# Multi-cloud vs AWS-first — decision matrix

Multi-cloud is an architecture tax you pay daily to insure against risks that
may never materialize. AWS-first-with-exit-ramps defers that tax until you
actually need to move. This matrix decides which trade is right for *your*
estate — not by ideology, but by counting hard requirements.

Score each row **0** (false), **1** (somewhat), **2** (hard requirement).
Sum the column. Read the result below.

| Driver | Pushes toward MULTI-CLOUD | Pushes toward AWS-FIRST |
|--------|---------------------------|-------------------------|
| Contractual: customer/regulator mandates >1 provider | 2 / 1 / 0 | — |
| Specific best-of-breed service only on another cloud (e.g. a unique ML/data product) | 2 / 1 / 0 | — |
| Acquired entity already runs production on another cloud | 2 / 1 / 0 | — |
| Sovereign/geographic coverage AWS cannot meet for a Region | 2 / 1 / 0 | — |
| Concentration-risk policy (board/financial-regulator, e.g. DORA) | 2 / 1 / 0 | — |
| Team headcount < 50 engineers | — | 2 / 1 / 0 |
| No dedicated platform team to run 2 control planes | — | 2 / 1 / 0 |
| Heavy use of managed/proprietary services for speed | — | 2 / 1 / 0 |
| Cost discipline matters more than provider diversity | — | 2 / 1 / 0 |
| You have never run a successful cross-cloud failover drill | — | 2 / 1 / 0 |

## Interpretation

- **Multi-cloud column ≥ 6 (with at least one hard "2"):** Deliberate
  multi-cloud is justified for the *specific* workloads driving the score.
  Keep the rest AWS-first. Multi-cloud is rarely an all-estate decision.
- **AWS-first column ≥ 6:** Go AWS-first. Buy insurance through
  **reversibility** (funded exit ramps per anchor workload), not through
  standing dual-cloud infrastructure.
- **Both columns 4-6:** Hybrid by workload. Name the 1-2 workloads that go
  multi-cloud; everything else stays single-cloud with exit ramps.

## The honest cost of "just in case" multi-cloud

| Cost | AWS-first + exit ramp | Standing active-active multi-cloud |
|------|------------------------|-------------------------------------|
| Daily ops surface | One control plane | Two control planes, two IAM models |
| Architecture | Best AWS-native services | Lowest common denominator |
| Egress | Within AWS / waived on exit | Continuous cross-cloud egress |
| Talent | One deep skill set | Two, or shallow in both |
| Exit cost | Deferred (re-platform project) | ~$0 (already everywhere) |

You are choosing **when** to pay: a one-time re-platform project later, or a
continuous tax now. For most mid-market estates, defer.

## Repatriation reality check (2026 published figures)

- Barclays CIO Survey (Q4 2024): ~83-86% plan to move *some* workloads back.
- IDC (Oct 2024): only ~8-9% plan *full* repatriation. The rest is selective.
- Gartner: ~90% of organizations expected on multi-cloud/hybrid by 2027.

Read these correctly: the trend is **workload-specific placement**, not a cloud
exodus and not blanket multi-cloud. Place each workload where its
cost-performance is strongest; keep an exit ramp on the anchors.