Introduction

A Building in Thirteen Months

Construction of the Empire State Building began on March 17, 1930. Thirteen and a half months later, on May 1, 1931, it opened. At the peak, its steel frame rose four and a half stories a week, and in one stretch of ten working days the crews erected fourteen floors. The final bill was about $41 million, well under budget.

The story is usually told as one of nerve: ironworkers on open beams, a race for the title of tallest building in the world. Read as economics, it is a story about time. The owners had borrowed $27.5 million from the Metropolitan Life Insurance Company, and the stock market had crashed five months before the first steel went up. Every month the site stayed open added interest. Every month the offices stood unfinished was a month without rent. The builders' real achievement was not height. It was compressing the coordination of thousands of people, hundreds of suppliers and one crowded Manhattan lot into a single, relentless schedule.

Those thousands of people were the schedule. More than 3,500 worked on the site at its peak, 3,439 of them on a single day in August 1930. Many were Irish and Italian immigrants, and a sizable group were Mohawk ironworkers from Kahnawake, near Montreal. Officially, five of them died. The photographer Lewis Hine went out onto the steel beside them, and his pictures still show the building as human work rather than as a line in a ledger. Both descriptions are true. A building's cost is mostly the hours of the people who make it, and its speed is the speed at which those hours can be arranged.

Almost a century later, we build slower. A comparable landmark today can take five years or more from first design to opening, and it costs far more per square foot in real terms. The steel is not the problem. Steel, glass and concrete are among the cheapest things per ton that an industrial economy makes. What has grown expensive is everything around the matter: the drawing, the negotiating, the sequencing of trades, the insurance against error, and the interest that runs while all of it is arranged.

This book is about that surrounding layer, which I will call coordination cost , and about what happens to buildings, cities and economies when machines make it close to free.

The Floor Beneath a Life

That can sound like a question for developers. It is not only theirs. The price of building sits underneath three things that shape a life: the shelter a household can afford, the work by which millions of people earn a living, and the public buildings a city can pay for, from schools and clinics to stations and libraries.

The first of these is already under strain. In the United States, half of all renter households in 2023, a record 22.6 million, spent 30 percent or more of their income on housing and utilities. Worldwide, UN-Habitat estimates that about 2.8 billion people live with some form of inadequate housing, 1.1 billion of them in informal settlements and slums, and that more than 300 million are homeless. Not all of that is the cost of construction. Much of it, as Chapter 10 shows, is the price of land and the rules about what may be built on it. But the cost of turning materials into rooms is part of every rent and every mortgage, and it has refused to fall for a very long time.

The second is the other side of the same ledger. Every dollar this book proposes to remove from a building's cost is, today, somebody's wages: the carpenter's, the welder's, the glazier's. A book about making buildings cheaper is also a book about who stops being paid to make them, and Chapter 12 weighs that loss as carefully as the saving.

The Maha Principle began this series with the architecture of a flourishing life. This book follows one part of that architecture down to its foundations: the material floor beneath shelter, work and public life, and the question of who owns it.

The Stubborn Industry

Over the last century, almost everything that comes out of a factory has become cheaper in real terms. Cars, clothing, electronics and appliances have all followed the same curve, because their makers learned to repeat the same steps millions of times and then handed those steps to machines. Buildings did not follow. In 2023 the economists Austan Goolsbee and Chad Syverson described "a large and decades-long decline in construction sector productivity" in the United States. Measured by physical output, such as houses built per worker, productivity has fallen or, at best, stood still for several decades. They also found that construction firms have grown less efficient at turning materials into finished buildings.

Two older ideas explain much of this. In the 1960s William Baumol observed that industries built on human labor, where productivity barely rises, grow steadily more expensive relative to industries that automate. A string quartet takes as long to play today as it did two centuries ago, yet the players must be paid wages set by an economy that has grown far more productive. Construction behaves much the same way. Earlier still, in 1937, Ronald Coase asked why firms exist at all, and answered that organizing work inside a firm is sometimes cheaper than arranging it through contracts. Construction shows the costly side of his argument. Nearly every building is a one-off, assembled by dozens of separate firms bound together by contracts, and a large share of its cost is the price of making those firms work as one.

None of this changed with the computer. Drafting software, 3D modeling and prefabrication each promised a revolution, and each improved some part of the process without bending the overall curve. Any argument that this time will be different has to explain why. This book's answer is that earlier tools made the drawing cheaper but left the site and the factory untouched. The combination arriving now reaches all three.

Six Layers of a Price

The central tool of this book is a simple breakdown. The price of any building, and in principle of anything, can be split into six layers:

1. Matter : the raw materials themselves, such as ore, aggregate, sand and metal, priced as commodities.

2. Energy : what it takes to extract, refine, move and place that matter.

3. Land : the ground the thing occupies, which no one can manufacture.

4. Coordination : every human hour spent designing, fabricating, assembling and managing, plus the interest that runs while those hours are spent.

5. Risk : the premium paid against error, through contingency budgets, insurance and the margins that contractors add for uncertainty.

6. Permission : the cost of being allowed to build, through permits, codes, public review and the professional liability of the engineers and architects who sign the drawings.

In a modern landmark tower, the fourth layer is the largest, and much of it is hidden. A developer's budget usually lists "materials" as a single line. Open that line and most of it turns out to be coordination too. A double-curved glass panel costs far more than the glass in it, because someone had to bend, laminate, inspect, crate and ship it. Rolled bronze costs more than bronze. Fabricated steel costs more than steel.

The material floor is what would be left if the coordination layer, and the risk that comes from coordinating badly, fell to almost nothing. It is not zero. Matter, energy, land and a core of permission remain. But it is far lower than what we pay today, and it is not fixed. As automation reaches the factories that fabricate components, the floor itself moves down toward the commodity price of matter and the energy needed to shape it.

Physical AI is the name this book uses for the combination that could push costs toward that floor: generative models that design and simulate, robotic systems that assemble on site, and automated factories that fabricate the parts. Each has arrived partly on its own before. What is new is that all three now draw on the same kind of intelligence and can work as one system.

The Tower in This Book

To keep the argument honest, this book is built around a model rather than a slogan, and the model is built around a single building. You should know from the start whose building it is.

The reference tower that appears in every chapter takes its form from the Caldera, my own unbuilt design concept for a headquarters: a flared, volcano-like tower on a waterfront site, wrapped in double-curved glass and a rolled bronze exoskeleton, with a light-filled central atrium, sky gardens and public piers at its base. The model gives it two million square feet on a five-acre site. It is exactly the kind of building this book is about. Its form is affordable only if the cost of complexity falls, and its site, foundations and materials are expensive at any level of automation.

I have an obvious interest in the answer, so here is what the interest has and has not touched. The cost figures are illustrative assumptions for a building of that size and form, stated openly in Chapter 8 and Appendix A, and meant to be challenged. They are not an estimate of what the Caldera itself would cost, and the argument would run the same way for any similar design. And the first draft of this book promised a more flattering result than the model would give: that a landmark costing $3 billion to build today would cost $1 billion when built by machines. The model says otherwise, and the correction stays in the book, because it is the most useful thing the model taught me.

The Borrowed Light asked, in its tenth chapter, which imagined worlds can actually be built, and found that a design can be coherent in every joint and still impossible to build. This book asks that question of one building, with a ledger in place of a conjecture: what would have to become cheaper for a design like the Caldera to leave the drawing board, and what never will?

What the Numbers Say, and What They Do Not

Chapter 8 follows the reference tower through four stages of automation. Under today's labor-gated methods it costs about $3.1 billion and takes seven years from entitlement to opening. The model then removes cost one layer at a time, in the order automation is most likely to arrive.

The first result surprised me. Generative design on its own, the part of this story that receives the most attention, saves only about 9%. Drawings are expensive, but they are a small share of the whole. Robotic assembly on site does more, bringing the total down by about 23%. About half of that step comes from lower labor costs, net of what the robots cost, and half from a shorter schedule, with less interest, site overhead and land carrying cost. The largest step comes when automation reaches the factories, and the hidden coordination inside "materials" begins to fall away. At the theoretical limit, the tower costs about $1.2 billion and takes under three years.

The second result matters as much as the first. About $750 million of the cost does not move at all, whatever the level of automation: the commodity content of the materials, the land, and the core cost of permission and liability. That is the material floor for this building, and the rest of the book keeps returning to it. It is also why the $1 billion tower of the first draft is out of reach: to get there, the land or the raw materials themselves would have to become cheaper.

Some things this book does not claim. It does not predict that landmark towers will cost a third of today's price by any particular year. The model's inputs are illustrative, stated openly and meant to be challenged. It does not assume that people disappear from building: engineers and architects will still sign the drawings, and the law will still hold them responsible. And it does not assume that cheaper construction means cheaper buildings for the people who use them. That last point turns out to be the most important of all.

Who Keeps the Saving

The architect Cass Gilbert called the skyscraper "the machine that makes the land pay." The phrase cuts both ways. If building becomes cheaper everywhere in a city at once, developers compete for the same scarce sites, and the saving tends to flow into the price of land. David Ricardo described this mechanism two centuries ago, and Henry George built a political program around it. In the book's model, if land prices absorb half of the saving, the developer's cost at the theoretical limit is not $1.2 billion but about $2.15 billion, and land grows from under a tenth of the budget to more than half.

So the question "how cheap can building become?" is followed by a harder one: who ends up with the difference? Landowners, developers, the companies that own the robotic fleets, the people who live and work in the buildings, or the public? For the households that already spend a third or more of their income on rent, that second question is the only one that matters. The answer is not set by technology. It depends on land policy, planning rules and ownership, and the later parts of the book treat it as the political question it is.

Beyond Buildings

The six layers are not special to architecture. A hospital visit, a university degree and a legal contract can be broken down the same way, and like buildings they sit far above their material floor because most of their cost is coordination. Manufacturing went through this compression over the last century. Construction may be next, and it is the clearest place to watch because its coordination layer is so large and its output so physical.

The Synthetic Self asked what happens to work and value when competence stops being scarce. This book asks the same question of the physical world, where the scarce thing has been the coordination of hands.

That is why the book ends by widening its view from the architecture of buildings to the architecture of economics. When coordination is cheap, value moves to whatever cannot be automated: land, energy, raw materials, the capital that owns the machines, and permission. An economy built that way has a different shape from the one we know. It rewards different people, raises taxes on different things and is limited by different constraints. Buildings let us see that shape early, in steel and glass, before it appears everywhere else.

How This Book Fits, and How to Read It

The Volcanic Engine argued that a habitable planet is not a finished state but a process, kept alive by a heat engine that continually melts, outgasses and renews the crust. This book applies the same way of seeing to the built world. A city is also a process, maintained by a steady flow of energy and matter, but for most of history that flow has been throttled by the cost of coordinating human hands. Take away the throttle and what remains is the underlying flow: matter moved by energy, on land, within rules. The material floor is the point where a building's cost is set mostly by physics and hardly at all by organization.

Each chapter begins by saying what kind of claims it makes. Some report what has been measured; those are empirical. Some build arguments and models from those measurements; those are theoretical. Some reason about what may follow; those are speculative. And some weigh what it all means; those are interpretive. Some chapters move between two of these, and say which. The labels are there so that you can judge each claim by its proper standard, and so that the speculative parts do not borrow authority from the empirical ones. When an argument becomes a guess, I will say so.

The book moves from the concrete to the general:

Part I, The Old Balance Sheet, sets out where the money in a building goes and why a century of better tools has not moved it. Part II, The Zero-Marginal Brain, examines what generative design and simulation remove, and why on their own they remove little. Part III, The Descending Floor, follows automation onto the site and into the factory, builds the model, and marks the costs that do not fall. Part IV, The Megastructure Renaissance, asks what cheap complexity does to land prices and to architectural form. Part V, The Human Habitat, turns to the workers displaced, the people housed, and how long automated buildings can last. Part VI, The Architecture of Economics, applies the six layers to the wider economy and asks where value settles once coordination is cheap.

The Empire State Building went up in thirteen months because its builders treated coordination as the real material, and paid for it in the hours of thousands of people. This book asks what happens when coordination stops being scarce, and who will be holding the saving when it does.

Notes and sources

1. "Empire State Building," Wikipedia, https://en.wikipedia.org/wiki/Empire_State_Building (construction dates; pace of four and a half stories a week and fourteen floors in ten working days; cost of $40,948,900; the $27.5 million Metropolitan Life loan of December 1929; first steel on April 1, 1930; more than 3,500 workers at the peak and 3,439 on August 14, 1930; five deaths by official account; Lewis Hine's photographs, in the writer Jim Rasenberger's description quoted there). The article gives the budget both as $50 million and as $60 million, so the text says only that the building came in well under it.

2. Joint Center for Housing Studies of Harvard University, The State of the Nation's Housing 2025, as reported by the National Low Income Housing Coalition, "Latest State of the Nation's Housing Report Finds Record Number of Housing Cost-Burdened Renters," June 30, 2025, https://nlihc.org/resource/latest-state-nations-housing-report-finds-record-number-housing-cost-burdened-renters ; UN-Habitat, Strategic Plan 2026–2029, summary, https://unhabitat.org/sites/default/files/2025/10/un-habitats_strategic_plan_2026-2029_one_pager.pdf

3. Austan Goolsbee and Chad Syverson, "The Strange and Awful Path of Productivity in the U.S. Construction Sector," NBER Working Paper 30845 (2023), https://doi.org/10.3386/w30845

4. William J. Baumol and William G. Bowen, Performing Arts: The Economic Dilemma (Twentieth Century Fund, 1966); R. H. Coase, "The Nature of the Firm," Economica 4 (1937): 386–405.

5. The Caldera is described in Chapter 11. The reference tower's assumptions are set out in Chapter 8 and in full in Appendix A; the first draft's $1 billion figure and why the model cannot reach it are discussed there under "Why Not One Billion."

6. Cass Gilbert, "The Financial Importance of Rapid Building," The Engineering Record, June 30, 1900, pp. 623–624, documented at https://www.barrypopik.com/index.php/new_york_city/entry/a_skyscraper_is_a_machine_that_makes_the_land_pay/ ; David Ricardo, On the Principles of Political Economy and Taxation (John Murray), on rent; Henry George, Progress and Poverty (1879).

7. Within this series: The Maha Principle: The Architecture of Human Flourishing; The Synthetic Self, Chapter 9, "The Economics of Synthetic Abundance"; The Volcanic Engine; The Borrowed Light, Chapter 10, "The Worlds That Cannot Be Built."