Teshima Art Museum

A 197-foot concrete shell, ten inches thick, cast on a mound of the site’s own dirt.

To build Teshima Art Museum, Kajima first built a hill.

They mounded the soil cut from the site into the shape of the room, shaped it with heavy equipment, surveyed it at 3,600 points, troweled mortar over it until it was smooth enough to cast against, tied 187 tons of rebar onto it, and poured a ten-inch shell over the top in one continuous pour lasting twenty-six hours. Once it cured, they sent excavators inside and hauled the hill out through two holes in the roof.

Most writing on Teshima covers the water droplet form, the quiet, the single artwork. Very little covers how it was built, which is the part worth knowing.

The finished shells in the terraces.

Photograph: Epiq, 12 March 2011. CC BY-SA 3.0.

The idea came from a ship’s propeller

Sōichirō Fukutake wanted a museum where architecture, art, and environment were one thing. Ryue Nishizawa designed it. Rei Naito made the single artwork inside. Mutsuro Sasaki did the structure. Kajima built it, starting in 2009, with Ikumi Toyoda running the site. Toyoda had run the Chichu Art Museum site five years earlier.

A free-form surface this size cannot be built on conventional shoring. There is no practical way to assemble scaffold and plywood formwork that holds a compound curve to tolerance across two hundred feet.

Toyoda’s solution came from the foundry. A ship’s propeller blade is not machined; its curves are cast in a mold made of packed sand. He asked whether the same approach would work for reinforced concrete: skip the falsework and make the mold out of the ground.

So Teshima is a sand casting. The site was the mold, and the mold was destroyed to release the part.

Floor plan.

One of the two unglazed roof openings.

Photograph: Bea Phi, 2025. CC BY-SA 4.0.

Flat was the hard part

A dome gets structurally better as it gets taller. Raise the crown and the section approaches a true arch, and load runs through the shell as compression. Flatten it and that goes away. A shallow shell carries much more bending, resists buckling far less, and pushes harder outward at the springing. The usual fix is more steel and a thicker section.

The design would not allow either. Nishizawa held the crown to about fifteen feet four inches over a longest span of roughly 197 feet. The shell came in at ten inches thick. That is a span-to-thickness ratio of about 236 to 1, at a rise of about one in thirteen.

Building section.

Two shells in the meadow.

Photograph: Kentaro Ohno, 12 October 2013. CC BY 2.0.

The steel it took

Kajima gives the roof concrete as roughly 785 cubic yards and the reinforcing steel as about 187 tons.

About 475 lb of steel per cubic yard

A conventional reinforced slab runs 150 to 250. If the 187-ton figure also covers the floor slab and foundations, the roof still lands near 280.

That is roughly double the normal rate, in a section ten inches deep. The flatness was paid for in steel. Kajima notes the rebar was placed by four hundred workers over the course of the job.

One more structural point. This is my reading, not a published claim. The build sequence shows the perimeter foundation was poured first, with the mound raised on top of it. The shell therefore lands on a continuous footing around its whole edge rather than on discrete supports, so the outward thrust is spread along the entire perimeter. That is a large part of why ten inches works at this rise. Sasaki’s shape-finding handled the rest, including the two roof openings, without visible local thickening. I could not find a published technical description of the Teshima analysis; his 2024 monograph covers the project.

Building the mound

Toyoda had his staff mock up every unknown at full scale: how to treat the mound surface so the concrete came off with a fine skin, how to detail reinforcement for a thin section on a complex curve, and how much heavy machinery could be driven inside afterward to dig the mold out without damaging the shell.

The cut material from grading was hauled to a temporary stockpile yard elsewhere on Teshima and held for later use as the mold.

Foundations went in first. The soil was then mounded on top of the completed foundation and shaped with heavy equipment, then surveyed in three dimensions at 3,600 points.

The mortar skin

Packed earth will not give a cast concrete surface, so the mound was skinned. Kajima names the material: bassa mortar, troweled on in repeated coats until the curve was smooth, with a release agent applied over the top. The mound and the mortar together made up the formwork. The dirt gave the mass; the mortar gave the surface.

Bassa mortar is a tilesetter’s material, not a formwork material. It is ordinary cement and sand, roughly one part to three or five, mixed with far less water than usual, until it is crumbly instead of plastic. The trade test for the mix is to squeeze a handful in your fist: it should hold as a ball. Too wet and it falls apart when you open your hand. Too dry and it never packs. Water is adjusted by feel, by hand, on site.

Its usual job is the setting bed under stone and floor tile, and the reasons it is used there are the reasons it worked here.

• Very little water means very little shrinkage, so it does not crack as it sets. A crack in the mold face would have printed straight into the underside of a white concrete roof that nobody could go back and patch.
• It does not flow. An ordinary wet mortar would have sagged down the dome. Bassa mortar stays where the trowel leaves it, which is why the trade uses it on slopes and steps.
• It screeds to high dimensional accuracy, which is the entire point when the target is a compound curve.

There is one more thing about the material worth noting, because it answers a question Kajima never addresses directly. The trade rule is that bassa mortar goes over a firm, compacted substrate; it cannot hold accuracy over loose fill. No compaction specification for the mound has been published anywhere I can find. But the choice of this material is indirect evidence that the earth underneath was compacted to a real standard, and that the mound was engineered fill rather than a pile of dirt.

After the mortar went on, the surface was surveyed again, all 3,600 points a second time, to check the mold rather than the earthwork. The allowable error against the drawings was set at 5 mm, about three-sixteenths of an inch, across the whole two-hundred-foot span. The reinforcing was then tied directly onto that surface.

The mold in section. Five layers. Three of them were removed.

Twenty-six hours

A joint anywhere in the roof would have ruined the effect, so it had to go in one pour. A small island has no batch plant that can feed one, so Kajima supplied ready-mix from a plant ship moored offshore.

The date was picked for clear winter weather. Work began at nine in the morning on 11 March 2010, with two pump trucks placing 26 to 39 cubic yards an hour. The mix was white concrete. Plasterers hand-finished the surface behind the pumps, working around falling leaves, through the afternoon and overnight. It finished at eleven the next morning, twenty-six hours after it started.

Digging the mound out

Machinery and conveyors worked through the two roof openings to remove the soil. Those openings are the only way in or out for several thousand cubic yards of dirt.

The floor was poured only after the mold was gone: roughly 562 cubic yards, again continuous. Worth noting, because English-language accounts often claim a single pour produced a seamless floor-to-wall transition. It did not. The roof shell is the seamless element. The floor is a separate, later pour, with a construction joint where it meets the shell.

The water

Rei Naito’s artwork, Matrix, is the floor itself. Here is how it works.

A well was drilled on the property. The museum sits partway up a hill above Karato, an area known for its groundwater, and there is enough natural head in the ground to push water up through the slab without a pump. It comes out through 186 holes in the floor, each about a sixteenth of an inch across, at irregular intervals through the day.

The slab is not ordinary concrete. It was given a water-repellent finish, so the water cannot soak in or spread out into a film. It beads instead, and a bead can roll where a film cannot.

Water beading on the floor.

Photograph: Iwan Baan.

Three things then move the beads. The floor was laid to a very slight slope. It also carries small undulations across its surface. And because the two roof openings are unglazed, there is always a light breeze in the room. Kajima credits the breeze along with the undulations for pushing the droplets around.

So the beads drift, run into each other, and merge. Larger drops move more readily than small ones and pick up others as they go. Over the course of a day the water gathers at the low points and forms a small pool. Benesse calls it a spring. On wet days rain comes through the openings and joins the water already on the floor.

The interior, with visitors seated on the floor.

Photograph: Iwan Baan.

So the shell was cast in soil from the site, and the artwork runs on water from under the site. There is a third thing, and it is probably the most useful one.

Wright returned again and again to the old idea that the reality of a building is not its walls but the space inside them. It is almost always a figure of speech. At Teshima it is a description of the work. The room was built first, as a solid hill of dirt. It was shaped by machine, surveyed twice at 3,600 points, and troweled smooth. Then it was carried out, and the volume of the room is the volume of the material that was removed.

Most buildings arrive at their interiors by subtraction after the fact. You put up structure and enclosure, and the space is whatever is left over. Teshima reverses the order. The void was the thing that got designed and built. The concrete is the record of where the hill used to be.

The usual argument for Teshima belonging to its landscape is about shape, and shape arguments work for any building with a curve in it. This one is harder to dismiss. The ground gave up the material, the water, and the room.