When a vendor moves the floor overnight

The shock

The whole business rode on one vendor's price

A game-hosting business comes down to one number. What does a gigabyte of RAM cost per month? That gigabyte is the thing you are really reselling.

So the vendor underneath you matters more than it looks. This one raised its base prices overnight, and it was not the first hike that year. The cheap machines started selling out too. That was no rounding error. It reset what the whole business cost to run.

One vendor had quietly become a single point of failure for the money. It could break your costs and your supply at the same time. The business was priced against one company's list, and that list had stopped standing still.

The part that's easy to misread

The hibernation did its job

You could read this as "the cost feature didn't save enough." It is the opposite.

The hibernation I built did exactly what it was meant to. It drops the idle CPU first. Then it moves idle memory off to disk while the server keeps running. On the platform's own numbers, a server still holding its idle memory clears around 58% margin. Hand that memory back to the pool and the same server climbs toward 86%. That side was already handled.

A vendor raising its hardware prices is a different problem, and it splits into two questions. How much memory do you waste on each machine? And what does the machine itself cost? You can get the first one right and still get hit by the second. Keeping the two apart is the only reason the next conversation stayed clear-headed instead of panicked.

The method

Check the price at the source

The reflex under a price shock is to pull up a comparison site and grab the cheapest row. Those sites were wrong. Stale prices, phantom stock, regions with no real capacity.

So I went to the source instead. I checked every provider's per-GB price on its own pages. I converted currencies on a fixed date, so a euro plan and a dollar plan were honestly comparable. I counted DDoS protection as a real cost line, because for game servers it is not optional.

The cheap EU and US regions all sit near the same floor. Somewhere like Australia runs two to three times that. You price that premium in on purpose. You do not want to discover it after you have promised a tier there.

The answer was not a different single vendor. It was to stop depending on any one of them.

That means three things at once: a primary provider, a backup ready to take over, and cloud capacity to soak up the spikes. With that in place, the next overnight hike is an annoyance instead of an emergency.

The exact prices in that analysis are already out of date. That was always going to happen, and that is fine. The point was never the numbers. It was not being tied to one company in the first place.

The lesson

The same math, on your AI bill

Two things carry out of this. The first is about vendors. Leaning on a single one is a risk whether or not you have ever priced it, and the time to build a way out is while the price is still holding.

The second is about which cost to chase. Every business has one resource that actually caps it, the thing you run out of first. Call that the binding cost. Find yours, and engineer against it rather than against the dashboard total.

It is the same muscle that protects an AI product. There the binding cost is inference. The questions do not change: what does a unit of work really cost when you measure it at the source instead of the invoice, where is the waste, and what happens to your margins the day a provider reprices. Cost work belongs in the system, wired into how the thing runs day to day.

If your bill is already the thing that hurts, that is where I start. A triage costs EUR 1,500, takes 48 hours, and the fee counts toward the fix if you go ahead.