Ten days in Wuhan

Huoshenshan hospital went from ground-breaking to handover in about ten days. It is the cleanest natural experiment we have in what actually limits construction speed.

2 September 2026

In late January 2020, at the start of the COVID-19 outbreak, China built Huoshenshan hospital in Wuhan: 1,000 beds1 and about 34,000 square meters1, from site clearance to handover between 24 January and 3 February1. A second hospital, Leishenshan, followed days later at larger scale. The images of excavators working in formation became famous. The delivery lesson behind them is more interesting than the spectacle.

The design was not invented that week. Huoshenshan adapted the standard plan of Beijing's Xiaotangshan hospital1, built for the 2003 SARS outbreak and itself completed in seven days. The information flow that dominates most projects, design decisions, approvals, requests for information, was largely resolved before day one.

Wards were prefabricated container modules, assembled before arriving on site1 and fitted out in parallel, off the critical path, while civil crews prepared the ground. Labor and equipment arrived in whatever quantity the schedule wanted, around the clock, under national mobilization: 35 excavators and ten bulldozers1 on site levelling alone. Permits and utility connections were cleared by decree.

In other words: for about ten days, every one of the seven flows ran unconstrained toward a single, standardized building. What remained was nearly pure assembly, and assembly is fast.

The ten days

Sequence approximate, assembled from contemporaneous public reporting, January to February 2020.12

The caveats are the point, not an apology

This was a single-purpose field hospital under emergency powers, with scope and finish standards no commercial program would accept, in a command economy that could redirect a national supply chain overnight. It is not a template for a data hall, a railway, or a reactor. Anyone presenting it as one is selling something.

It is something more useful: an isolation of the variable. Huoshenshan demonstrates what duration looks like when distribution failure is removed, when information, materials, labor, equipment, space, precedence and external conditions are all ready the moment the work wants them. The answer is that the construction itself, the thing we usually blame, nearly vanishes as a constraint.

The same seven flows, two programs

Set Huoshenshan beside Crossrail's endgame, using the flow categories from lean construction4 as operationalized at Stanford CIFE3.

Author's assessment, applying the Koskela and Garcia-Lopez flow categories to the public record of each project. Not a figure from either source.

Five of the seven flows were fine on both programs. Crossrail did not fail on labor, materials, equipment, workspace or weather, and Huoshenshan did not beat it on those. The entire difference sits in two rows, and both of them are the rows a schedule does not carry.

Where your program sits

Every program lives somewhere between Wuhan and Crossrail. The position on that line is set less by how hard the crews work than by how well the flows around them are fed. That is a management choice, made long before the first pour, about what the program measures and what it leaves invisible.

For teams building AI data centers the useful question is not "how do we go faster". It is "which of the seven are we actually feeding, and which two are we going to discover late". Crossrail found out at trial running. That is an expensive place to find out.

This project in The Flow Index

This project's row in The Flow Index, which states the method and the evidence behind each cell. Author's assessment against the public record, not a figure from any cited source.

by Ahmed Schrute