China is already exporting buildings. In 2024, it shipped $3.22 billion of products classified under the customs category for prefabricated buildings, more than any other country. Nearly $200 million went to the United States. The category includes far more than houses, but that is precisely the point. Buildings are beginning to move through the world as manufactured goods. Walls, rooms and steel modules can leave a factory, cross an ocean and arrive ready for assembly.
The house is becoming a product. The land beneath it cannot be.
The real-estate trade hiding inside the age of AI begins with that split. Artificial intelligence and robotics will make more of the physical world reproducible. They will compress the labor, time and waste required to build a home. A structure that once demanded a rotating cast of trades may eventually arrive as a kit and be assembled by machines.
When the cost of producing the structure falls, however, the value of a desirable home does not have to fall with it. More of the price can migrate into the part nobody can manufacture: the location. Real estate has always contained two assets that we insist on quoting as one:
Property value = structure value + land value.
The first is a depreciating, reproducible object. The second is a claim on a specific place, along with its access, legal rights, schools, jobs, power, water, fiber, weather, neighbors and future uses. AI will widen the distance between the two assets.
The building is already entering the factory
Construction has resisted productivity gains longer than almost any major industry. A house is still assembled outdoors, in changing weather, by crews that arrive in sequence and frequently wait on one another. Each project is treated as a local exception, leaving the whole arrangement vulnerable to automation.
At Wolf Ranch in Georgetown, Texas, ICON and Lennar completed a 100-home development using building-scale robotic printers. The machines extruded concrete walls around the clock. By the second year, eleven printers were producing two homes a week and had cut printing time in half. The first project cost more than expected while the companies worked through foundations, roofs, utilities and finishing trades. By completion, Lennar executive Stuart Miller said costs and cycle time had fallen by half through the learning process. The partners are planning a larger community.
China supplies the other clue. Its prefab industry takes construction work that once happened only on-site and moves it into controlled factories. Chinese exports in the broad prefabricated-building category rose from $1.47 billion in 2015 to $4.34 billion in 2025, according to customs figures reported by Xinhua. Foshan alone has more than 300 exporters in the sector. Hotels, dormitories and offices already leave these production lines as modules.
A factory can standardize a room. Software can turn a design into a bill of materials. Robots can cut, weld, print, inspect and assemble. AI can optimize the floor plan around cost, climate and local code before a worker reaches the site.
Elon Musk takes the curve much further. In a July 2026 interview with The Economist, he predicted an “age of amazing abundance” by 2036. Asked what money would be needed for, he named food, housing, transport and entertainment, then imagined robots and AI providing more goods and services than any person could consume.
Musk did not make a narrow forecast that every house would be free by 2036. He made a much larger and more speculative claim about AI and robotic production. Ten years is probably too short for permitting systems, utilities and local politics to move at machine speed. Yet the direction is plausible even if the date is wrong.
Homes will become easier to produce. Location will remain stubbornly finite.
We have seen this split before
The historical record already tells us which half of a property tends to absorb appreciation. Morris Davis and Jonathan Heathcote decomposed the US housing stock into land and structures from 1975 through mid-2006. Their definition of land was broader than soil. It captured the plot, the location and whatever made a home worth more than the replacement cost of its building.
Over that period, inflation-adjusted residential land prices rose by a factor of 3.7, or about 270%. Existing-home prices rose 96%. Structure replacement costs rose 33%.
The final decade was even more revealing. From 1996 through mid-2006, real house prices rose 70%. Real structure replacement costs rose 29%. The implied price of residential land rose almost 160%.
The authors concluded that both the long-run rise and the cyclical movement in US house prices were driven primarily by residential land rather than structures. Land prices were more than three times as volatile as structure prices.
A follow-up study by Davis and Michael Palumbo exposed how local this process can be. They decomposed home values across 46 large US metropolitan areas from 1984 to 2004. The average land share rose from 32% to roughly 50%. By the end of 2004, land represented about 75% of home value in West Coast cities and 65% on the East Coast, compared with about 40% in the Midwest, Southeast and Southwest. The house was the visible object, while the appreciation sat beneath it.
Katharina Knoll, Moritz Schularick and Thomas Steger widened the lens to 14 advanced economies from 1870 to 2012. Their decomposition attributed about 80% of the rise in house prices between 1950 and 2012 to land. Even when they reduced the assumed initial land share to 25% as a sensitivity test, land still explained more than 70% of the increase.
Marc Francke and Alex van de Minne estimated that a typical structure with little or no maintenance lost about 43% of its value after fifty years. A very well-maintained home showed almost no long-run physical deterioration in their model. A building can preserve value. Some buildings can become scarce assets themselves.
The ordinary structure, however, starts with a problem land does not have. It wears out. It becomes functionally obsolete. Its kitchen ages, its roof leaks and its layout stops matching what buyers want. New structures compete with it at replacement cost.
Land does not depreciate in the same physical sense. A location can still lose value, sometimes permanently, but not because a newer acre was manufactured next door.
AI will move value toward whatever cannot be copied
Every age of abundance produces a new scarcity. The internet made information cheap and attention expensive. Streaming made songs abundant and live access scarce. AI is making competent digital work cheaper, which raises the value of proprietary data, trusted distribution, energy and physical access.
Housing should follow the same pattern. Imagine that robotic construction cuts the cost of a standard structure by half. On plentiful land with permissive zoning, the full price of housing can fall. Austin has already shown the first half of this process without robots: a large apartment-building wave pushed rents down. If machines make supply even more elastic, many ordinary housing markets should become more affordable.
Now move the same cheap structure to a parcel within walking distance of Stanford, beside Central Park, on the Pacific coast, inside a top school district, or next to a power substation with secured capacity and long-haul fiber. The structure remains cheap. The parcel remains scarce.
A lower cost of construction may even increase land value in high-demand places. If buyers can spend less on the building, they can bid more for the right to put it somewhere valuable. Economists see the same residual logic in development today: the finished property value, minus the cost of construction and required profit, determines what a developer can pay for land.
Robotics does not abolish housing scarcity. It changes its address. Scarcity migrates away from drywall, framing labor and standardized floor plans, then gathers around entitlements, grid access, clean water, insurance, school boundaries, coastlines, transit, culture and proximity to other people. The best land will store the value released by cheaper production.
There will be two maps of prime land
The AI era is creating one map for people and another for machines. The human map still rewards dense networks of talent, capital and culture. The San Francisco Bay Area remains the deepest AI company and research cluster in the country. New York combines finance, media, enterprise customers and global talent. Boston and Cambridge join universities, biotechnology and technical labor. Seattle retains cloud infrastructure and engineering depth. These markets are expensive and heavily regulated, but their scarcity is real because the network already exists.
The machine map looks different. An AI data center cares less about restaurants and more about megawatts, fiber, water, permitting and a large contiguous parcel.
CBRE reported that North America’s primary data-center vacancy rate fell to 1.4% at the end of 2025 even as capacity grew 36%. Northern Virginia absorbed 1,102 megawatts during the year. Dallas absorbed 470.8 megawatts. Recent and pending site transactions in Northern Virginia and the Northeast exceeded $8 million per acre, while grid capacity for existing projects in most markets was largely committed through 2030. A separate clue explains the premium: greenfield sites able to secure power within 18 to 36 months are now highly sought after. Four US land zones deserve attention.
Northern Virginia and its expansion ring
Ashburn remains the backbone of the cloud because fiber, customers and technical infrastructure have accumulated there for decades. Powered and entitled land is exceptionally scarce. The direct market is expensive, so the more interesting signal may be the expansion toward Richmond and Pennsylvania, where developers can remain connected to the Northeast while searching for power and larger parcels.
Dallas–Fort Worth and the Texas Triangle
DFW combines population growth, corporate demand, logistics, fiber and a deregulated power market. Dallas became the third North American data-center market to pass one gigawatt of inventory in 2025. Austin and San Antonio add semiconductor investment, technical talent and land. The interesting parcels have a credible route to power and water inside a corridor that people and infrastructure are already choosing.
Central Ohio
Columbus has become a cloud availability zone rather than a speculative dot on a map. JLL reported that hyperscalers acquired more than 2,000 acres across the region in the two years through mid-2025. CBRE noted new fiber routes linking Columbus to Chicago and Ashburn. Grid constraints are pushing activity outside New Albany and into the rest of Ohio, where a parcel with real interconnection prospects can be more valuable than one sitting beside a busy highway.
The Carolinas and the Mid-Atlantic frontier
Charlotte–Raleigh combines universities, finance, life sciences, population growth and comparatively low power costs. Pennsylvania offers deregulated electricity, proximity to both New York and Northern Virginia, and more greenfield land. These markets may capture both sides of the map: places where skilled people want to live and places where AI infrastructure can still be built.
The screen is simple to describe and difficult to execute. Look for durable population and income growth, a deep employment base, transport access, low climate exposure, verified water, multiple fiber routes, zoning that permits valuable use and power that is deliverable rather than merely visible on a map.
A title deed establishes ownership. The value comes from the rights and connections attached to it.
The house becomes cargo
For most of modern history, a house looked permanent and the land beneath it looked passive. AI and robotics reverse the picture.
The structure becomes editable. A model redesigns it. A factory produces it. A robot assembles it. New modules replace old ones. The building begins to behave more like a car, an appliance or a piece of software with a physical shell. The location becomes the durable asset.
The historical evidence was already pointing there. In the United States, land prices outran structure costs. Across 14 advanced economies, land explained most of the postwar house-price increase. Across major US cities, the land share climbed fastest where desirable locations were hardest to reproduce.
AI adds a new force to an old pattern. It will make intelligence abundant, then use that intelligence to make more physical goods abundant. Houses will be among them, though probably later and less completely than Musk expects.
Someday a buyer may choose a home online, watch a factory build it and see robots assemble it in a week. The expensive decision will still be where to put it.
Sources and further reading
Observatory of Economic Complexity, “Prefabricated buildings in China trade,” 2024 trade data. The HS 9406 category includes residential and non-residential prefabricated buildings.
World Integrated Trade Solution / UN Comtrade, “Prefabricated buildings exports by country,” 2024.
Xinhua, “Faster, greener, more affordable: China’s modular building solution goes global,” May 30, 2026; and “China’s factory-built buildings find growing markets overseas,” August 12, 2026.
CNBC, “Inside the world’s largest 3D-printed housing development,” March 12, 2025.
The Economist, “An interview with Elon Musk,” July 24, 2026; full video published by The Economist on YouTube.
Morris A. Davis and Jonathan Heathcote, “The Price and Quantity of Residential Land in the United States,” Journal of Monetary Economics 54, no. 8 (2007): 2595–2620. Author manuscript.
Morris A. Davis and Michael G. Palumbo, “The Price of Residential Land in Large U.S. Cities,” Journal of Urban Economics 63, no. 1 (2008): 352–384.
Katharina Knoll, Moritz Schularick and Thomas Steger, “No Price Like Home: Global House Prices, 1870–2012,” American Economic Review 107, no. 2 (2017): 331–353.
Marc K. Francke and Alex M. van de Minne, “Land, Structure and Depreciation,” Real Estate Economics 45, no. 2 (2017): 415–451.
CBRE, “North America Data Center Trends H2 2025,” 2026.
JLL, “North America Data Center Report, Midyear 2025,” August 2025.
US Census Bureau, “Growth in Metro Areas Outpaced Nation,” March 13, 2025.