Cloud lock-in is usually discussed as a technology choice: use the managed service and accept the dependency, or build the portable version and accept the operational cost. For a business, it is a commercial choice. How much would it cost to leave, how long would it take and what would stop working during the move?
The answer is not "avoid lock-in at all costs". A managed database, queue or AI platform can be the right decision. The problem is unmeasured dependency: a contract with no exit clause, data that cannot be exported in a usable format and a team that has never tested a restore outside the provider.
Where things stand in 2026
Cloud estates are hybrid and multi-provider. Flexera's 2026 report puts hybrid adoption at 73 percent, with cost management still the top challenge and waste at 29 percent. AI workloads are adding a new layer of dependency: model APIs, vector stores, evaluation tools and GPU capacity. The lock-in question now includes the AI platform, not just the infrastructure.
For an SME, the practical concern is not moving an entire estate. It is the handful of workloads that would be expensive to move and the data that would be difficult to extract. Those are the ones to classify and plan for.
2027: classify workloads and write the exit clauses
2027 should start with a workload inventory. Classify each workload by portability: easy (containers, standard databases, object storage), moderate (managed services with a clear export path) and hard (proprietary AI platforms, deeply integrated serverless, data gravity). For the hard ones, document why the dependency is justified and what the exit would cost.
Contract terms matter. Ask for data export in a standard format, reasonable egress pricing, notice periods, deletion certification and support during migration. A contract that guarantees uptime but not exit is only half a contract.
- Data export: test it, do not just read the clause.
- Egress: model the cost of moving the data you actually have.
- Infrastructure as code: keep the deployment reproducible outside the provider where practical.
- Identity: avoid making the provider's identity system the only way to access critical data.
2028: AI platform lock-in and data gravity
By 2028, AI platforms will be a major source of lock-in. A vector database, a model API, a prompt management tool and an evaluation suite can become tightly coupled. The model may change, the price may change and the provider may deprecate a feature. The business needs a way to switch without rebuilding the workflow.
Keep the application layer model-agnostic where it matters: a common interface for prompts, retrieval and evaluation. Store the raw documents and embeddings in a format you can export. Keep the evaluation set separate from the vendor. If the vendor's dashboard is the only place where quality is measured, you cannot compare alternatives.
Data gravity is the quiet lock-in. The longer data lives in one platform, the more integrations, permissions and reports depend on it. Review the data architecture every year, not only at renewal.
2029: sovereignty, residency and negotiation
2029 is the year data residency and sovereignty become procurement requirements. Bangladesh's Personal Data Protection Act, 2026 and its residency provisions may affect where personal data can be stored or transferred. The cloud provider's region map is not enough; you need to know which services process data outside the region and which subprocessors are involved.
Negotiation leverage comes from evidence. If you can show that a workload is portable, the provider has a reason to offer better terms. If you cannot, the renewal conversation is one-sided. Run a tabletop exit for one workload and use the result in the negotiation.
Do not confuse multi-cloud with portability. Two providers can still be two silos with different identity, networking and data models. The question is whether the workload can move, not how many logos are on the architecture diagram.
2030: portability by design
By 2030, portability should be a design requirement for new workloads, not a rescue plan. Standard interfaces, open data formats, infrastructure as code and tested exports make the business a better customer and a better negotiator. The provider still matters; the dependency is measured.
The mature position is a portfolio: some workloads run on managed services because the value is clear, some are portable because the risk is high and some are hybrid because the data or latency requires it. The decision is documented, reviewed and priced.
What to do in the next 90 days
- List the top ten cloud workloads and classify them as easy, moderate or hard to move.
- For the top three hard workloads, write the exit cost and the business reason for the dependency.
- Test an export of one critical dataset and measure the time and cost.
- Review the contract for data export, egress, deletion and migration support.
- Keep the AI evaluation set and raw documents outside the vendor's walled garden.
How to measure progress
- Portable workload share: the percentage of critical workloads with a tested exit.
- Export time: how long it takes to extract a critical dataset.
- Egress cost: modelled and actual.
- Contract flexibility: notice, deletion and migration terms.
- Provider concentration: the share of critical workloads on one provider.
Lock-in is not a failure of judgment. Unmeasured lock-in is. Know what you are buying, what it would cost to leave and which dependencies are worth the trade.
Where this fits in our work
If you want help applying this to your business, start with Managed Technology & Cybersecurity. We publish scope and starting prices on the pricing page. For the next step in this series, read FinOps for small teams: 2027, 2028, 2029, 2030.
Sources
- Flexera, "2026 State of the Cloud Report", 18 March 2026.
- DataGuidance, "Bangladesh: Personal Data Protection Act 2026", accessed 16 November 2026.
- Gartner, "Worldwide IT Spending to Grow 15% in 2027", 21 October 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 *.