I saw something recently that I think growing mid-market companies should be paying attention to. ComfortDelGro is moving some of its production Kubernetes workloads out of the public cloud and back into private infrastructure. It also expects much of its AI workload to eventually run there. Compute and data sovereignty is coming back in fashion.

This isn’t the first company to do it, and it certainly won’t be the last. In fact, I think cloud repatriation is going to become a much bigger technology conversation over the next few years. I don’t think we were wrong about the cloud or because everyone is about to start rebuilding data centers, but I think it’s because companies are starting to understand their workloads and the economics behind them much better.

Think about how most businesses got here. When you were growing quickly, moving to the cloud made a lot of sense. You didn’t know exactly how much infrastructure you were going to need, and you didn’t want to make large capital investments every time the business changed. The ability to add users, applications and capacity without constantly worrying about the infrastructure underneath it was incredibly valuable.

But something changes as a company grows. That unpredictable 100-person business becomes a 500-person business. The application you weren’t sure anyone would use becomes critical to the operation. Your storage needs aren’t much of a mystery anymore, and the amount of compute you use every month becomes fairly consistent. You suddenly have something you didn’t have five years ago: predictability.

And predictability is an opportunity to control expense.

You’re probably not going to build your own hyperscale infrastructure, and you probably don’t have a team of people constantly modeling where every workload should run. You shouldn’t need one. But you are reaching a size where technology decisions made when you were smaller can become very expensive decisions when you’re bigger.

AI is an accelerant, and I’m a huge fan of experimentation. If you have 30 people trying an AI tool, I don’t want you building infrastructure around something you don’t even know will work. Consumption-based cloud services are almost perfect for this stage because you can experiment, learn and move quickly without making a large bet upfront.

Now imagine that experiment actually works. Thirty users become 300. One AI workflow becomes 20. Your employees start using agents throughout the day, token consumption climbs and data starts moving between more systems. What started as a relatively insignificant monthly expense eventually becomes large enough that it is making a huge impact on your P&L.

At that point, I wouldn’t immediately ask whether we should get out of the cloud. I’d ask something more useful: Are we still paying for flexibility that we actually need?

I don’t know what the answer will be for your business, and that’s really the point. You may run the numbers and find that Azure or AWS is still easily the best answer. You might find that one large, predictable workload would be dramatically cheaper somewhere else. You may end up with a hybrid environment where different workloads live in different places for very good reasons.

What I would start thinking about now is portability.

As you’re building the next application, deploying AI, selecting platforms or signing longer-term cloud commitments, spend some time thinking about what would happen if the economics changed. Can you easily move your data? Are you designing around proprietary services that would be difficult to replace? If the workload becomes ten times larger, do the economics still work? If you discovered three years from now that there was a much better place to run it, how difficult would it actually be to move?

I’m not trying to keep you out of the cloud, but these questions are designed to keep today’s decision from eliminating tomorrow’s options.

So I wouldn’t start pulling workloads out of the cloud because repatriation is becoming a trend. I’d do something much less dramatic. Take your three largest cloud workloads and sit down with whoever leads technology in your organization. Ask them:

“If we had to decide where this should run today, knowing everything we know now, would we make the same decision?”

If the answer is yes, that’s valuable confirmation that you’re still in the right place.

If everyone has to think about it for a while, you may have just found a pretty interesting conversation to have. And, if you’re not sure… you know where to find me!

Until next week,

—Jared

Text Me: 314.806.3912

A fast answer and a coordinated one are not the same thing. When AI resolves a B2B customer issue without looping in the teams who have to deliver on it, you get confident responses nobody actually signed off on.

A new briefing paper from Harvard Business Review Analytic Services, sponsored by Front, examines the coordination gaps that open up when transactional AI tools meet multi-team B2B service, and how leading companies are using AI to close those gaps instead of widening them.

Read the briefing paper for the questions to ask before your next AI investment.

Read the Briefing Paper