Hugh Grosvenor's central London estate was valued at around £9.9 billion in mid-2026 — roughly 300 acres of Mayfair and Belgravia held unbroken since 1677, per a July 2026 Westminster profile. Yet the number that matters to an apartment buyer is sharper: the gap between two similar-sized flats — one in a Berkeley-built tower, one at 1 Mayfair. Near-identical floorplates, very different pounds per square foot.
Treat the premium like a model output rather than a vibe: measure both flats on a stated area basis, pull six datasets — sold prices, floor areas, planning consents, service charges, marketing renders, resale pairs — and split the difference into four priced components. The estate's own logic explains why the dirt dominates: the Grosvenors do not sell, they lease, a discipline running since the 1720s, so even the freehold beneath Grosvenor Square remains the Duke's while Eaton Square flats trade for tens of millions on terms their owners never fully own.
One input rarely gets audited: the photorealistic CGI that off-plan buyers absorb as ground truth. Stress-test the gap against it and only part of the spread survives contact with a resale — the rest is new-build polish that evaporates on the way to the second-hand market.
Anatomy of the Gap
Fix both arms before touching any decomposition. Arm one: 1 Mayfair, a super-prime W1 launch quoting per square foot on a gross internal basis. Arm two: a matched basket of Berkeley Group volume schemes — St Edward's Kensington Row and St George's 250 City Road — transacting at their own recorded rates. The distance between the two arms is the multiple every later section decomposes. The lazy reading — you're just paying for the Mayfair postcode — dies on contact with the arithmetic: decomposed properly, only part of the premium maps to land value. The rest is capitalized service charges, leasehold-versus-freehold structure, and specification that exists in the render but not in the contract schedule.
Start with the measurement layer, because it silently edits both arms. RICS has been migrating residential measurement from legacy GEA/GIA conventions toward IPMS, and switching basis — balconies counted or excluded, wall thicknesses included or omitted — can swing quoted area materially. That mechanically moves pounds-per-square-foot with no change in price or product. A Mayfair unit re-measured onto a generous basis can look cheap against a tightly quoted Berkeley flat, or the reverse. No premium claim means anything until the area basis is stated in writing.
Developers price four things into the gap. First, freehold land scarcity: according to a July 2026 profile of the Westminster estate, the family has held the Mayfair and Belgravia freeholds unbroken since 1677, the ground beneath Grosvenor Square and most of Eaton Square remains Grosvenor-owned, and the estate carries a £9.9 billion valuation — developable W1 freehold is nearly exhausted, while Berkeley sells leasehold flats on regenerated sites at volume. Second, specification delta: stone, bespoke joinery, ceiling volumes well above a standard volume-spec shell. Third, staffed services — concierge, spa, parking — funded through the service charge. Fourth, a trophy-address liquidity premium attached to one building's brand. Only the first lands fully in the title; the second lands partially in the specification schedule; the third and fourth sit almost entirely outside the contract.
The marketing layer then inflates perceived specification. Off-plan sales run on CGI renders governed by the ASA CAP Code's misleading-imagery rules — but a render is judged as an advertisement, while delivery is governed by the contractual specification schedule, and the two documents rarely describe the same room. Anyone who evaluates synthetic imagery professionally recognizes the failure mode: modern renders are photorealistic generative output, optimized for perceptual plausibility rather than dimensional accuracy, so buyers systematically credit units with lighting, views, and spatial generosity that appear nowhere in the schedule. The CAP Code polices deception; it cannot police optimism baked into a diffusion pipeline.
The carry mechanism converts operating cost into apparent asset value. The trophy unit's staffed-building service charge runs at multiples of a standard Berkeley scheme's, recurs annually, and gets capitalized by the next buyer at exit — a recurring cost stream discounted into the price the outgoing owner commands. A slice of the headline premium is therefore deferred operating cost wearing an asset-value costume.
Tenure is the quietest driver. Since recent leasehold reform forced peppercorn ground rents on new leases, ground rent no longer separates the arms; what still prices is leasehold versus share-of-freehold. Berkeley's volume model sells leasehold; W1 trophy stock increasingly offers a share of the freehold. This is legal structure, not cosmetics — and it is the oldest structure in the book. The Grosvenors began granting long building leases in the 1720s, retaining the freehold while buildings revert at expiry; as the same July 2026 profile puts it, "the Grosvenors do not sell. They lease."
On these numbers the default winner is the Berkeley arm; the Mayfair case must be proven unit by unit, starting with the table below.
| Arm | Quoted level | Tenure | What the quote bundles |
|---|---|---|---|
| 1 Mayfair (W1 super-prime launch) | Developer quote, gross internal | Increasingly share-of-freehold | Land scarcity, staffed services, trophy brand |
| St Edward's Kensington Row (Berkeley) | Recorded scheme rate | Leasehold | Volume-regeneration spec, standard ceilings |
| St George's 250 City Road (Berkeley) | Recorded scheme rate | Leasehold | Volume-regeneration spec, standard ceilings |
| Implied multiple at basket midpoint | Derived from the two arms | Cross-tenure | The gap the next section sizes bucket by bucket |
Action before you trust either number: demand the IPMS-basis measured area and the contractual specification schedule alongside the full render set, then let the decomposition that follows assign each pound of the gap to a bucket that either survives resale or evaporates.
The Benchmark Set
Six datasets decide this trade, and only two of them record actual transactions. Everything else is a house view — useful for direction, worthless as ground truth. Anyone who has tuned a model against self-reported evals knows how that ends, so assemble the benchmark set the way you'd assemble an eval suite: ledger first, indices second, filings and surveys as bounds.
The ledger arm comes from LonRes. According to LonRes, achieved rates in W1 — Mayfair and St James's — sit at a clear premium to what Berkeley's London schemes clear at scale, on thin volume. Both clusters come from recorded completions rather than asking prices, which is what lets them anchor both arms of the ratio above.
The trend check cuts against the postcode story. According to Knight Frank's Prime Central London Index, PCL values peaked in an earlier cycle and have spent the years since below that peak, with annual growth only recently turning slightly positive. If the premium were simply "you're paying for the Mayfair postcode," the postcode would have compounded. Instead the index shows years of range — the Mayfair arm has been range-bound, not compounding, so the case has to rest on recoverable value rather than momentum.
Savills supplies the forward view, with a caveat attached. According to Savills' current five-year prime London forecast, prime stock should outgrow mainstream cumulatively across the forecast period, front-loaded early. Front-loaded forecasts concentrate their confidence in the first year and stretch thinnest at the horizon, and the error bars around a handful of super-prime W1 sales run far wider than anything around the Berkeley basket, where homogeneous volume completions pin the distribution tight.
Then verify independently against the public ledger. Pull Land Registry Price Paid records for W1K and each Berkeley-completion postcode and clean them the way you'd clean a training set: drop duplicate registrations (new-builds typically appear more than once — first sale and later resale), join on postcode district, trim part-building and portfolio transfers. The same clustering LonRes reports falls out of the raw ledger, confirming it without touching a single agent-reported index.
The cost side has a filing behind it. According to Berkeley Group's annual report, the company completes homes at volume, with repeatable floor plates, a national supply chain, and fixed costs amortized across that output — the mechanism that lets Berkeley deliver at a fraction of W1 land-adjusted cost, a denominator no boutique Mayfair scheme shares.
Quality, finally, is a null result. According to HBF's customer-satisfaction star ratings, Berkeley holds a top-band rating with strong recommendation rates, and build-quality satisfaction is statistically indistinguishable across both arms. Better walls cannot explain the premium, so the premium has to live in land, tenure, and capitalized services — exactly where the recoverable-value test bites.
Weight the set accordingly: LonRes is the working source for achieved-price comparables, Land Registry is the ground truth that adjudicates any dispute, the two house views set the trend scenario, and the filing plus the survey bound the cost and quality sides. When a brochure quote disagrees with the ledger, the ledger wins — which is why the decision rule demands verified achieved-price comparables before any Mayfair unit clears the bar.
| Benchmark | What it records | Key reading | Role in the decision |
|---|---|---|---|
| Knight Frank PCL Index | House-view index of PCL values | Peaked in an earlier cycle; below peak since | Trend check: Mayfair arm is range-bound |
| Savills five-year prime forecast | Forward house view | Prime outgrows mainstream cumulatively; front-loaded early | Scenario input; widest error bars |
| LonRes | Achieved prices, W1 and Berkeley schemes | W1 clears at a premium to Berkeley schemes | Primary source for comparables |
| Land Registry Price Paid | Public transaction ledger | W1K + Berkeley postcodes, recent years | Ground truth; adjudicates disputes |
| Berkeley annual report | Audited corporate filing | Volume completions at mainstream price points | Cost-base sanity check |
| HBF star ratings | Buyer satisfaction survey | Top rating band; strong recommendation | Quality null result |
Four-Bucket Decomposition
Run the gap through an ablation before you pay it. Anyone who has debugged a vision pipeline knows a benchmark delta rarely belongs to one component — you zero out pieces one at a time and see what survives. Treat the spread between the two arms above the same way: decomposed into four buckets, only part of it turns out to be the Mayfair postcode at all. The remainder is capitalized service charges, leasehold-versus-freehold structure, and specification that exists in the render but not in the contract schedule. Triangulating Land Registry transfers, LonRes achieved prices, and Knight Frank's index view produces the allocation below; read each slice as a judgment call with wide error bars, not a quote.
| Bucket | Gap allocated | 1 Mayfair side | Berkeley-basket side | Winner |
|---|---|---|---|---|
| Land / location scarcity | Largest slice | Estate-core W1 freehold; squares laid out 1715 | No scarcity rent on regional plots | Mayfair, conditionally |
| Specification | Midsized slice | Bespoke fit-out beyond the functional line | NHBC-warranted volume spec | Split |
| Services and carrying costs | Smaller slice | Capitalized service-charge differential | Standard estate charge | Berkeley |
| Tenure and exit liquidity | Smaller slice | Trophy-brand buyer pool | Broad mainstream resale market | Conditional Mayfair |
| Survives resale | Part of the total | Land plus durable spec only | Functional spec largely holds | Buy Berkeley by default |
Bucket 1 is the only line item with a three-century pedigree. According to Bradshaw's Hand Book, both Grosvenor and Hanover Squares were laid out in 1715, and ground that old in W1 behaves like a fixed training set — nobody is minting more of it. But the winner is Mayfair only conditionally, because scarcity attaches parcel-by-parcel: according to a July 2026 Westminster profile of the estate, the family does not own every building on every street, having pruned plots off over the centuries. In flats, the land value attaches cleanly only to share-of-freehold stock; a leasehold unit rents the scarcity without ever owning it.
Bucket 2 splits. Berkeley's NHBC-warranted new-build specification delivers most of the functional quality — envelope performance, plant, acoustics — at a fraction of bespoke cost, and returns diminish sharply once a fit-out pushes past the point where spend migrates from measurable function to taste. Taste does not transfer at resale. The survival logic implies only a minor slice of this bucket comes back at exit.
Bucket 3 is the cleanest call in the table. A service-charge differential that recurs annually gets capitalized into the asking price at prevailing rates — the buyer pre-pays an annuity of running costs. Under the canonical rule, that makes it a liability wearing an amenity's clothes, and no concierge desk recovers the capitalization at sale. Berkeley wins outright.
Bucket 4 is conditional Mayfair. A trophy brand genuinely narrows the buyer search, but only while the trophy-buyer pool is being bid up; in a downturn that pool thins first and fastest, so the liquidity premium is an option that expires precisely when you would exercise it.
Sum the survivors: land plus durable specification is the share of the gap that endures — the verdict cell above. Everything else is consumption, not carry. The framework's explicit winner is therefore buy Berkeley, and the premium is rational only for the specific unit that clears the recoverability threshold and passes the tenure test defined in the closing section. Run all four buckets on any candidate before either number leaves your account.
What the Data Doesn't Tell You
Land Registry tells you what changed hands; it never tells you why. Price-paid records omit condition, floor level, aspect, and — the silent killer — seller concessions. A launch-phase print can embed furniture packages or duty contributions that never surface in the registered figure, so the "achieved price" you match against may be gross of money the buyer never effectively paid. LonRes narrows further: it logs only member-agent transactions, and super-prime stock routinely trades off-market, so the panel is a selected sample, not a census. Knight Frank contributes forecasts — direction, not ground truth, as the benchmark section established. Anyone who has fit a model on a biased sample knows the fix isn't more rows; it's knowing which questions the sample structurally cannot answer.
This blindness is why the laziest explanation — you're paying for the Mayfair postcode — survives contact with the data. A single recorded price cannot separate buckets, so the entire premium collapses into "location" by default. Decompose it and only part maps to land value; the remainder is capitalized service charges, the leasehold-versus-share-of-freehold structure, and specification that lives in the marketing renders but not in the contract schedule. The datasets don't refute the postcode story; they can't see the components, which is why the burden of proof belongs on the specific unit, never the district.
Treat the recoverable share as a portfolio mean sitting on wide unit-level dispersion. Floorplate position, aspect, and level push individual outcomes in both directions around that mean — two laterals in the same building can land on opposite sides of the recoverable-value bar while averaging to the same headline. Tenure amplifies the spread: much of Mayfair's freehold rests with the great estates — According to the July 2026 Westminster profile, the 7th Duke of Westminster inherited the title at twenty-five, and his family's holdings anchor the district — so genuine share-of-freehold stock is structurally scarce, and each qualifying unit trades as its own micro-market with its own comp depth.
| Failure mode | Detection test | Effect on the rule |
| Incentive-inflated comparable | Demand the reservation form's concession disclosures; reconcile against the registered price | Comp counts as unverified until reconciled |
| Off-market deal absent from LonRes | Require three verified achieved prices, never agent appraisals | Below three, the rule defaults to the Berkeley arm |
| Area measured on mixed bases | Re-measure both arms to one gross-internal basis before dividing | Mismatched measurement silently distorts the gap |
| Tenure dressed up in marketing | Read the title register, not the brochure | No share of freehold, no premium — comps become irrelevant |
| Repriced service-charge capitalization | Pull three years of service-charge accounts; stress the recoverable share downward | The threshold's calibration goes stale between editions |
None of these flips the default; they change how much verification you owe. The rule breaks conservatively in thin markets — where one perfect in-building sale exists, insisting on three verified comps discards real information, and you will occasionally walk away from a justified premium. It breaks dangerously under regime change: if reform caps how service charges capitalize into values, or a building-wide refurbishment converts render-spec into contract-spec, recoverable value moves faster than Land Registry prints arrive. And it breaks through selection — distressed and probate sales print low, trophy trades print high, and neither extreme describes your unit.
So interrogate the documents, not the narrative: order the title register and recent service-charge accounts, secure concession disclosures in writing, and re-measure both arms on a single area basis. Clear every test plus the threshold and the premium is defensible; fail any one and the difference between the two arms stays pocketed — which, in 2026, remains the base case.
Render Risk and Thin Samples
LonRes logs only a thin trickle of high-value W1 transactions each quarter — call it small n. Anyone who has evaluated a model on a small test set knows the pathology: the confidence interval around the point estimate runs wider than the effect being measured. Every single-address £/sq ft premium quoted for a 2026 W1 launch carries exactly that noise, so publish the n beside every number. An estimate without its sample size is marketing, not measurement.
The specification bucket has a worse problem than small n: it stays partly unverifiable until practical completion. Off-plan marketing sets, audited line by line, routinely show material CGI-to-contract divergence — ceiling coffers omitted from the schedule, porcelain substituted for stone, view corridors blocked by consented-but-unbuilt neighboring schemes. According to Westminster's July 2026 area profile, Mount Street trades on its red-brick Queen Anne facades and Michelin-starred frontage; a render can frame that streetscape beautifully while the contract schedules none of it. Diff every CGI against the specification schedule line by line and price only what the schedule names — that is also where the "you're just paying for the Mayfair postcode" reflex dies, because so much of the marketed specification exists only in the image, not in the document you legally buy.
No historical series contains the current policy regime. Non-dom status has ended under the FIG replacement, and a confirmed high-value council-tax surcharge is set to land on England's most valuable homes, banded by property value. Both land disproportionately on the trophy tier that underwrites Mayfair exits, so any backtest fitted on earlier prints extrapolates a buyer base that no longer exists in the same shape — with an annual, value-banded charge compounding against exit liquidity.
Off-plan adds a mechanical wedge: completion-stage surveyor valuations on prime W1 off-plan units routinely come in below contract price, because valuers anchor to achieved comparables and reach for the conservative end when quarterly prints run this thin. The realized premium can therefore diverge from the contracted one before any market movement occurs — stress the recoverable-value screen down accordingly before trusting a marginal pass.
The model cannot price sterling. Prime London demand is levered to global equity markets and the pound, and a sharp currency move shifts the effective £/sq ft paid by dollar-based buyers by more than the entire specification component of the gap above. The bucket you can least verify is smaller than one bad week in equities.
Hedonic regressions on data this thin leave most of the variance unexplained, so read the four-bucket allocation as a structured prior with named judgment calls — not a measurement. That is precisely why the decision rule holds together: the recoverable-value threshold and the three verified achieved-price comparables are margins of safety sized to absorb these error bars, and share-of-freehold tenure is the one premium input verifiable from the title register rather than a render.
| Failure mode | Hard signal | Effect on the premium call | Gate it justifies |
|---|---|---|---|
| Thin comps | LonRes: only a thin quarterly tape of high-value W1 prints | Error bars wider than the gap | Publish n; three verified achieved-price comparables |
| Render divergence | Line-by-line audits of off-plan sets routinely find divergence | Spec share unverifiable until practical completion | Diff CGI vs. contract schedule; pay for scheduled items only |
| Valuation wedge | Completion-stage surveyor vals routinely land under contract price | Realized premium below contracted premium | Stress the recoverable-value screen down by that wedge |
| Policy shock | FIG regime in force; high-value council-tax surcharge confirmed | Trophy-tier exit liquidity thins | Require threshold headroom plus share-of-freehold |
| Currency/equity swing | A sharp sterling move exceeds the entire spec component for USD buyers | Effective £/sq ft moves outside the model | Time entry to sterling, not only to listings |
| Model ceiling | Hedonic fits on this data explain little variance | Allocation is a structured prior | Treat the recoverable-value screen as margin of safety |
Before paying either number, demand three artifacts: the n behind the premium quote, the specification schedule behind the render, and the surveyor's completion-stage comp set. If any one of the three is missing, the Berkeley-built arm wins by default — pocket the difference.
Worked Case
Hold size constant — matched floor areas on the IPMS basis
```
Frequently Asked Questions
How large is the Grosvenor estate underlying Mayfair, and how long has the family actually owned it?
Per a July 2026 Westminster profile, Hugh Grosvenor's central London estate was valued at around £9.9 billion in mid-2026 — roughly 300 acres of Mayfair and Belgravia held unbroken since 1677, with long building leases granted since the 1720s so buyers hold long leases and never the ground.
Of the six datasets used to decompose the gap, how many actually record real transactions?
Only two record actual transactions — the LonRes ledger of recorded W1 completions and Land Registry Price Paid records for W1K and each Berkeley-completion postcode — while indices, forecasts, filings and surveys serve as direction and bounds rather than ground truth.
Where do the four priced components of the premium actually land in the legal paperwork?
Freehold land scarcity lands fully in the title, the specification delta lands only partially in the contractual specification schedule, and staffed services plus the trophy-address liquidity premium sit almost entirely outside the contract.
Does ground rent still explain any of the price difference between the Mayfair launch and the Berkeley schemes?
No — since recent leasehold reform forced peppercorn ground rents on new leases, ground rent no longer separates the two arms, and what still prices is leasehold versus share-of-freehold, with 1 Mayfair increasingly offering a share of the freehold against Berkeley's leasehold volume model.
Can the same apartment legitimately be quoted at two different pounds-per-square-foot figures?
Yes — because RICS has been migrating residential measurement from legacy GEA/GIA conventions toward IPMS, switching basis (balconies counted or excluded, wall thicknesses included or omitted) can swing quoted area materially and mechanically move pounds-per-square-foot with no change in price or product, which is why 1 Mayfair quotes on a gross internal basis.
If the premium were really just about the Mayfair postcode, what would the Knight Frank index show?
It would have compounded — instead Knight Frank's Prime Central London Index shows PCL values peaked in an earlier cycle and have spent the years since below that peak, with annual growth only recently turning slightly positive, leaving the Mayfair arm range-bound so the case must rest on recoverable value rather than momentum.
Also worth reading: Historic 2006 Rural Ohio Estate How 4,814 Square Feet of Modern Living Adapts to Today's Real Estate Market: Historic 2006 Rural Ohio Estate · Leesburg VA Real Estate Market Analysis Average Home Size Reaches 3,800 Square Feet in Q4 2024: Leesburg VA Real Estate Market · Greeneville, TN Real Estate Analysis How $3774 Per Square Foot Shapes Local Market Dynamics in Late 2024: Greeneville, TN Real Estate Analysis