’s PE/REIT deals are overwhelmingly done, never touch CQC’s registered provider at all. Verified against Danforth Care: its provider entity pre-dates Welltower’s 2025 acquisition of it by nearly two years. This is exactly why ownership-chain data (R17), not CQC registration data, was the only route to this paper’s central analysis (R22).
The original contribution is step one. Everything else depends on replacing the binary with a structural taxonomy.
Every figure here traces to a citable source except the following.
The pilot can separate for-profit from third sector, and can measure scale, chain size, company age and secured borrowing. It cannot identify private-equity ownership at the location/provider level: CQC publishes no such field and no current public list exists. **Largely addressed**: R16 hand-traces the 26 largest named operators with individual citations, and R17 extends the same structural classification (multi-level shell ladders, named sponsors, not keyword-matching) to the full register — 6,666 of 7,221 providers (92.3%) resolved to a named investor type. Private equity specifically is 0.64% of providers and 1.01% of beds once resolved. What remains open: 7.7% of providers (mostly single-home operators) have no PSC trail to follow, and this is a snapshot — PE ownership turns over on fund-cycle timescales, so it will date the same way R16 has already dated once. R23 is a separate, structural limit rather than a resolution gap: an opco/propco sale-and-leaseback split (as at Berkley Care) is invisible to any PSC walk, since the regulated operating company genuinely is management-owned and the institutional money sits entirely in an unconnected property company.
The leverage null is most likely a measurement failure. Charge counts count charges, not amounts, and sit on the registered-provider company rather than the propco or offshore parent where the structures CHPI describes sit. **Partially addressed**: the Companies House PSC snapshot reveals who controls the registered provider directly. A disclosed corporate controller is significant but inseparable from chain size (49% of single-site providers vs 78% of 51+ chains). The non-disclosure categories are confounded with sector — 76% of statement-only homes are charities, which are PSC-exempt and rate better — so they are reported decomposed rather than as one opacity dummy. Offshore controllers are too rare to test (35 of 3,128 companies). **Further addressed** (R21): two free, public routes reach real fund-level leverage data for specific named operators — Voyage Care’s bond-investor Annual Report (net debt/EBITDA 6.6x→5.0x, forced by its own Senior Secured Notes covenants) and Cortina Race LLP’s (Orchard’s parent) Companies House group accounts (interest cover ≈0.6x on trading profit). Both are single-operator case studies, not a market-wide figure — reaching the fund for the market as a whole still needs paid group-structure data, but the technique (bond covenants; UK LLP/group accounts, not just the operating subsidiary) is now proven and reusable.
The only figures traced so far (c.£180,000–£270,000 total development cost per bed) come from commercial care-home construction firms’ own marketing pages, which are not citable. **Largely addressed** (R18): LaingBuisson’s Care Cost Benchmarks model, via the DHSC-mandated 2022 Fair Cost of Care exercise, gives £110,000 per bed for new-build at a standard mid-market specification; a later, October 2024 LaingBuisson report for Care England gives £150,000 per bed for new, investment-grade stock. Neither is a full itemised build-cost figure (fees, FF&E, contingency not separately shown), so both are cited as named, dated anchors alongside — not a replacement for — the existing marketing-sourced range. Knight Frank’s own reports (R3, and its 2025 Capital Markets report) were checked directly and confirmed to contain no development-cost-per-bed figure at all, only land value.
Multiplying the 200,000-bed shortfall by a cost per bed is Bridgehead’s own arithmetic, not a published estimate. It is the single most powerful number available and therefore the one most likely to be attacked, so it needs sourced build costs, a stated range, and the method shown in full. **Confirmed dead end**: Knight Frank (two reports, checked directly), Savills’ own June 2026 UK Care Home Development Report (fetched and searched directly, zero £ figures anywhere in the text), Carterwood, Care England, RICS, the Health Foundation and the King’s Fund were all checked directly; none has published a total capital-gap estimate for UK elderly care. Savills does give an independent shortfall figure worth citing alongside R3 — 139,000 additional beds needed by 2036, a different horizon from Knight Frank’s c.200,000 by 2050, not a contradiction. Any headline £ capital-gap figure in this paper should be stated explicitly as Bridgehead’s own arithmetic (bed shortfall × the R18 cost-per-bed range), shown as a range with the method visible, not attributed to a published external source.
No existing dataset classifies English care homes by leverage, propco separation, ultimate parent jurisdiction and holding period. This has to be constructed from Companies House filings and the charge register, joined to CQC location-level ratings. **Built and tested at full scale** — see R17. The CQC-to-Companies House join works at 99.5%, the scale dimension is solid, and the group-structure layer above the registered provider now resolves for 92.3% of providers by investor type (family/founder, charity, mutual, EOT, REIT, private equity, pension/infrastructure fund), up from 76.5% after two pipeline revisions (5 and 7 September 2026). Rejoined to CQC ratings under the same controls as R15: mutual, charity/third-sector, REIT and local authority now all survive as significant, all in the protective direction; private equity does not. What remains: this sees who controls the regulated provider, not who owns the real estate underneath it — a real and likely under-counted gap, not an over-counted one (see R16/R19 on Barchester’s sale-and-leaseback, and R23 on Berkley Care's identically-shaped split).
The ‘buyer of last resort’ claim is plausible but untested. It requires an analysis of CQC deregistrations against ownership transfer, to establish whether acquisition demonstrably prevents closure. If it does not, the claim does not stand. **Tested directly, and found untestable from this data at any frequency** (R22): a 29-point quarterly panel built from CQC’s own historical archive (2019–2026) shows zero locations changing CQC Provider ID across 6.75 years, despite this exact window covering a large, independently-documented wave of institutional acquisitions. Verified against a known case (Danforth Care): the registered provider entity pre-dates Welltower’s 2025 acquisition of it by nearly two years and is unchanged since. The reason is structural — share-purchase acquisitions, which is how this sector’s PE/REIT deals are overwhelmingly structured, do not touch the registered provider at all, so no snapshot frequency would make this claim testable from CQC data; only ownership-chain data (R17) can see it. Separately, exit rates by provider size remain survivorship-biased and unusable until a provider cohort is fixed at a past date.
Twenty-three in total. Each entry notes what it supports, so any figure here can be traced back to it.