In short
Cloud migration business cases count servers and licences and leave out the costs that decide the result. Five recur: egress, a missing FinOps function, lift-and-shift sizing, committed-spend lock-in and data residency.
In 2021, Andreessen Horowitz estimated that cloud costs were taking roughly $100 billion of market value from the top fifty public software companies that use it, and called this the "trillion-dollar paradox". Companies that moved to the cloud to save money meet a smaller version of it. Their business cases missed the same five items.
A migration ROI model counts infrastructure line items. The costs that decide the outcome sit in operations, people, architecture and compliance, and they show up after the migration is declared complete. The gap comes from what the model is built to count, so it repeats from one company to the next.
Data moves into a cloud for free. Moving it out, between regions or to another provider is charged by volume. Backups, analytics extracts, disaster-recovery replication and traffic between services all add to it. A Gartner report found that most of its clients spend 10 to 15% of their cloud bill on egress, a line the original business case seldom models at realistic volumes.
The larger cost is strategic. In a 2025 Dimensional Research survey of 403 IT leaders, commissioned by Backblaze, 55% named egress and data transfer the biggest barrier to switching storage providers. Once enough data sits with one provider, leaving it costs money. That is what a16z called data gravity.
The business case treats cloud skills as a one-time cost of retraining and a few hires. The provider ships changes faster than an internal team learns them, so the cost keeps recurring.
FinOps, the practice that assigns cloud costs to business units, enforces tagging, rightsizes workloads and negotiates commitments, rarely exists when a migration starts. It tends to appear after the first budget overrun. Flexera's 2024 State of the Cloud Report ranks managing cloud spend as the top cloud challenge for the second year running, ahead of security. Without an owner, spend grows faster than the workload.
Most enterprise migrations are lift-and-shift, because of deadlines and the risk of rewriting production systems. The business case assumes re-architected workloads that scale with demand and use managed services. In practice, monoliths move to cloud virtual machines with few changes and run on instances sized for peak load, at cloud prices.
Gartner notes that most lift-and-shift migrations fall short of the business outcomes expected of them, and that elasticity depends on re-architecture work the original migration seldom funds. The deck priced the elastic end state. The invoice reflects the interim state, which often becomes permanent.
Reserved instances, savings plans and committed-use contracts make cloud pricing competitive, and all of them require multi-year commitments to a provider, an instance family and sometimes a region. The discounts on on-demand prices are real. So is the lock-in, the same lock-in the migration was meant to leave behind.
If a workload later needs another instance type, region or provider, the committed spend becomes a sunk cost that bends the decision. Business cases count the discount as a saving and leave out the constraint.
Kazakhstan's personal data law requires personal data to be stored in databases inside the country, and AWS, Azure and Google Cloud have no region there. A company holding regulated data cannot run on public cloud alone. It runs a hybrid: regulated data in local data centres or with local cloud providers, other workloads in international regions, and secure links between them.
A hybrid has its own costs: dedicated lines or VPN, two control planes, two security models, and latency on every workflow that crosses the boundary. Business cases built on a Western template assume a local region and no residency rule, so none of this appears in them.
The usual reply is that each of these can be fixed. Egress can be designed down, FinOps can be built, lift-and-shift can be followed by re-architecture, and a hybrid can be designed well.
Each fix has a price. Architecture reduces egress and leaves some of it. FinOps matures slowly, and the learning is paid for in overruns. Re-architecture is a second migration, often larger than the first, and few organisations find the budget for it once the first is declared done. The discount requires the commitment. In Kazakhstan, residency is set by law. Together these describe how cloud economics work at enterprise scale.
Before signing a migration business case, ask for four things. A three-year cost of ownership that includes egress at realistic volumes, hybrid connectivity, FinOps people and tools, and committed spend as both a saving and a constraint. A FinOps owner and operating model agreed before the move. A decision per workload to lift-and-shift, re-platform, re-architect or stay on-premises, with the re-architecture funded in the same case. And in Kazakhstan, a residency decision at the start: which data stays in the country and what the hybrid costs to run.
The cloud remains the right place for most enterprise workloads. These four items make the business case hold up when the invoices arrive.
Flexera's 2024 State of the Cloud Report found public cloud spend running over budget by 15% on average. The usual drivers are egress that was never modelled, instances oversized during lift-and-shift, and no FinOps owner enforcing tagging, rightsizing and commitments.
Flexera's 2024 report estimates that about 27% of IaaS and PaaS spend is wasted on idle compute, oversized instances, orphaned storage and commitments that do not match use. The share has fallen from 32% in 2022 as FinOps practice spreads.
The law requires personal data to be stored in databases inside Kazakhstan, and no major hyperscaler runs a region there. Compliance therefore means a local cloud provider or a hybrid that keeps regulated data in local data centres. Both add connectivity, operations and latency costs that the standard model leaves out, so build them in from day one.
No. These are reasons to build an honest business case and an operating model around it. The cloud still gives elasticity for variable workloads, managed services, faster provisioning and a catalogue that keeps growing. The argument here is with ROI models that understate the cost.
A migration model that survives the first year of invoices counts egress, hybrid connectivity, committed spend and the people who own the cost. opengate designs and builds cloud and hybrid architecture together with the cost model behind it, and keeps regulated data where it has to stay. Start with one workload, and we measure what it costs before and after the move.
