Data centre: the kilowatt nobody pays for

“For five years I thought I was selling space in a rack and a bit of electricity. It turns out I sell electricity, and the space I give away almost for free. And half the electricity I give away for nothing at all.”

Nazar Ostapchuk runs a small data centre in Uzhhorod. Colocation: the client brings his own servers, Nazar gives him the rack, the power, the cooling and the uplink. Twelve years in IT infrastructure, the last five as the owner. The hall is three quarters full, clients do not leave, the kit is his own. And the profit is such that you cannot see it.

A hall of forty racks

There are 40 racks in the hall. 29 are let, 11 stand empty.

That is 72.5% occupancy, and the figure has held for a second year. The eleven empty racks are not a shortage of demand: two are reserved for a client about to bring his kit in, and the rest stand idle because Nazar deliberately does not sell the last row. He knows something is wrong with it, and until now could not say what.

Every let rack draws 3.5 kW on average. That is a normal figure for mixed load: one client keeps two servers and nothing else, another packs the rack to the ceiling and asks for an extra phase.

The client pays on two lines. Rent for the space — 9 000 ₴ a month. Electricity — by the readings of his own meter, at 6.1 ₴ per kWh.

• space in a rack: 9 000 ₴ a month, fixed
• electricity: 6.1 ₴ per kWh by the client's own meter
• uplink and cooling: included in the price of the space
• access to the hall and an engineer's hands: included

The contract is transparent, clients understand it and nobody argues. The transparency is exactly where the trouble sits: the line “by the readings of your own meter” looks honest and is at the same time the main hole in the hall's economics.

What the client pays for

Twenty-nine racks at 3.5 kW each is 101.5 kW in the servers. Over a year, if they run without stopping, that comes to 889 140 kWh.

You can count it from either end and get the same answer. One 3.5 kW rack across the 8 760 hours of a year gives 30 660 kWh. Twenty-nine such racks — the same 889 140.

That is the figure the client sees. That is what he pays for.

The client's invoice is written from his own meter

The meter in the rack measures what this client's servers consume. It does not measure the cooling those servers demand, the losses in the UPS, the extract fans or the lighting of the hall the rack stands in.

All of that draws electricity too, bought at the feed along with everything else.

What the data centre pays for

There is a different instrument at the feed. It measures everything that came into the building.

Servers make heat, and that heat has to be carried out of the hall or the kit shuts down in twenty minutes. Chillers, pumps, UPS units, controls — all of it runs off the same line, and none of it has a meter tied to a client.

Nor is this a hidden fault. Every hall works this way: a server turns electricity into heat with almost nothing left over, and the heat then has to be thrown out into the street, which costs more electricity. The UPS is not free either — it heats up on its own and loses part of what passes through it.

A kilowatt in a server drags half a kilowatt behind it

Nazar's ratio is stable: for every kilowatt in a server the hall draws another half a kilowatt on everything else. The industry calls this PUE, and 1.5 for a small hall with no cold-aisle containment is an ordinary result.

So the feed does not carry 101.5 kW but 152.25.

The difference is 50.75 kW. Over a year that is 444 570 kWh which pass through the feed and never appear on any client's invoice.

The coefficient that decides everything

The third line is worth a careful look. The data centre buys 1 333 710 kWh and invoices for 889 140. The rest — 444 570 kWh — belongs to nobody in particular, which is why it lands on no invoice.

PUE is not a technical figure but a pricing one

Engineers read PUE as a measure of how good the hall is: the closer to one, the better the cooling was designed. For the owner it is something else — the multiplier you have to put on the purchase price of a kilowatt-hour before naming your own price to the client.

“I was shown that figure about three times in five years. They say: Nazar, your PUE is one and a half, not bad for a hall like this. I nodded. I heard a technical compliment and did not hear that somebody had just told me my cost price.”

The year in figures

The report is put together correctly. All six lines are reconciled with bank statements and signed certificates, and any audit will confirm them. It simply never puts purchase and billing in the same units, because one arrives in hryvnias a month and the other in kilowatt-hours per client.

6.1% net. For a business with its own building and its own kit that is little, and Nazar knew it. He just thought the reason was the empty racks.

Nazar took this report to the bank and twice borrowed against it for equipment. The bank looked at the turnover, at the building he owns, at the steady clients, and asked nothing. Six percent net for an infrastructure business with assets like that is not a reason to worry — it is a reason to talk about scaling up.

Where the difference goes

Let us look at the electricity on its own.

Sold for 5 423 754 ₴. Bought for 6 935 292 ₴.

The difference is 1 511 538 ₴, and it has a minus in front of it.

The markup sits on the wrong kilowatts

Nazar bills 6.1 ₴ while buying at 5.2 ₴. A markup of 0.9 ₴ per kilowatt-hour, almost eighteen percent, and on paper that looks like a perfectly normal resale.

But the markup is charged on 889 140 kWh, while the data centre pays for 1 333 710. The markup sits on the clients' kilowatts; the cost sits on all of them.

Reselling electricity with a markup produces minus 1 511 538 ₴ a year.

The overhead kilowatts

The annual net profit of the data centre is 520 462 ₴. The hole in the electricity is 2.9 times bigger.

So the hall earns on letting out space and hands part of what it earns to the power company for kilowatts no client ever saw on an invoice.

Put that on one line with the things Nazar actually controls. All the fixed costs of the hall — the building, seven people, security, the uplink, the licences — come to 1 100 000 ₴ a year. The hole in the electricity is bigger than all of it, and noticeably so.

“For two years I hunted for savings in the fixed costs. I went through the security firm, the uplink, the insurance, I even squeezed the cleaner. One million one hundred thousand for the whole lot. And next door a million and a half was leaking away every year, and I never looked there, because it came as a utility bill.”

For comparison: all eleven empty racks, let at the old tariff, would give 40.7% of what the right price on the racks already let delivers. The hall physically does not hold as much space as the mistake in the contract costs.

The break-even tariff

The right price is easy to work out. The purchase price of 5.2 ₴ multiplied by the coefficient of 1.5 gives 7.8 ₴ per kilowatt-hour.

That is the price at which electricity comes out at zero. Not at a profit — at zero.

The arithmetic here is so simple that it is hard to believe. If you buy one and a half kilowatts for every kilowatt you sell, then your cost of a kilowatt is the purchase price times one and a half. There is no other mathematics here.

Nazar bills 6.1 ₴. The shortfall is 1.7 ₴ on every kilowatt-hour, and across 889 140 kWh that is exactly the 1 511 538 ₴ the difference came to.

• purchase: 5.2 ₴ per kWh
• break-even price for the client: 7.8 ₴
• actual price for the client: 6.1 ₴
• shortfall: 1.7 ₴ on every kilowatt-hour

The contribution of a single rack

Now you can see what one let rack really brings in.

The third line of the table is the same in both columns, and that is the whole point. The purchase does not depend on what Nazar billed the client: 30 660 kWh per rack multiplied by 1.5 and by 5.2 ₴ gives 239 148 ₴ whatever was agreed.

The space becomes a contribution only at the right tariff

At a tariff of 6.1 ₴ a rack brings in 55 878 ₴ a year — 18.9% of everything invoiced from it. The rest is eaten by the difference on the electricity, and the rent for the space effectively subsidises the power company.

At a tariff of 7.8 ₴ the electricity comes out at zero and the space itself becomes the contribution: 108 000 ₴. That is 1.93 times more.

Why none of this is visible

The reason is not carelessness. A cut like this never appears on its own, neither in the accounts nor in the billing.

• electricity is purchased on a single monthly invoice, with no split by client
• billing comes off the meters, which is a different source of data
• PUE lives in the technical documentation, not in the financial model
• rent for the space and electricity sit on the same “revenue” line in the report

To see the minus you have to put together two figures that stand side by side in no report: the kilowatt-hours from the feed and the kilowatt-hours from the invoices.

What had been tried before

“The first three years I just sold. I thought: fill the hall and everything will add up. I filled it to twenty-nine, and nothing added up. Then I decided the problem was in the fixed costs and started cutting them. That did not add up either.”

“A consultant even did the sums for me: fill the hall and you will come out at twelve percent margin. I believed it back then. What he actually did was take my revenue per rack and multiply it, without asking what I pay for a kilowatt and what I sell it for.”

The logic was right for any other business. There is capacity in the hall, the capacity is underused, so you have to sell. In a data centre that logic works differently, because every new rack brings with it not only revenue but its own share of the overhead kilowatts.

Empty racks cost nothing. Occupied ones cost more than they look.

Two roads

When Nazar finally saw the figure, he had two options. Find more clients, or fix the tariff.

Five more racks are realistic: there are enquiries, there is spare power, there is space. Each gives 55 878 ₴ of contribution, so 279 390 ₴ together, and net profit would become 799 852 ₴.

Fixing the tariff means raising the price of electricity from 6.1 ₴ to 7.8 ₴ for all twenty-nine. That is 1 511 538 ₴, and net profit would become 2 032 000 ₴.

The arithmetic of the choice

The tariff is 5.41 times stronger than selling.

And this is not a comparison of “five racks against the tariff”. To catch the tariff up through sales, you would have to let 27.05 new racks.

The hall has a physical limit, the tariff has none

There are 11 empty racks in the hall. Even letting all eleven gives less than half of what the right price on the ones already let delivers.

This hole physically cannot be closed by selling: there is not enough space in the hall.

“That is what floored me. For five years I ran after clients, and the figure I needed was sitting in a contract I wrote myself. And to fix it I did not have to run anywhere.”

The limit of the hall

There is a technical limit too, and it deserves a word of its own.

The cooling capacity of the hall is designed for a certain amount of heat. The closer the hall gets to that limit, the worse the cooling works and the higher the coefficient climbs. So the last racks fill up dearer than the first, and PUE of 1.5 in a full hall will no longer hold.

The first racks in a hall fill up almost for free: the chillers run anyway, the UPS units are under load anyway. The last ones fill up expensively, because each needs more cooling than the one before. That is exactly why Nazar was not selling the last row — the instinct was right, there were just no figures under it.

So sales are not merely weaker than the tariff. They also spoil the very multiplier the whole economics stands on.

What was left was deciding how to put this to clients. A price rise of 27.9% looks like a price rise of 27.9%, however well founded it is.

What Nazar did

Clients do not read PUE, they read the price

Raising electricity from 6.1 ₴ to 7.8 ₴ in one move meant handing clients a reason to go looking for another hall. Nazar did it differently: he rewrote the contract so that the price of a kilowatt-hour is tied to the purchase price and multiplied by the coefficient of the hall, which he publishes every quarter.

The conversation with clients turned out easier than he had feared. Not one of them argued that cooling costs money too — everybody understood that, nobody had ever put the question that way before. Two asked to see the readings at the feed, and Nazar showed them.

A formula instead of a figure. The client sees what he is paying for, and sees that the hall does not earn on electricity.

• price per kWh = purchase price × coefficient of the hall, updated every quarter
• the coefficient is published together with the readings at the feed
• rent for the space stayed at 9 000 ₴ and is now genuinely a contribution
• every new client is costed with his own share of the overhead kilowatts

Two clients out of twenty-nine left. The rest stayed, because a formula is more honest than a figure taken off the ceiling, and that is visible from the very first invoice.

The remaining twenty-seven stayed where they were. One of them, the largest in the hall, asked to see the calculation himself and then put the same formula into his own contracts with his own clients.

The money does not disappear — you simply do not see it

“The funniest part is that I optimised nothing. I did not change a single machine, did not re-lay a single cable, did not let anybody go. I just started multiplying the purchase price by the coefficient the engineers had been quoting me for five years running.”

There is one more thing worth saying. The coefficient of 1.5 is no longer a technical characteristic but a number the owner answers for: every tenth of it is hryvnias in the tariff for the client. When the hall becomes more efficient, the price for the client will fall by itself, and that is no longer a gesture of goodwill but a formula.

Nazar's hall has not grown bigger and has not become more efficient. There are still 40 racks, 29 of them let, the coefficient is the same 1.5. The only thing that changed is that the price for the client is now derived from the figure at the feed, and not from a feeling about how much the market will bear.

How to see this in your own hall

All the arithmetic above rests on one comparison that appears on no invoice: how many kilowatt-hours pass through the feed, and how many of them were priced. While power sits in the books as one monthly sum and billing follows the clients' own meters, a 1 511 538 ₴ hole looks like an ordinary utility bill. When purchase, resale and hall occupancy come together in one place on their own, an underpriced tariff is visible in the second month — which is exactly why owners of small and medium businesses use Finmap: to see the financial picture instead of guessing at it.

Read also

• Greenhouse growing: cost per kilo, not kilos per square metre
• Coworking Profit: Per Seat and Occupancy, Not a Full Room

Topic foundation: Management Accounting Explained: What It Is and Why Owners Need It