Cloud computing turned twenty years old just last month. And yet, two decades in, it’s still common for cloud spending to be treated like a data center bill, wrapped in elaborate chargeback formulas, explained away with multi-tab spreadsheets and governed by financial thinking that was built for a different kind of asset altogether.
Many teams can benefit from a shift to a more opex-oriented way of looking at their cloud spending. It’s not just a different set of numbers. It’s a different way of thinking. And to get there, we need to talk about pencils.
A World of Fixed Costs
Imagine you own an old pencil factory. It’s fully self-contained: there’s a graphite mine on the property, a forest full of trees, an old building and some even older pencil-making machines. Every few years you buy a new digger for the graphite or a new logger for the timber. You have a crew of machinery operators, some factory workers plus a financial office and a general manager. The only raw materials you actually pay for are paint for pencils, fuel for the machinery and power for the factory. Everything else is stuff like taxes, insurance, etc.
In this world, almost every cost is fixed. It barely matters whether you cut down one tree or a thousand, it’s basically the same labor bill, the equipment wears out quicker but depreciation is much bigger than any replacement blades and you’re not paying anyone for the wood because, well, it literally grows on trees. The pencil-making machines are ancient, in fact they might not even be made anymore but replacing them is unthinkably expensive, so you put up with their quirks and their breakdowns and make do.
Business is steady. Maybe some local schools grow and shrink. Last year the local stationery store closed but you had started selling on Amazon. Your customer base is mature and predictable, your business is easy to understand and financial planning is straightforward. Even the lumpy capex of a new digger every few years can be plotted out, with a built-in contingency for surprises. Sales can be a little bit more of a challenge. It’s easier to lose a customer than to win one.
Otherwise your P&L consists of your total expenses, divide it by shipments of pencils and that’s the pricing floor. Anything above that is profit. The whole business plan fits on the back of an envelope.
The Big Overseas Order
One day a big overseas customer prospect contacts you. They love your product and are talking about a size of order that could potentially double the size of your business.
In the capex world this is a scary prospect, and probably not advisable. Doubling output means doubling everything: more diggers, more loggers, maybe a second factory line, more crews, all of it paid up front before a single additional shipment. And if the customer walks in two years, you’re going bust.
You have a predictable steady business, and in return, high levels of growth are actually a real risk for your business. Low risk = low growth. You politely decline, and life goes on as before.
Same Business, Different Shape
Now imagine the same business that’s run with an opex mindset.
You don’t dig up the graphite. You don’t chop down the trees. You don’t even own the pencil-making machines. Instead you buy the wood, the graphite and the paint, and you pay a giant pencil factory somewhere else to make pencils for you, which you then market and sell. You focus on customer research, innovation, sales, marketing and distribution. You’re a brand and a customer book, not an industrial empire.
Sounds like a worse business doesn’t it? You’ve given up the romance of the mine and the forest. You’re “just” a middleman. And you can’t possibly make money when you’re paying for stuff that in the other world are both “free”, right?
What if the giant factory and the dedicated graphite supplier can each deliver each piece cheaper (or even at a similar cost) to doing it in-house, because they’re located in a cheaper locale, specialize and/or operate at scale? Your unit cost may actually go down. And even if it doesn’t, your capital needs are much, much smaller. The big overseas order isn’t a bet-the-farm capex decision anymore, it’s just a bigger purchase order. If the customer walks, you stop paying the costs. The risk profile inverts. You seek growth, which before really wasn’t a good idea.
You may have increased your costs but in return you have a new capability: lower risk, higher growth, new products and markets are now open to you. In one way, your business is safer than in the simpler, capex world.
The decision on whether to make it yourself or buy it in is just a financial one: which is more cost-effective. Common sense says owning the inputs has to be cheaper: no middleman markup, no shipping from far away, no profit margin baked in for someone else. And yet, time after time, specialists at scale somewhere on the other side of the planet beat the in-house version. We touched on the comparative advantage logic previously, over here.
This Isn’t About Pencils
This essay isn’t really about whether to outsource the pencil factory. It’s about how the financial worldview has to change when you flip from capex-dominated to opex-dominated. And that flip is not a million miles from the comparison between running your own data center and using the cloud. I’ll leave the “which is better for you” question to the reader. What I want to focus on is just how important it is to shift your thinking.
For instance, when you start buying raw materials instead of producing them yourself, cost is just one variable among many. The cheapest wood isn’t necessarily the best. How does the graphite run on the page? What are the breakage rates in transit? How quickly can you go from order to manufacturing? Does a more expensive paint sell more pencils on the shelf?
Welcome to the world of inputs and outputs.
These things have always mattered, of course. You just didn’t have so much control over them. When you picked your co-lo, your hardware, your network architecture, those decisions locked in your choices for years. And once you’d paid for the gear, the incremental cost of deploying a new feature was effectively zero, which can also affect decision making.
In the new opex world, every feature incurs ongoing cost. Every transaction, every user, every minute it runs. The questions sharpen accordingly. Will users actually value it? Can we charge more? Will it reduce churn? You move from “can we afford to build it once” to “is it worth what it costs us forever”.
Five Golden Rules
Most cloud practitioners understand the shift in principle. But when it comes to actually doing something about it, well, that’s not so easy. So here are five rules to help you get your head around it.
1. Accurate unit costs are essential, and harder than they look. Avoid the trap of using averages whenever you can. Develop real bills of materials (BOM) based on what was actually used and what it actually cost at the moment it was used. If your costs vary by time of day, each data point has to reflect that. Averages don’t just lose information, they can actively send you the wrong way on real decisions.
2. Don’t cross the beams. Try not to mix capex and opex in your unit cost analyses. Anyone who’s been doing this for a while knows the elaborate chargeback rules used to amortize a single big investment across users or business units. That’s a fine financial practice and it has to happen somewhere. But when it bleeds into your input/output analysis the noise of a big dumb chargeback can easily drown the signal of varying opex spending. You end up looking at numbers that are part real consumption and part accounting allocation, and you can’t make a clean decision off that. Keep them in separate buckets.
3. It’s not just about margins. A move that improves margin can quietly cost you somewhere else. It might increase churn. It might reduce your ability to shift posture in response to customer demand. It might change your leverage with suppliers, customers or new entrants. Margin is one variable, not the whole picture.
4. A higher overall bill might not be a bad thing. If you get this right and grow your business, your cloud bill should grow with it. When Toyota sells more cars, it spends more on steel. Nobody sees that as wrong. The number that matters is unit cost, and even unit cost going up can be the right answer: for instance if it meant you took a chunk out of a competitor, or caught a demand spike that would otherwise have been lost. I have seen too many CFOs get stuck demanding to know why cloud spending is up, when in many businesses, cloud has become a raw material for the product and is a sign of strong growth.
5. Even a simple model is better than no model. Try to avoid analysis paralysis, especially at the start. Most teams already have a pre-existing tangle of both capex and opex, and trying to build the perfect model from scratch is how you end up six months in with no answers. Even some cloud systems can be hard to chargeback. I was CEO at one SaaS company where we processed a lot of data. We would never know if a user would actually use the feature that data was powering, and we had to make compromises in assigning these costs. Pull out just the variable parts of your opex, divide by customers, and you’ve already learned something. A rough number you can actually use beats a perfect number you’ll never finish.
Where to From Here
Each of these rules will likely be covered by a future article on its own, and could easily be a sixty-minute talk at a conference or podcast. My intention here was to offer a framework on how to think differently and strip the concepts down to something legible for someone who isn’t already steeped in unit economics. We can shift our mental model away from old habits of complicated chargeback formulas, and move towards a simpler quest, namely the proper accrual of actual costs versus real world impact.
The shift is the point. The depth comes later.
If you’ve ever felt like the financial side of cloud was confusing in ways your old infrastructure never was, this might be why. It’s not that the numbers got harder. It’s that they started asking a different set of questions. And once you hear what they’re really asking, the answers get a whole lot easier to find.
