Predictable Infrastructure Costs for Agencies and MSPs

Most writing about cloud cost predictability is aimed at the company paying the invoice. Agencies and MSPs have a harder version of the problem, because they don't absorb the variance themselves. Agency infrastructure costs and MSP infrastructure costs behave less like an operating expense and more like a bet placed months before the results arrive. Whoever your MSP cloud provider is, the exposure sits with you. They give a client a price, and the difference between that price and the actual cost is their margin.

That changes the form of the problem entirely. For the end customer, a variable invoice is a nuisance and a forecasting challenge. For an agency or MSP, it is a direct transfer from margin to provider, discovered a month after the work was done and the price agreed.

This article is about that version. I run marketing at InMotion Cloud, so treat the last section as what it is. The rest holds regardless of who you buy from.

The quoting problem

The primary issue is a timing mismatch. You quote before you know.

A client asks how much it will cost to host and manage their environment. You must answer now, in a proposal, against others who are also answering now. The true cost will be determined over the following months by traffic you cannot forecast, data transfer driven by end-user behavior, storage that accrues, and support incidents that come when they come.

Three options follow, and each costs you something. Quote low and win the work at a margin that erodes as usage grows. Quote high with a buffer and lose to someone who didn't. Or quote with pass-through variability, which moves your problem onto the client and turns every invoice into a conversation.

Most agencies wind up with some form of the second or third, and the choice quietly defines their MSP pricing model. The buffer is money left on the table when usage remains low, and it is the most common place agency hosting margin goes to die. MSP margin erodes the same way, one quiet month at a time. The pass-through is billing friction and client dissatisfaction that shows up as churn rather than as a line item, which makes it the more expensive of the two and the harder one to see.

This is why managed service provider pricing tends to look different from the pricing of the businesses those providers serve. You are not pricing infrastructure. You are pricing your own exposure to someone else's usage.

Where the variance really comes from

Worth being specific, because the sources are not equally controllable.

Compute is the most predictable part. You size an environment, and unless the client's traffic changes materially, it stays roughly where you put it. Most agencies already do a good job of estimating this.

Data transfer is the most uncontrollable and by far the biggest source of unpleasant surprises. End-user behavior drives egress, and your architecture decisions barely touch it. A client marketing campaign. A large file becoming popular. A mobile app update pulling assets. An integration partner polling an API more frequently than expected. None of that is in your sizing exercise, and all of it is on the invoice.

The scale is not trivial. InMotion Cloud's cost comparison calculator, working from published list rates of $0.085/GB for the workload sizes it compares, models egress at approximately eight percent of a hyperscaler monthly total at typical usage. That number follows the client's traffic profile, which is the whole point. It is the line item you have the least ability to predict at quoting time.

Storage growth is slower but compounds. Backups pile up, snapshots stick around, logs last longer than anyone planned. Any single month looks unremarkable. The year does not.

Support is where most agency models break, because the cost is usually absorbed by your own team members instead of being billed. Support teams carry it as hours. If a client environment has a bad month, the cost is your engineers' hours, and those hours were priced into the retainer at an assumed incident rate. Nobody revisits that assumption per client. A difficult environment can run well above the incident rate you modeled and stay invisible, because it surfaces as a busy team rather than as an overrun on that account. It hits the bottom line without ever appearing on a statement. The clients who consume the most support are rarely the ones paying the most.

There is a second-order version of this worth naming. Response times you promise a client are a cost you have agreed to before you know what will trigger it. A provider whose own technical support is slow pushes that cost onto your team, because your engineers absorb the gap between what you promised and what your vendor delivers.

The management cost nobody quotes for

There is a second cost that never shows up in an infrastructure comparison, and for agencies it is often larger than the infrastructure itself.

Client hosting management is work, and someone on your team is doing it. Patching, monitoring, capacity checks, responding in real time when something breaks at a bad hour. Customer support for your clients is downstream of technical support from your provider — if theirs is slow, yours is too, and the gap is absorbed by your team. If you handle managing hosting in-house across a set of customers, you are carrying a staffing cost that scales with customer count and shows up on no per-client infrastructure quote. It comes straight out of the revenue streams those clients represent.

The number the calculator puts on this is worth looking at directly. For a team that does not manage its own infrastructure, it models the DevOps overhead absorbed when running on a hyperscaler at $2,000 per month, around thirty percent of the modeled total. For a team with its own dedicated DevOps function, the modeled comparison total falls considerably.

That fall is worth reading carefully, because the work has not gone anywhere. It has moved from an invoice to payroll. Agency pricing models often bake in the first and quietly absorb the second, which is why portfolio margin can look good per client and disappointing in aggregate.

What a fixed monthly model actually changes

Converting variable cost to committed cost does one thing. It moves the variance from your margin to the provider.

That is the point, and it is worth being specific about it instead of just saying the arrangement is cheaper. Sometimes it is, sometimes it isn't. What it does reliably is let you quote a client knowing the number you quoted is the number you will pay.

When this does not help you

Several situations where fixed monthly infrastructure is the wrong answer for an agency, and they are worth checking before you assume otherwise.

If your clients are small and their environments are small, the overhead of a managed private cloud commitment is more than what you are saving. There is a point below which it is not worth bothering.

If your client work is project-based and short-term rather than ongoing retained hosting, you want the ability to spin environments up and down without a standing commitment. Public cloud fits that mold better.

If your main motion is reselling at margin — white label cloud hosting with your own billing layer on top — that is a different commercial arrangement from a managed engagement and should be negotiated as one. The reseller hosting vs managed hosting distinction matters more than it looks: one is a product you mark up, the other is a service you deliver.

And if your clients genuinely have spiky, unpredictable traffic, the elasticity is doing real work. Fixing the cost means either provisioning for peak or accepting that peaks will need handling some other way.

How to check your own numbers

Three things to do before changing anything.

Take your three largest client environments and pull twelve months of actual infrastructure cost against what you quoted. The gap, positive or negative, is the variance you are currently carrying. Most agencies have never done this per client, and the distribution is usually far more uneven than you would expect.

Break every invoice down into compute, storage, data transfer, and support. Data transfer is the line to watch, because your clients' end users drive it and you don't.

Estimate the internal hours spent supporting client environments across a quarter and price them at your loaded rate. That number belongs in any comparison. Leaving it out is the most common reason a cost model looks good on paper and doesn't match the accounts.

If the variance is low and your overhead is low, the current setup is working and you should leave it alone. If either number feels uncomfortable, that is what to fix, and fixing the variance is a different exercise from fixing the total.

What we sell, since I said I'd be plain about it

InMotion Cloud plan pricing is $399 per month for Explorer, $1,349 for Small, $2,669 for Medium, and $7,049 for Large. These are flat rates, not starting points, which is what makes them cost effective to quote against — you are not buffering against a number you cannot see. You can start a client proposal with a number you already know.

All of it runs on dedicated hardware in InMotion-owned data centers rather than resold capacity. Flex Spend Advantage adds a pooled model. You commit to a monthly spend and rebalance resources across workloads as they change, instead of holding a fixed allocation for each environment. For an agency portfolio, that maps to how the work actually behaves: one client's project ramps while another's quiets down, and the total stays fairly stable even though the distribution doesn't. This is the part that is specifically an agency cloud hosting product rather than a hosting plan sold to agencies. The same applies to cloud hosting for MSPs. In MSP cloud hosting, a book of client environments behaves differently from one environment of your own, and multi tenant client hosting compounds the difference.

Two things worth knowing before you evaluate any of this, whether you run a digital agency hosting a handful of client sites or a web agency hosting dozens. Managed hosting for agencies only pays off if the management is real — ask any provider what their engineers actually own versus what stays with your team, because the answer varies more than the marketing does. And the limitation, stated on our own page: this is not a promise of unlimited use. It is a committed spend with flexible allocation inside it. If your portfolio grows significantly, the commitment gets reviewed. What it ends is the month-to-month surprise, not the need to plan.