Forrester Says You Cannot Fully Leave US Cloud in 2026. Here Is What You Can Do Instead.
Forrester Says You Cannot Fully Leave US Cloud in 2026. Here Is What You Can Do Instead.
Two findings that sit awkwardly together.
Gartner surveyed CIOs and IT leaders in Western Europe and found 61% expect geopolitics to increase their reliance on local or regional cloud providers. 1 And according to Forrester, no European company will be able to completely move away from US hyperscalers by 2026, because Europe's push for sovereignty won't dislodge US tech dominance this year. 2
Both are true, and the conclusion most people draw from them is wrong. The usual reading is that sovereignty is aspirational, so park it. The better reading is that full migration was always the wrong unit of work.
All-or-nothing keeps failing
A complete migration off a hyperscaler is a multi-year programme touching managed databases, identity, queues, serverless glue, observability and a decade of accumulated IAM policy. It's expensive, it's risky, and it competes with every revenue-generating project in the roadmap. So it gets proposed, costed, and quietly deferred. Every year.
Meanwhile the real exposures, the ones that show up in an audit or a regulator's letter, are usually concentrated in a small number of places rather than spread across the whole estate. A handful of datasets, a handful of endpoints.
The unbundled version
Treat it as a set of independent, individually cheap moves.
Move the data that carries the obligation, and whatever processing the obligation follows with it. Personal data, customer records, and anything with a residency requirement attached, remembering that some regimes bind where the processing happens and not only where the bytes rest. That's often a small fraction of total volume and the overwhelming majority of the legal risk.
Get one credible exit off zero. The hardest part of leaving isn't the second workload, it's the first. One production service running somewhere else, anything real with monitoring and a runbook, converts "we couldn't leave" into "we know how long it takes".
Put backups outside the primary provider. This is one of the cheapest sovereignty and resilience moves available, and it doubles as the ransomware control we wrote about in May. If your only copy lives inside one provider's account, sovereignty isn't your most urgent problem.
Own your addressing and routing. If you hold your own IP space and announce it yourself, moving between providers stops being a renumbering project. This is the single most underrated exit-cost reduction in infrastructure.
Split the control plane from the data plane. Running DNS, or authentication, or your status page somewhere other than your main provider means a provider-level failure doesn't also take away your ability to respond to it.
What this costs with us
None of the above requires a migration programme, and most of it is priced like a rounding error.
AVS ISP runs Tirana, Skopje, Amsterdam and Dublin, which is a genuine choice of jurisdiction per machine, including two locations outside the EU. A VPS starts at €5/mo, a VDS runs from €40/mo for 4 vCPU to €160/mo for 16, and bare metal starts at €150/mo. Crypto accepted, no KYC, deployment in minutes.
On the addressing point, we operate AS210464, run RPKI validation, publish a full BGP community set for controlling exactly where your prefixes are announced from, and sell BGP sessions from €5/mo. Every VM comes with a routed /48 of IPv6, and we lease space to customers who need something of their own to announce. If you bring your own /24, you can move it between our sites without touching a single application config, and take it with you if you leave us, which is rather the point.
You can measure our network before committing anything through the looking glass, which runs live ping, traceroute and BGP lookups from all four sites with no account.
The test worth running
Forget the migration business case for a moment and answer a smaller question. How long would it take you to stand up your most important service somewhere else, if you had to?
If you don't know, that's the finding. And the way to find out isn't a strategy document. It's one small server at another provider, in another country, with one real thing running on it.
That experiment costs €5 a month. The strategy document costs a lot more and tells you less.