Definition
Multi-Cloud Portability Erosion Index (MCPEI) measures the rate of increase in provider-specific service dependencies as a function of AI placement decisions.
The Measurement Gap
A global enterprise adopted multi-cloud architecture three years ago with an explicit "avoid lock-in" strategy. The architecture standards mandated abstraction layers over provider-specific services, with the intent that workloads could move between providers without major rework. The cloud architecture team validated each new service against the portability standards. The reviews ran for two years.
Then the workload optimization initiative launched. An AI placement engine analyzed workloads and recommended consolidation against provider-specific services where the recommendations would reduce cost or improve performance. The architecture review function shifted from gatekeeping to "best practices guidance" because the optimizer's recommendations were data-driven. Twelve months in, the cost savings are real ($4.7M annualized) and the optimization platform is highlighted as a success story.
Then a procurement renegotiation surfaces a question. The team needs to estimate the cost of moving a particular set of workloads off the primary provider if the negotiation goes badly. The cloud architecture team starts the assessment. The workloads they expected to be portable have accumulated dependencies on provider-specific services: managed databases with provider-specific APIs, serverless functions written against provider event formats, machine learning services with proprietary model formats, identity services tied into provider IAM. Each individual binding was justified by the optimizer. The cumulative effect is that the workloads now have hundreds of provider-specific bindings where they had a few dozen before.
The estimate comes back: moving the workloads would take 18 months and cost $12M in engineering work. The negotiating leverage the multi-cloud strategy was supposed to provide has been silently spent on cost savings of $4.7M annualized. The procurement team has substantially less leverage than the strategy assumed.
The legacy multi-cloud metrics report a healthy deployment.
Cost per cloud: optimized, declining quarterly.
Workload distribution by provider: tracked, slightly more concentrated on the primary provider.
Cross-cloud bandwidth: stable.
Cloud spend efficiency: improved.
Optimization recommendations adopted: high.
Multi-cloud strategic compliance: tracked qualitatively, considered healthy.
None of these report on the cumulative lock-in being created by the optimization recommendations. Each recommendation was individually small. The cumulative effect was significant. The portability erosion is happening one workload binding at a time, invisible at the workload level, only visible at the portfolio level.
These metrics were designed when workload placement was made by architects who weighed cost against portability explicitly. The architect rejected provider-specific bindings that compromised portability even when they offered cost savings. With AI-driven optimization, the placement decisions are made by an engine optimizing for cost and performance, with portability as a constraint that is either binary (allowed / not allowed) or invisible (not in the objective function).
There is no metric that tells you whether AI workload placement decisions are incrementally increasing provider lock-in, which means multi-cloud strategy erodes silently.
Formula
Multi-Cloud Portability Erosion Index (MCPEI) measures the rate of increase in provider-specific service dependencies as a function of AI placement decisions. In plain language: count the workload bindings to provider-specific services at baseline (when AI placement was enabled). Count them now. The percentage increase, normalized by total bindings, is MCPEI. A positive value means lock-in is increasing. A high positive value means lock-in is increasing rapidly.
The formula:
MCPEI = (Proprietary_bindings_current − Proprietary_bindings_baseline) / Total_service_bindings_current
Where:
- Proprietary_bindings_current = the count of bindings to provider-specific services at the current measurement
- Proprietary_bindings_baseline = the count of bindings to provider-specific services at the baseline (typically when AI placement was enabled or when the multi-cloud strategy was last reviewed)
-
Total_service_bindings_current = the count of all service bindings at the current measurementUnits are a percentage change. Positive values mean increasing lock-in. Negative values mean decreasing. A worked example. The global enterprise computes MCPEI after the procurement-driven review:
-
Baseline (24 months ago): Total bindings = 412, Proprietary bindings = 78
- Current: Total bindings = 487, Proprietary bindings = 165
- Bindings added since baseline: 75 total, of which the count of new proprietary bindings = 87 (some old generic bindings were replaced with proprietary)
MCPEI = (165 − 78) / 487 = 87 / 487 = 17.9%
Per the catalog guidance (below 5% annual increase stable, above 10% systematic deepening) this MCPEI of 17.9% across 24 months represents roughly 9% per year of lock-in deepening, but the trend appears to be accelerating in the most recent year. The action trigger at 8% has been crossed. The team needs to add a portability constraint to the optimizer that penalizes new proprietary bindings unless they pass an architecture review. Three edge cases need explicit treatment. First, define "provider-specific" per service type before measurement: what counts as proprietary varies (managed Kubernetes is portable; provider-specific managed databases are usually not; serverless functions vary by how they are written). Document the categorization rules and apply consistently. Second, exclude pre-existing single-provider workloads: workloads that were never multi-cloud should not count toward MCPEI because they were never portable to begin with. Third, the minimum is 50+ workloads under AI placement for the percentage to be meaningful. Cross-references. Downstream of MCPEI is
Replica Consistency Divergence Score (RCDS, GMI-IT-055): portability erosion often involves replica-placement decisions that have data consistency implications. The misinterpretation to watch for: a team reports MCPEI = 3% (low) and concludes their portability is stable. They miss that the 3% is computed only over workloads explicitly tagged as multi-cloud workloads, while the AI optimizer has been recommending new workloads be deployed as single-cloud from the start, never entering the multi-cloud-tagged set. Track MCPEI alongside the rate of new workloads being deployed as single-cloud; both numbers describe portability decisions.
Proposed formula, pending validation pilots.