A cloud bill can look healthy while a single architectural decision quietly raises the cost of leaving. That is vendor lock-in: not one contract clause, but the accumulated weight of data formats, identity systems, managed services, network paths and team habits that make a move expensive. For a Bangladeshi business, the risk is rarely that a provider will disappear. It is that a renewal, a price change or a service limit arrives when switching is least convenient.
This comparison sets out four common operating models and the criteria that decide how much lock-in each one carries. It is written for owners and managers who need a defensible position, not a slogan about being cloud-neutral.
What lock-in actually means
Lock-in is a spectrum, and it sits in several layers at once. A system can be portable at the data layer and stuck at the identity layer. Assess each layer separately.
- Data: proprietary formats, export limits and relationships that break when records leave.
- Compute and managed services: functions, queues, managed databases and AI APIs that have no direct equivalent elsewhere.
- Identity: the directory, single sign-on and role model that everything else trusts.
- Network: private links, reserved addresses and egress charges that make data movement expensive.
- Commercial terms: minimum commitments, renewal discounts and termination assistance.
- Skills: a team that knows one console, one toolchain and one way of debugging.
Four models, compared
Single cloud, fully managed
One provider supplies compute, storage, database, identity and analytics. It is the fastest route to a working system and the cheapest to operate with a small team. The trade-off is concentration: every managed service you adopt becomes a migration dependency. This model fits a young workload, a lean team and a business that values speed over optionality.
Multi-cloud by design
Workloads are split across two providers, often for resilience or procurement reasons. It reduces single-vendor failure, but only when the split follows workload boundaries. Duplicating identity, networking and observability often costs more than the second cloud saves. Multi-cloud suits a small number of independent systems, not one application scattered across providers.
Portable baseline
The core runs on commodity compute and open formats, with managed services used at the edges. Containers, a relational database with standard export, object storage with lifecycle rules and infrastructure defined in code. You give up some convenience and pay in engineering discipline, but the exit path stays credible. This is the usual choice for a business that expects to change providers or run part of the estate itself.
Hybrid or local plus cloud
Sensitive or latency-sensitive systems stay in a local data centre or private cloud, while public cloud carries channels, analytics or seasonal capacity. This is common where data residency, existing hardware or connectivity make a full move unattractive. The hard part is the seam: identity, backup and monitoring must work across both environments.
At a glance, the trade-offs run in opposite directions:
- Single cloud: fastest to build and deepest managed features, but the highest concentration.
- Multi-cloud: reduces single-vendor failure, but adds the most operating complexity.
- Portable baseline: lowest exit cost, but requires engineering discipline.
- Hybrid: fits residency and existing hardware, but makes identity and backup harder.
The criteria that decide the risk
- Data export: can you obtain all records, relationships and history in a usable format, and how often?
- Egress and network: what does it cost to move data out, and does private networking tie you to one provider?
- Managed-service depth: how many critical workflows depend on a service with no realistic equivalent?
- Identity: can users and roles be recreated elsewhere without rebuilding access rules by hand?
- Automation: is infrastructure reproducible from code, or does it live in a console?
- Contract: renewal terms, minimum spend, notice period, exit assistance and deletion evidence.
- Skills: who can operate the alternative, and what would they need to learn?
When each model fits
Choose single cloud when the product is still finding fit and the team cannot carry platform overhead. Choose multi-cloud when you have genuinely independent workloads or a regulatory reason to avoid concentration, not because it sounds resilient. Choose a portable baseline when the system is core to the business and a future move is plausible. Choose hybrid when existing infrastructure, data residency or unreliable connectivity make the public cloud an incomplete answer.
The cost lines nobody quotes
Lock-in has a carrying cost and an exit cost. Carrying cost includes duplicate tooling, portability work that delivers no customer feature, and the engineering time spent keeping options open. Exit cost includes data migration, parallel running, retesting integrations, retraining staff and overlapping contracts. A credible business case compares both, not just the monthly bill.
Ask for unit costs rather than totals: per gigabyte exported, per engineering day, per integration retested and per week of parallel running. A vendor that cannot break the number down is selling a feeling, not a plan.
A 90-day lock-in review
- Days 1-15: inventory systems, data stores, identities and integrations, and name the owner of each.
- Days 16-30: score every layer for export effort, dependency depth and contract constraints.
- Days 31-50: test one real export into a second environment. A plan that has never been tried is a hypothesis.
- Days 51-75: fix the cheapest high-risk items: contract notice, data export schedule, infrastructure in code.
- Days 76-90: rehearse a partial move and report the residual risk to management.
Metrics
- The share of critical data with a tested export and restore procedure.
- Time to restore a core service in a second environment.
- The number of critical services with no documented alternative.
- Egress cost as a share of total cloud spend.
- The share of infrastructure defined in code rather than configured by hand.
- Contract notice periods compared with the realistic migration lead time.
Where this fits
If you need help assessing the estate or designing a more portable baseline, start with Managed Technology & Cybersecurity. The pricing page explains how we separate assessment, implementation and support. For the commercial pattern behind these decisions, read how cloud vendor lock-in changes year by year.
Sources
- Flexera, 2026 State of the Cloud Report, accessed 17 November 2026.
- NIST SP 800-145, The NIST Definition of Cloud Computing, accessed 17 November 2026.
- UK Competition and Markets Authority, Cloud Services Market Investigation, accessed 17 November 2026.
- FinOps Foundation, FinOps Framework, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.