Cloud or On-Prem, What Matters Economically
7 min read
The cloud vs. on-prem debate has been running on two tracks for fifteen years. At the top, architects talk about latency, scaling, and microservices. Below, management quietly crunches the numbers-only to find the AWS bill 30 percent over the business case by year three. By 2026, the top layer will vanish from most mid-market decisions. What remains is a TCO question with three or four honestly calculated variables.
Key Takeaways
- Cloud vs. on-prem is a cost-curve decision, not a stack decision: If you’re reducing the question to containers, Kubernetes, or VMs, you haven’t asked it seriously yet. The real dividing lines lie in acquisition, personnel, growth, and regulatory costs-not architecture.
- Cloud margins vanish for predictable workloads: AWS, Azure, and GCP hold the pricing edge only when demand is unpredictable or highly variable. For stable mid-market workloads, the advantage often flips in favor of on-prem or colocation between month 24 and month 36.
- NIS2 and DORA shift the decision-but rarely in the way everyone assumes: Compliance isn’t pushing mid-market firms into the cloud by default. In regulated sectors, on-prem will become the default option again by 2026, as data residency and auditability neutralize hyperscaler arguments.
Related:M&A rarely fails over price / S/4HANA migration: before the decision
Cloud vs. on-prem is a TCO question, not an architecture question
Anyone sitting in mid-market executive meetings who approved a cloud migration in the past twelve months knows the pattern. The initial business case projects double-digit savings at 24 months. By month 36, that shrinks to single digits-often with the sign flipped. Hyperscaler discounts evaporate, egress fees emerge, and the tool stack multiplies.
The most common misconception in these calculations is that the cloud variable somehow accounts for personnel costs. It doesn’t. A modern cloud setup requires cloud engineers, FinOps specialists, and security experts who, in the DACH region by 2026, will cost between 95,000 and 140,000 euros in annual gross salary. Companies historically running on-prem with two sysadmins rarely get by with fewer than four heads in a cloud model.
The Four Cost Curves No One Honestly Calculates
A robust TCO calculation separates four curves that often merge into one in standard business cases. This blending is the main reason why migration decisions later turn out to be miscalculations.
- Upfront Costs. On-premises has them as a lump sum, while the cloud spreads them over the contract term. If you face capex constraints (bank covenants, investment budget pressure), the cloud becomes the default choice-regardless of the tech stack.
- Staffing Costs. The cloud doesn’t reduce sysadmin workloads-it shifts them into specialized roles with higher market rates. A three-person on-prem team costs less than one cloud engineer, a backup cloud engineer, and a FinOps lead combined.
- Growth Costs. For highly unpredictable workloads (seasonal spikes, product launches, peak loads), the cloud has a structural advantage. But for stable, predictable demand, on-premises becomes cheaper with each passing year as hardware depreciation outpaces cloud OpEx.
- Exit Costs. The most underestimated curve. Migrating to the cloud is cheaper than migrating away from it. Egress fees, refactoring for proprietary services, new license negotiations-exiting an 18-month AWS setup can cost 1.5 to 3 times the original migration effort. Yet this is never factored into the business case.
Speed: Why the Phrase “18-Month Migration” Means Nothing
Migration timelines are the most abused figure in cloud vs. on-prem debates. An “18-month migration” could mean 18 months of technical execution with a fully functional setup at the end-or 18 months of parallel operation with escalating double-license costs plus another 12 months of stabilization.
In the migration projects I’ve observed among mid-sized DACH companies, the difference was rarely technical. It almost always came down to the application portfolio. Migrating 40 to 80 SaaS apps plus 15 legacy custom applications is a completely different beast than moving three core apps and an ERP system.
The honest number isn’t the migration duration-it’s the stabilization horizon. When does the new setup run without daily engineer intervention, without new tickets, without performance surprises? If you don’t factor this into your decision matrix, you’re comparing apples to more expensive apples.
Regulation in 2026: NIS2 and DORA Reshape the Decision
For years, hyperscalers’ key arguments were security, compliance, and auditability. But by 2026, this narrative will be on shaky ground in three sectors: banking, critical infrastructure, and healthcare.
NIS2 requires verifiable supply chain control and access audits. DORA mandates threat-led penetration testing and reproducible recovery tests for financial institutions. Both are harder to achieve in a hyperscaler model than in on-prem setups with clearly documented ownership chains-not because the cloud is less secure, but because the audit burden is distributed differently.
The DORA stress tests that German banks are currently failing serve as an early warning. If you list a cloud provider as a critical third-party vendor, you must meet ICT-TPRM requirements-and those have tightened in 2026. For many mid-sized banks, returning to on-prem isn’t an anti-cloud reflex-it’s compliance-driven pragmatism.
Three Questions Executives Must Answer Before Making the Call
Anyone tackling the cloud vs. on-prem question in 2026 honestly can’t avoid these three questions. They’re uncomfortable because they challenge the CFO’s office more than IT.
- How predictable is my workload over the next 36 months? If you have 80 percent predictability, on-prem or colocation is worth serious consideration. If you’re at 50 percent or below, the cloud is structurally the better fit, as its flexibility justifies the premium.
- What regulatory obligations will I face in 2027 and 2028? If you fall under CRA, NIS2, DORA, or industry-specific regulations, you should explicitly calculate compliance costs for both models. In regulated sectors, the cloud premium in 2026 will no longer be affordable for many mid-sized companies, as audit readiness becomes more expensive than the hyperscaler’s scalability advantage.
- What will my exit cost after 24 months if the chosen model fails? If you can’t answer this, you haven’t thought the decision through. Both cloud exits and on-prem decommissions are costly. Those who factor exit clauses into the business case make different choices than those who only calculate the entry path.
In executive meetings where these three questions were answered honestly, the cloud-first assumption often shifted- not toward anti-cloud decisions, but toward more nuanced hybrid setups with clearly separated workload profiles.
Frequently Asked Questions
Are pure on-prem setups still economically justifiable in 2026?
Yes, for mid-sized companies with stable, predictable workloads and in-house sysadmin teams. The TCO calculation tips in favor of on-prem between months 24 and 36, once cloud discounts expire and personnel costs become the deciding factor. Anti-cloud reflexes rarely drive the decision-it’s usually just cold, hard math.
When is hybrid the right model, and when is it just an expensive compromise?
Hybrid works when workloads are clearly separated by demand profile: core, stable apps on-prem, and peak or seasonal loads in the cloud. It fails when the model emerges as a political compromise between cloud-first executives and on-prem skeptics. In those cases, it doubles complexity without delivering load advantages-and costs more than either pure approach.
What are the exit costs from a typical AWS setup after 18 months?
Realistically, between 1.5 and 3 times the original migration costs. Key expenses include egress fees, refactoring efforts for proprietary services (Lambda, DynamoDB, SQS), and renegotiating licenses for software that can’t be transferred from cloud to on-prem. Mid-sized companies almost always underestimate this.
Will NIS2 really push mid-sized companies out of the cloud and back to on-prem?
In regulated industries, yes-hardly at all in unregulated ones. Banks, healthcare providers, and critical infrastructure operators are recalculating audit burdens in 2026, as NIS2 and sector-specific regulations tighten supply chain obligations. For unregulated mid-sized companies, the cloud generally remains the more efficient option.
Can the cloud vs. on-prem decision be delegated, or does it need to be made at the C-level?
It’s a C-level decision. IT can evaluate architectures and stacks, but cost projections over 36 months, regulatory implications, and exit risks belong in the boardroom. Delegating it to IT results in a stack decision rather than a business decision-that’s the most common structural mistake in these projects.
Further Reading
- S/4HANA Migration: Mid-market Companies Face 2026 Deadline
- AI in Accounting: 78 Percent of Mid-market Firms Still Rely on Blind Posting
- The DORA Stress Tests German Banks Are Currently Failing
More from the MBF Media Network
Source header image: Pexels
