It's not a technology question. It's a question of control. I've migrated enterprise environments to the cloud. I've brought others back. I've built hybrids that have been running for years without intervention. The only constant: the right decision depends on who you are, not on what's trendy.
The illusion of simplification
Cloud sells itself as simplification. No more buying hardware, no more managing racks, no more replacing disks at 3 in the morning. That's true. But the simplification is a partial illusion.
What doesn't appear in the sales pitch: monthly invoices that creep up incrementally. Egress charges that nobody budgeted for. Vendor lock-in that turns migrating to another provider into a 6-month project. And a truth that few say out loud: your data is on someone else's servers, governed by someone else's terms, in someone else's jurisdiction.
I'm not saying it's wrong. I'm saying you need to go in with your eyes open.
What I've seen in 15 years of enterprise
I've managed production environments for clients in healthcare, insurance, and finance. Regulated environments, with quarterly audits and contractual penalties for downtime. Here's what I've observed:
Cloud excels at variability. Dev/test environments that spin up in the morning and shut down at night. Projects that last 3 months and then disappear. Burst capacity for campaigns or launches. This is where cloud is unbeatable. You pay for what you use, not what you have.
On-premise excels at predictability. Databases that have been running 24/7 for 5 years. Stable workloads with constant consumption. Core applications that don't fundamentally change from one quarter to the next. Here, a dedicated server pays for itself in 18-24 months. After that, the marginal cost is nearly zero. In the cloud, you pay the same in month 60 as in month 1.
Hybrid isn't a compromise. It's an architecture. The best-performing setup I've built used on-premise for the core (databases, critical applications, sensitive data) and cloud for the periphery (CDN, analytics, DR site, overflow compute). The total cost was 40% lower than full-cloud, with more control and the same availability.
The question nobody asks
Everyone asks "cloud or on-premise?" Nobody asks: "Who controls the exit?"
If tomorrow you decide to leave your current provider, how long does it take? How many systems need to be rewritten? How many dependencies are tied to proprietary services that have no equivalent elsewhere?
I've seen companies locked into a single cloud provider because they built everything on provider-specific managed services. Lambda functions, DynamoDB, Cloud Spanner. Every "simplification" decision was, in fact, a coupling decision. And coupling is paid at exit, not at entry.
Our rule: any architecture must allow migration within 90 days maximum. If it doesn't, it's not an architecture. It's captivity with a monthly invoice.
The real cost: 5-year TCO
I've done the TCO exercise dozens of times for different clients. The result is almost always the same:
In year 1, cloud looks cheaper. No CAPEX, rapid deployment, everything works. By year 3, costs equalize. By year 5, dedicated on-premise is 30-50% cheaper for stable workloads. The numbers vary, but the direction doesn't.
The exception: if your workloads are genuinely elastic (10x variation between minimum and maximum), cloud remains more efficient long-term. But that's rarer than most people think.
Data sovereignty is not optional
NIS2, DORA, GDPR, sector-specific regulations. The world is moving in one direction: more control over data, not less. I've worked in eHealth, where a database in the wrong jurisdiction means non-compliance. I've seen audits that demanded physical proof of data localization.
Cloud can be compliant. But compliance in the cloud requires explicit effort: choosing the right region, customer-managed encryption, specific data processing agreements. It doesn't come by default. On-premise offers compliance by design: the data is where you can see it.
What we recommend
We don't recommend cloud. We don't recommend on-premise. We recommend understanding what you own and what you rent. Knowing where your data is, who controls it, and how much it costs to move it.
If you have stable workloads, sensitive data, and a 3+ year horizon, it's worth investing in your own infrastructure. If you need agility, variability, and deployment speed, cloud is the right choice. If you have both, you build a hybrid with clear boundaries.
The only wrong decision is one made without data. Let's talk and analyze it together.