, Britain
A hosting company in Docklands ordered the switchgear for its second hall in the summer of 1999. The chillers went onto the roof that autumn, the raised floor went down over the winter, and by the early spring of 2000 the hall was finished. Power in, cooling running, fire suppression tested, cages built, nothing in them.
Of the two customers who were going to fill it, one had gone quiet over Christmas and the other was waiting on a funding round that never closed.
The first hall was let and paying. The second was a clean, cold, empty room that cost money every day and would carry on costing money every day until somebody put equipment in it.
The peak has a date, and the funding stopped within weeks
The technology market peaked in March 2000. The Nasdaq Composite reached its high in the second week of that month and then fell, not in one day but over two years, and the firms that had been raising money on a business plan and a growth rate discovered that the next round was not there.
The London end of it had a second component. The auction of third generation mobile spectrum in Britain closed in April 2000 and raised something over £22 billion for the Treasury. The money came out of telecoms balance sheets at precisely the moment those balance sheets stopped being able to raise more, and telecoms firms were both the largest builders of network capacity and a large part of the customer base for everybody else's.
What that looked like from a data centre in 2000 was not a share price. It was a diary that stopped filling.
A data centre is a power contract with a building around it
The thing that makes an empty hall so expensive is that almost none of the cost is variable.
You do not buy electricity in a data centre the way a shop does. You agree a supply capacity with the network, and a substantial part of what you pay every month follows that agreed capacity rather than the units you actually draw. You agreed it when the hall was designed, for a hall that was going to be full. Cooling runs anyway, because a hall has to be held within a temperature and humidity range whether it is earning or not. The generators are tested and serviced on a schedule. Security is a rota, and a rota does not get smaller because the building is emptier.
Then there is the building. Halls of that vintage were taken on long leases with upward-only rent reviews, and the rating assessment follows the fit-out rather than the tenant.
The rule of thumb that the trade learned in those years is that an empty hall costs a large fraction of what a full one costs, and earns nothing. There is no version of the business where the answer is to wait quietly.
The lead time is what does the damage
Capacity is not ordered in the month you need it. Between the decision to fit out a hall and the day it can take equipment there is well over a year: planning, power, plant on long delivery, the electrical work, the commissioning.
That means every hall in Britain that came into service in 2000 and 2001 was a decision taken in 1998 and 1999, in a market that no longer existed by the time the room was ready. The capital had been committed, the lease had been signed, and the only remaining decision was what to do with a finished asset nobody wanted.
The same lag worked in reverse on the way out. Capacity built for the boom stood empty for years, and the second-hand capacity that came onto the market from firms that failed reset the price for everybody who had survived. You could not re-let at the 1999 rate, because the hall down the road had been bought out of an administration for a fraction of its build cost and its owner could price accordingly.
The survivors were the ones with the cheapest debt
This is the part worth taking away, because it is not a lesson about technology at all.
Two operators with the same halls, the same tenants and the same empty space did not have the same outcome. The one that had built with equity, or with bank debt at a modest margin and a long tenor, could carry an empty hall for three years, keep it warm, and let it at whatever the market eventually paid. The one that had built on borrowing struck in 1999, when the market would lend to anything with fibre in it, had an interest bill that assumed occupancy and covenants tested against earnings that were not going to be there.
Neither of them had an asset problem. They had the same asset. What failed was the capital structure, and it failed on a date determined by a covenant test rather than by anything a customer did.
That is also why the sensible move looks like a defeat at the time. Letting a hall cheaply, on a long term, to a tenant who will still be there in five years, is worse business than holding out for the old price, right up until you notice that holding out costs you the full running cost of the hall every month and the old price is never coming back.
What you can actually do with an empty hall
There are four levers and all of them hurt.
- Mothball it. Shut the cooling down to a holding level, isolate what can be isolated, and accept that bringing it back takes months you may not have when a customer finally appears.
- Let it cheap and long, which fixes your revenue at the bottom of the market and pays your interest through it.
- Break it up and sell smaller, which costs more in cages, metering and management than a single tenancy would.
- Sell the lease or the company, which in 2001 meant selling into a market with more sellers than buyers.
The operators still standing in 2003 had generally chosen the second.
Where this shows up in the game
Racks for Rent is a historical start. It puts you in a data centre in London in March 2000 with two halls fitted out, one of them let, and gives you eighteen months. It is one of the hard ones. You begin with eighteen million pounds of borrowing that somebody else agreed to, and with one customer who is about sixty per cent of the revenue and has given notice. Two more prospective tenants withdraw in months one and three, which is the market arriving rather than bad luck. The cost of cooling and powering an empty hall is charged in full. You may mothball a hall, which cuts that cost and takes four months to reverse, and any customer you take now is taken at the price of now, for five years.
It is scored on cash at the end: finish the eighteen months with £1,500,000 in the bank, or you have not done it. Run out of cash before then and it ends early. The target rewards the unglamorous answer, because letting a hall cheaply on a long term scores better than holding out for a price nobody is paying, and the Racks for Rent page sets out the start, the rules and the failure conditions in full.
The seed is fixed at 600011, so everybody attempting it meets the same Britain, the same withdrawals in the same months and the same power bill, which is what makes the scores comparable. It runs under ironman rules: one save, and what happens happens.