An empty field, three years, ten thousand people

In 2015 I was handed a piece of land in central Taiwan and a deadline. The land was going to become an advanced semiconductor campus. Within three years, more than ten thousand people would arrive to work there every day. My job was everything those people would need that was not a fab tool: food, transport, security, housing support, buildings that stayed comfortable and safe, and the hundreds of small services that decide whether a campus feels like a place to work or a place to survive.

People assume the hard part of a project like that is construction. It isn’t. Construction is hard, but it is well understood. Engineers have schedules, contractors have penalties, and concrete cures at a known rate.

The hard part is that people do not arrive on the construction schedule. They arrive on the business schedule. And the business schedule, in a fab ramp, does not wait for anyone.

The real bottleneck was a mismatch of two curves

Draw two lines on the same chart. One is headcount on site, month by month. The other is service capacity: how many meals you can serve at noon, how many people your shuttles can move at shift change, how many access cards you can issue in a morning, how many work orders your maintenance team can close in a day.

On a greenfield site, both lines start at zero. Headcount climbs in steps, because tools come in, and the people who run them come right behind. Service capacity climbs in ramps, because every service needs a building, a vendor, a team, and a few weeks of rehearsal before it works under load.

Every crisis I saw on that campus lived in the gap between those two lines. When headcount jumps a step before service capacity has ramped, you get a queue, and the queue becomes the culture. People remember the first month of a new site for years.

So the question I learned to ask was never “how fast can we build?” It was “which curve leads, and by how much?”

Three decisions that mattered

Build the floor before the crowd. We treated day-one service capacity as a requirement, not a goal. If a building was going to open with a thousand people, the canteen, the shuttle loop and the help desk had to run at that load before the first person badged in. That meant paying for capacity that sat partly idle for a few weeks. It was the cheapest insurance on the project.

Design for the exceptions, not the average. The average day runs itself. What breaks a campus is the five percent: the night shift that needs a hot meal at 3 a.m., the visitor who arrives without clearance, the typhoon that closes the road. We wrote the exception paths down first, assigned an owner to each, and drilled them. When the exceptions have owners, the average day stays boring, which is exactly what you want.

Make every team a small control room. A campus growing that fast outruns any manager’s ability to approve things. So each service team got three things: a clear operating standard, the data to see their own performance daily, and the authority to fix problems inside that standard without asking. My job moved from deciding to designing the conditions for others to decide.

What I would do differently

I would have invested in shared data earlier. Each service ran its own records, and for the first year, joining them took manual work. If I had insisted on one common data layer from the start, we would have seen the headcount-versus-capacity gap weeks earlier, instead of feeling it in the queues.

I would also have written down more of what we learned while it was happening. A lot of the hard-won judgment from those years lived in people’s heads, and some of it left with them.

What transfers

You don’t need a semiconductor campus to use this. Any time something has to grow fast (a new office, a new team, a company after an acquisition) the same three ideas hold:

1. Find the two curves that must stay in step, and decide which one leads.
2. Give every exception an owner before you need one.
3. Turn managers into designers of the system, so decisions can happen where the problems are.

The goal of all this is a strange one. If you sequence well, nobody notices. The people who arrive in year three walk into a campus that feels like it has always been there. That invisibility is what good sequencing buys you.