Data centers and SaaS: why your bill is going up

Your software bill went up again this year and nobody explained why. The CRM raised prices 12%. The email marketing tool launched an "AI plan" that costs twice the old one. Cloud storage adjusted without notice. You look at the spreadsheet and think it is greed. Part of it is. But there is a physical machine behind all of it: the data center running your SaaS needs power, a lot of power, and the cost of that power became the hottest topic in tech.

In the United States, the environmental agency published guidance that makes it easier to license power generation dedicated to data centers. In practice: it gets faster and cheaper to fire up a gas turbine on the lot next to the server warehouse, with fewer environmental requirements along the way. It is a public policy decision that feels far away from your business in Curitiba or Fortaleza. It is not. It moves the cost of the fuel that keeps your SaaS running.

What power has to do with the price of your software

Software feels immaterial. You click, it shows up on screen. But every click is a server spinning in some warehouse, and a server burns electricity twice: once to process, once to cool the heat from processing.

A large data center today pulls between 100 and 500 megawatts. For comparison, a Brazilian city of 300,000 people consumes in that range. The projects announced for 2027 talk in gigawatts, the equivalent of a whole nuclear plant dedicated to serving AI models.

That cost reaches your bill through three paths:

• Direct, when the vendor raises the plan price and cites "infrastructure costs."
• Disguised, when they hold the price and cut your usage limit, those 10,000 emails that became 5,000.
• Bundled, when they create a new tier with AI built in and move you there at renewal.

The third one is the most common right now. And the most expensive.

Why AI changed the math for everyone

Before 2023, the marginal cost of serving a SaaS user was close to zero. You logged in, queried a database, the server answered. Pennies a month.

A query to a large language model costs between 100 and 1000 times more than a regular database query. That is not marketing hyperbole, it is hardware arithmetic: expensive GPU, expensive power, expensive cooling.

Then your CRM vendor put a little "summarize with AI" button on the screen. You use it three times a day. Your 40 employees use it too. Suddenly that product that cost $12 per user has a real variable cost, and someone pays the difference.

When a SaaS gives you AI for free, either it is burning investor cash or it already buried the cost somewhere else.

This is not a conspiracy. It is the normal cycle: subsidy phase to win the market, then price correction. We saw it with Uber, with delivery, with streaming. The difference is that this time the base cost is physical and does not disappear with scale. Power does not get free because you have more customers.

The firm that found out too late

I worked with an accounting firm of 22 people. They used seven SaaS tools. Nobody had the full list, each manager had signed up for their own.

We mapped everything: $4,870 a month. Of those seven, two did the same thing. A third had 18 active licenses for 9 people who still worked there. And the priciest one, a document management platform, had raised prices 34% in 14 months on the grounds of "new artificial intelligence features" that nobody at the firm used.

We cut the duplicates, fixed the licenses, and replaced the expensive platform with a custom indexing and search solution running on their own server. Final monthly cost: $1,640. Development investment: paid back in five months.

The point is not "build everything in house." It is that nobody had looked. The increase went through the corporate card for 14 months straight without a single person questioning it.

How to know if you are exposed

You do not need a formal audit. Set aside an afternoon and answer four questions.

What do you pay today, all in? Pull the last three months of corporate card statements and flag every recurring charge in dollars. You will find things you forgot existed.

How many licenses are active versus people working? This is the easiest fat to cut. In almost every client I looked at, 15% to 30% is waste.

How much of each tool do you actually use? If the team touches 20% of the features, there is a cheaper option. Or there is something you could do better with a custom build.

What happens if the price doubles next year? If the answer is "we pay because we have no choice," you have a dependency problem, not a price problem. And dependency is always more expensive than the subscription.

What you can do without becoming a hostage

I am not a "build it all yourself" fundamentalist. Swapping Gmail for your own mail server is a recipe for headaches and you will not save anything meaningful.

What works is sorting by criticality. Generic tool that the whole market uses and has five equivalent competitors? Stay with the SaaS, negotiate at renewal, threaten to leave for real (it works more than you would think, a 20% discount is common once the retention team joins the conversation).

Now, a tool that carries the core process of your business, with your data, your rules, your way of working? That is where custom software is worth considering. Not just for the monthly cost, but because the tailored version usually does what you need better and does not do what you do not need.

What changed in the last two years is the math. Building something custom used to take six months and a team. Today one developer with good tools ships in weeks what used to take quarters. The equation flipped. An automation that replaces a $1,200 a month subscription pays for itself in under a year and stays yours after that.

And there is the point nobody puts in the spreadsheet: the software you subscribe to can change owners, change price, kill your plan, or simply shut down. It happened to plenty of people who built an entire operation on top of a tool that vanished. What runs on your infrastructure, with your code, does not vanish.

The practical summary

The American agency's decision on power licensing is a signal of where the sector is heading: data center infrastructure became a strategic priority, the cost is high, and somebody has to pay. That somebody is the market that buys SaaS, which means you.

This does not mean panic or cancel everything. It means you stop treating software subscriptions as an invisible fixed expense. It is the second or third largest cost line at a lot of service companies and almost nobody looks at it with the same care they give rent or payroll.

Start with the spreadsheet. Then decide what is commodity and what is yours. And what is yours, make it truly yours.

If you want to talk about what you could automate or bring in house in your case, take a look at how I work.

Your SaaS bill went up again and nobody explained why.

Part of it is greed. But there is physics behind it: every click in your software is a server burning power, and power became the most contested cost in tech.

One AI model query costs 100 to 1000 times more than a regular database query. Someone pays that difference. Guess who.

Last week I opened the spreadsheet of an accounting firm: $4,870 a month across 7 tools. Two did the same thing. One had 18 licenses for 9 people. We closed at $1,640.

The increase sat on the corporate card for 14 months and nobody asked a single question.

Software became the second largest cost line at a lot of service companies, and it is the only one nobody audits.

If you do not know what you pay today, start there. And if you want to talk about what you could bring in house, reach out.

#SaaS #CostManagement #ArtificialIntelligence #Technology #Automation