Start with the number the whole scheme is a bet on: the tensile strength of a carbon nanotube. In the lab, individual nanotubes have been measured above 100 GPa — roughly two orders of magnitude beyond construction steel per unit mass. Every dream of building taller and lighter than steel allows runs through that figure. The Shimizu TRY 2004 Mega-City Pyramid is what happens when you take that lab number and extrapolate it to 2,004 metres of occupied structure over Tokyo Bay.

The specs, per the Shimizu Corporation proposal catalogued on Wikipedia, read like a systems diagram waiting for its physics to arrive: a pyramid 2,004 m high, five stacked megatrusses each sized to the Great Pyramid of Giza, 8 square kilometres of floor area, one million residents, and twenty-four-plus 30-storey towers not sitting on the ground but suspended from the truss lattice on nanotube cables. Foundation: 36 piers of special concrete into the seabed. Proposed start 2030, completion 2110 — an eighty-year build. It was patented internationally in October 1992, a decade after a Shimizu engineer reportedly walked out of Blade Runner fixated on the Tyrell Corporation’s twin pyramids.

Read it as a cartographer reads any large system: not by its silhouette but by its dependency graph. The pyramid’s entire load story is one edge. Conventional materials can’t carry it — the structure would crush under its own mass — so the design offloads the impossible weight onto “super-strong lightweight materials based on carbon nanotubes and graphene presently being researched.” That phrase is doing the structural work. The building is a placeholder pointing at a materials-science roadmap that, in 2026, has not delivered load-bearing structural CNT members at anything near this scale. The megatruss topology is real engineering; the tendon material is a promissory note.

←TODAY: In 2026 no one can spin a kilometre-scale, load-rated carbon-nanotube tension member — the pyramid is unbuildable for exactly one missing edge. →3012: The arcology that gets built is the one whose critical material shipped first and whose failure modes were mapped before the ribbon-cutting. Fulcrum: A structure is only as speculative as its single least-available dependency — find that node and you know the real completion date.

And this is where the megatruss choice is smarter than the fantasy around it. An open network of struts that lets wind and seismic energy pass through rather than resisting it as a solid mass is the correct answer to the site. Japan sits on the Pacific Ring of Fire; Tokyo is a city that has rebuilt itself around earthquake and tsunami as recurring inputs, not edge cases. The pyramid answers that with porosity — the same logic Heinz Isler chased in his thin concrete shells, where, as PAZ’s reference library keeps insisting, the form follows the forces rather than fighting them. It is also the logic behind every segmented-shell project ICD/ITKE has scaled from pavilion to public roof: distribute the load through many small, legible members, and the structure becomes something you can actually fabricate and reason about.

The Guardian’s Unbuilt Tokyo survey filed this pyramid alongside the city’s “depthscrapers” as a monument to congestion — and Koolhaas’s Delirious New York already gave us the frame: the metropolis as an unplanned project theorising itself through impossible objects. The Shimizu pyramid is that theory pushed to one million people. It is worth studying not as a thing to build but as a joined-up research question: which of its edges — CNT cables, algae power, robotic assembly, personal-pod transit through the trusses — is closest to shipping, and which is furthest?

Atelier: Your Büro will never build an arcology, but you already design structures whose feasibility rests on one un-arrived component — a material spec, a supplier, a proprietary connector, a solver license. The Shimizu lesson for a team living with AI-assisted design is the same as its lesson for a megatruss: the exciting node is rarely the fragile one. This Monday, take your next live project and mark, on the model, the single element whose failure or unavailability halts everything — then ask procurement for its second source before you draw another line.

Hack: Find the load-bearing node in your own stack before it finds you. A dependency graph has articulation points — nodes whose removal splits the network — and those are your single points of failure. Model your project’s real dependencies (not the org chart) and let the algorithm surface the third one you forgot.

import networkx as nx
g = nx.Graph([("seabed_piers","megatruss"),("CNT_cable","megatruss"),("CNT_cable","suspended_tower"),("PV_film","grid"),("grid","pumps")])
cuts = list(nx.articulation_points(g))
print("single points of failure:", cuts)  # -> 'CNT_cable', 'megatruss', 'grid'

Run it on your live model’s supplier and system list, not on a pyramid. The node that shows up in that list and has no second source is the one to fix. Japan’s genius has always been holding tradition and frontier in the same hand — timber joinery beside seismic base-isolation; the pyramid deserves the same discipline. Draw the dependency graph, name the fragile edge, and either resource it or design around it before it becomes the reason the project stalls at 2110.

Source: en.wikipedia.org