What a mobile proxy costs to run, line by line

About fifteen dollars a month. That is what it costs to keep one Singapore mobile line alive on my rack before a customer has connected to it, before anyone has been paid, before the line has earned a cent.

I get asked often enough that publishing the model beats defending it one message at a time. Every figure below is off my own cost sheet, and where I give a range the number moves month to month.

The bill for one line

Add the electricity and you land near fifteen dollars a month per line. Against what a line sells for, that puts gross margin somewhere in the seventy to seventy five percent band, which is the figure people quote back at me when they have decided this business prints money.

Hold that number.

The SIM is the floor

Ten dollars a month, per line, forever, and there is nothing clever I can do about it. Scale barely helps. A carrier does not hand me a better rate because I hold a lot of lines; I am a retail data customer with a lot of accounts.

That one number is why this market has a hard bottom. If someone sells you a genuine Singapore mobile line for less than the SIM behind it costs, either the SIM is not real or they are buying market share with their own money. Both of those end the same way for the person who bought the line.

Modems are a consumable, and heat sets the burn rate

Each line needs a modem, and modems die. They run continuously in a warm rack in a tropical climate. The honest way to cost one is the purchase price divided by the months it actually survives, not the price on the invoice.

Across my fleet that lands at roughly a dollar fifty per line per month once replacements are spread over the units that keep working.

Airflow moves that number directly. Shelves sitting where air actually moves have modems lasting noticeably longer than shelves in a dead pocket, and I learned that from the replacement rate rather than any manual. Ventilation is a cost control. Electricity is the smaller half of the heat story; the depreciation it drives is the bigger half.

It took me a while to accept that a fraction of the hardware is always dying. Failures stopped feeling like events the month I put them in the spreadsheet as a recurring line item, which is what they had been the whole time.

The costs that only work when you divide them

You cannot plug a dozen modems into a machine and expect them to stay there. They need properly rated powered hubs, and this is where a rack quietly falls apart.

I found that out expensively. A hub that could not supply enough current dropped an entire shelf off the bus at once and the capacity simply stopped existing in software. Every light on the front was still on. It read like a provisioning bug, and I spent an embarrassing stretch of time debugging code before suspecting the power.

Call it $150 of hub covering around thirty ports. Five dollars a port, once, for the component most likely to take a dozen customers offline while looking like your own software failing.

The server is the same shape of cost with a friendlier curve. Two hundred dollars buys the machine that manages the ports and passes the traffic. Across a properly loaded rack that is a few dollars per line per month. Across four modems it is ruinous.

Small deployments have terrible economics. Fixed costs are genuinely fixed, and they only stop hurting when you divide them by a decent number.

What the margin actually has to absorb

Seventy five percent sounds enormous until you list what comes out of it.

It pays for the lines nobody is renting. A fleet is never fully sold, and an idle line still burns its SIM and its share of the power. Take a hundred lines with thirty unsold and the effective margin drops by about a third before anything else goes wrong.

It pays for spare capacity that exists purely to be unsold. When hardware dies at an awkward hour I need somewhere to move that customer immediately, ideally on the same carrier so nothing about their setup changes. I keep a batch of cards reserved for exactly that, earning nothing, doing their job by being available.

It pays for failures at two in the morning. Somebody has to notice, and something has to move the customer onto working hardware before they write in. I have automation for that now. Building it cost weeks.

It pays for data overage. Plans have ceilings, and a customer quietly pulling seven hundred gigabytes on a plan sized for two hundred is a real cost that reaches me long before it reaches them.

And it pays for the hours. Diagnosis, swaps, carrier admin, the nights when the rack is doing something inexplicable and nobody else will work out why.

The line item with no hardware attached

There is a category of cost that does not exist at four lines and dominates at a hundred. Knowing what is actually true.

Which physical modem is which port. Which port belongs to which customer. Whether the thing every dashboard calls healthy is passing traffic right now. At small numbers you answer all of that by looking at the rack. At scale it is the hardest problem in the business, because three separate systems each hold an opinion about the state of a line and they drift apart quietly.

I have had a modem reported healthy by every check I run while returning 502s to the customer on every single request. I have had ports marked free that were carrying a paying customer, and ports marked occupied that had been idle for weeks. Each of those is either lost revenue or a support conversation, and none of it appears anywhere in a hardware cost.

So there is now a reconciler on a schedule forcing those systems to agree, plus watchdogs that catch a dead line before a customer does. Most of my time has gone into that, and it is invisible in any price comparison.

What a price under the floor means

If a genuine line costs fifteen dollars a month before anyone is paid, a service priced well under that is doing something else to make the numbers work.

Usually the address is shared with far more customers than you were told, which makes it neither exclusive nor sticky. Sometimes it is a datacenter address dressed up as mobile, which any competent ASN check catches in one request. Occasionally it is somebody’s home connection resold without their knowledge, which is a problem you inherit.

None of that is a moral argument. It is arithmetic. Real hardware and real SIMs have a floor, and anyone selling below it has removed one of those two things.

The carrier is a supplier I do not control

Nothing about a carrier relationship appears in a cost breakdown, and it shapes everything anyway.

Every line depends on a telco with its own rules about what a data plan is for, and those rules change without anyone consulting me. So part of what this business is doing is holding multiple Singapore networks at once, on separate accounts, so a policy change at one does not take out the whole fleet. That diversity costs me. More accounts, more admin, more hardware quirks to learn.

It also means a customer asking for one specific network is asking for something genuinely constrained. New cards on some carriers take days to activate. That lead time is a real limit rather than a stalling tactic. If a supplier can hand over unlimited quantity on any network you name, instantly, ask where those cards came from.

What to ask before you hand anyone money

Ask what happens the day your line dies, because it will. The answer you want is a swap onto equivalent hardware, quickly, with no argument about whose fault it was.

Ask whether the address is exclusive to you, and if it is shared, how many others are on it. That drives cost and quality more than anything else here, and a vague answer is itself the answer.

Check your actual data usage before you pick a plan rather than the number you assume. I have had customers on a two hundred gigabyte plan running three times that, and it always ends in a conversation neither of us enjoys.

What I got wrong

For a long stretch I priced by looking at what competitors charged and sitting slightly underneath. That is how most small operators price. It is backwards.

A bad month forced me to actually build the cost model, and the model showed immediately that two of my plans were barely above the cost of the hardware serving them. I had been pushing those plans hardest, because they moved. Of course they moved. They were cheap for a reason I had never sat down and calculated.

If you run anything with hardware underneath it, work out the cost per unit before you write the price list. One boring afternoon, and it tells you which of your popular products the others are quietly funding.

If you want Singapore mobile lines on real carrier SIMs, with the swap process and the reserve capacity behind them, that is what we sell.

Get new guides and videos first — join the Telegram channel.