The homebuilder that refuses to own the land

NVR gives its own history one sentence: "NVR, Inc., a Virginia corporation, was formed in 1980 as NVHomes, Inc." That is the whole of it. The filing contains no account of how the company came to build the way it does, and it is worth noticing the silence rather than filling it in: the widely repeated story that the land-light model was adopted in response to the company's early 1990s troubles appears nowhere in the document. What the filing gives you is the strategy as it stands, not its origin.

Today it sells houses in thirty-seven metropolitan areas across sixteen states and Washington, D.C., under three names: Ryan Homes to first-time and first-time move-up buyers, and NVHomes and Heartland Homes to move-up and luxury buyers.

Read the first page and the strategy is stated as a refusal. The company says it generally does not engage in land development, and instead acquires finished building lots from third-party developers under fixed price lot purchase agreements that require deposits which may be forfeited if it fails to perform.

That is a strange sentence for a homebuilder, because land is the industry's defining asset. The conventional builder buys raw ground years before it will sell a house on it, pays to entitle and develop it, and carries it on the balance sheet through whatever the market does in between. It is a construction business bolted to a levered land position, and in most cycles the land position is the larger bet.

NVR's filing gives the reason it declined that bet, and the reason is about capital rather than caution. The lot acquisition strategy, it says, has allowed the company to maximize inventory turnover, which it believes lets it operate with less capital and so raises returns on equity and total capital. The point of not owning land is not that land is dangerous. It is that land is slow, and slow assets ruin a return.

There is a second half to the arrangement that is easy to miss and matters enormously. NVR says its sole legal obligation and economic loss for failing to perform under one of these agreements is the deposit, under a liquidated damages provision. It may choose, for any reason and at any time, not to buy the lots. Read as a financial instrument rather than a contract, that is an option: a premium paid for the right to buy at a fixed price, abandoned if the price stops making sense.

So the question to carry through the rest: if the loss is capped at the deposit, who is bearing the land risk, and what are they charging for it?

Two segments, and the smaller one is not decoration. Homebuilding revenue was 10,094 million dollars in fiscal 2025 and mortgage banking fees were 230 million, on consolidated revenues of 10,324 million.

The homebuilding half works like any builder's: sell a house for about 461 thousand dollars, build it for less, keep the difference. In fiscal 2025 that difference was 2,141 million dollars of gross profit on 10,094 million of revenue, a margin of 21.2 percent, against 23.7 percent the year before and 24.3 percent the year before that. Selling, general and administrative expenses took 600 million.

The mortgage half is the part that makes the first half unusual. NVR originates the loan for its own buyer, and its capture rate, the share of NVR homebuyers who finance through NVR Mortgage, was 86 percent. That is not a modest attachment rate on an add-on product. It means the company is present at both ends of essentially every transaction, so it sees the buyer's income and credit before it sees the contract, and it earns a second fee on a customer it has already acquired.

Now the part that is genuinely different, and it is not on the income statement at all. Because NVR options lots rather than owning ground, its balance sheet does not carry years of land. Total assets were 5,857 million dollars against consolidated revenues of 10,324 million, which is a builder turning its entire asset base close to twice a year. A conventional builder carrying a multi-year land bank cannot approach that, and the whole return depends on it.

So the money is made three ways stacked: a normal construction margin, a second fee on the financing, and a rate of turnover that a land-owning competitor cannot match. Only the third one is a moat, and it is bought rather than granted, which is what the next section is about.

Per house, fiscal 2025 looks like this. NVR settled 21,915 homes at an average price of 461 thousand dollars, wrote 20,410 new orders at 456 thousand, and ended the year with 8,448 homes in backlog worth 4,008 million dollars. Orders came in below settlements, so the backlog shrank from 9,953 homes, and the cancellation rate rose to 17 percent from 14 percent the year before and 13 percent before that.

A 21.2 percent gross margin on a 461 thousand dollar house is about 98 thousand dollars of gross profit per settlement, before the 600 million of overhead spread across 21,915 houses, which is roughly 27 thousand each. That is the per-unit economics, and by the standards of the industry it is good but not extraordinary.

The extraordinary number is what those ordinary units earn on the capital behind them. Net income of 1,340 million dollars sits on 3,865 million of shareholders equity. Do that division yourself before reading on, because the gap between a 21 percent gross margin and the answer is the entire argument for the model.

Two things drive that gap, and it is worth keeping them separate. The first is real and operational: assets turn nearly twice a year because there is no land bank, so the same equity supports far more houses. The second is financial: NVR bought back 243,082 shares for 1,819 million dollars during the year, which is more than the year's net income, and equity fell from 4,210 million to 3,865 million as a result. A shrinking denominator raises the ratio without anything improving.

That distinction is the discipline this dossier is really teaching. A high return on equity produced by turnover is a statement about the business. A high return on equity produced by retiring the equity is a statement about the capital structure. Both are legitimate, they are not the same claim, and only the first one survives a bad year.

The wrong way to size this is to count American households. NVR is capacity-constrained in a specific and unusual way: its growth depends on other companies choosing to develop finished lots in the places it wants to build.

So the honest measure of its runway is the option book. NVR controlled about 180,100 lots at the end of fiscal 2025, up from 162,400, and it says it generally seeks to control a supply of lots suitable to meet a five-year business plan. Against 21,915 settlements a year, 180,100 lots is roughly eight years of production, so the constraint is not currently lots in aggregate.

It is lots in the right places, and the segment split shows the shape of that. The South East carries 72,900 controlled lots, the largest position, and produced the lowest gross margin at 18.3 percent. The North East carries 19,000, the smallest, at the highest margin, 25.5 percent. Where the lots are cheap and plentiful, the houses are cheap and the margins are thin. Growth in the option book is not automatically growth in profit.

There is a cost to holding the runway that a land-owning builder does not pay in the same form. Gross contract land deposits were 962 million dollars at year end, up from 785 million, and NVR expects to place a further 734 million of deposits under existing agreements as development milestones are met. Optioning capital is still capital. It earns nothing while it waits, and unlike land it cannot appreciate.

The genuine ceiling, then, is the supply of third-party developers willing to carry ground for a builder that has structurally declined to. NVR's own filing names this as the risk, and there is no version of the model that grows past it, because the model is defined by not doing that work itself.

Four homebuilding segments and a mortgage business, and the four are not variations on a theme. They are four different housing markets that happen to share a balance sheet.

Mid Atlantic is the core: 4,372 million dollars of revenue, a 23.3 percent gross margin, 723 million of segment profit, on 60,100 controlled lots and 8,287 settlements at an average price of 528 thousand dollars. North East is the smallest and the richest per house, 1,202 million of revenue at a 25.5 percent margin, settling 1,860 homes at 647 thousand each. Mid East runs 1,875 million at 21.1 percent. South East is the second largest by revenue at 2,645 million and the weakest by margin at 18.3 percent, settling 7,290 homes at 363 thousand.

Read the price column and the margin column together and the pattern is clean: average selling price and gross margin move in the same direction across the four. That is what it looks like when land cost is a fixed share of a house and the house is the variable. A 363 thousand dollar house in the South East cannot absorb the same lot premium as a 647 thousand dollar house in the North East.

Mortgage Banking sits across all four, capturing 86 percent of NVR's own buyers and closing 6,040 million dollars of loan volume for 156 million of segment profit. Its scale is set by the homebuilding segments and it has no independent growth path, which is the point: it is a second fee on a customer already acquired, not a business being built.

The deposit column is where this map gets interesting. South East holds 318 million dollars of net contract land deposits against Mid Atlantic's 348 million, on revenue barely over half as large. The lowest-margin segment carries proportionally the most optioned capital, which is exactly the combination that gets written down when a market softens.

Fiscal 2025 is the answer to this section, which is why it is the right year to read. The model did not break, but it was billed, and the bill is instructive.

NVR took net pre-tax charges of about 76 million dollars against its contract land deposits during the year, against 7 million the year before. The allowance for losses on those deposits nearly doubled, from 59 million to 111 million, and roughly 18,200 controlled lots now carry an impairment. Management attributes the margin decline to higher lot costs, pricing pressure from affordability, and those impairments, and it guides that the adjustments will have a materially negative effect on gross margins through the first half of 2026.

So hold the capped-loss argument up against that. Every individual loss was capped at a deposit, exactly as the contract promised. A capped loss taken 18,200 times is still 111 million dollars of allowance, and the abandonment is not free even when the deposit is small: the community stops, the overhead stays, and the lots that were the next two years of production are gone. A cap on the size of each loss is not a cap on the number of them.

The subtler cost is the one that never appears as a charge. Somebody has to bear the land risk NVR declined, and the developer who bears it charges for it in the lot price. That premium is inside the 21.2 percent gross margin every year, in good markets and bad, which is why NVR's margin sits below the best land-owning builders' peak-cycle margins. The model is not free insurance. It is insurance with a premium paid continuously and a claim collected rarely.

Then the structural dependency, which NVR names itself: the results depend on continuing to control an adequate number of lots in desirable locations, and it warns there is no assurance that supply will stay available on terms like the past, or that it will not have to commit more capital to controlling lots than it has historically. That last clause is the whole failure mode in one line. If developers demand larger deposits, or step back from carrying ground entirely, the model converges toward the land-owning model it was built to avoid, and the turnover advantage that produces the return goes with it.

Finally, watch the two numbers that were moving in opposite directions in fiscal 2025. Deposits at risk grew from 785 million to 962 million while the allowance against them nearly doubled. Capital went into the option book in the same year the option book was being written down. That is either conviction at the bottom of a soft patch or good money after bad, and the way to tell them apart is not the story: it is whether the impairment allowance keeps climbing and whether the cancellation rate, at 17 percent and rising for two years, turns back down.