Cloud migration services — strategy for private and hybrid infrastructure

Open-source Cozystack (a CNCF project we create and maintain) Ænix Platform, the supported commercial distribution Aenix builds, operates and migrates it.

The right question in a 2026 cloud migration is which workloads run best where — and increasingly the answer is a mix of public cloud, private cloud, and repatriated on-premises capacity, not a mandate to move everything to public cloud. Aenix runs structured cloud migrations across three common patterns: public-cloud repatriation driven by cost or sovereignty, VMware exit under Broadcom subscription pressure, and greenfield private-cloud builds. Each starts with a Platform Readiness Assessment — workload classification, honest TCO modelling, and a target architecture — before a single workload moves. The team that ships your migration is the same team behind Cozystack, the open-source platform most migrations land on.

Quick facts

  • What it is A structured cloud migration engagement — repatriation, VMware exit, or greenfield private cloud — led by workload placement, not a fixed destination
  • Primary patterns Public-cloud repatriation, VMware/VCF exit, sovereignty-driven relocation, AI/GPU economics, greenfield build
  • Destination platform Cozystack (Apache 2.0, no per-CPU licensing) on customer-controlled hardware — other targets where a better technical fit
  • Assessment timeline 14-28 days for Platform Readiness Assessment and written target architecture
  • Implementation timeline 3-18 months depending on estate size, integrated with your team
  • Areas served EU, DACH, Central Asia — time-zone-aligned engineering
  • Production reference A European academic-computing SaaS moved off a hyperscaler onto owned bare metal with no user downtime and cut GPU cost about fivefold (see the multi-cloud academic GPU case study); tier-1 European bank engagements remain NDA-protected until mid-2027

Source: Cozystack docs, CNCF Landscape

Cloud migration in 2026 is a workload-placement decision, not a race to public cloud. Ænix runs structured cloud migrations — public-cloud repatriation, VMware exit, and greenfield private-cloud builds — where the destination is chosen from the workload, not assumed up front.

The team that ships your migration is the same team behind Cozystack — the open-source platform most private-cloud migrations land on. We work alongside your engineers for assessment, sequencing, and implementation.

Pairs with: one of the Ænix platforms — the destination follows the buyer profile. Organizations that sell cloud to external customers (hosters, MSPs, telcos, national operators) land on the Public Cloud Platform; regulated organizations running cloud for their own developers land on the Private Cloud Platform, whose developer self-service layer replaces the internal PaaS; GPU and inference estates land on the AI Platform.


When a cloud migration makes sense

Migration is worth the disruption when a concrete trigger is driving it. The common ones in 2026:

  • VMware exit under Broadcom subscription pressure — 2-5x renewal increases, ELA breakage, and mandatory VCF bundling push infrastructure teams to a platform they control. See VMware alternative for the destination, and the dedicated VMware migration hub for cohort sequencing tied to subscription expirations.
  • Public-cloud repatriation driven by cost or sovereignty — steady-state workloads that were cheap to start in a hyperscaler become expensive at scale, and data-residency rules increasingly require customer-controlled infrastructure. See cloud repatriation.
  • Sovereignty requirements — DORA, NIS2, and sectoral rules force critical workloads onto infrastructure with a clear jurisdiction and audit trail. See data sovereignty.
  • AI and GPU economics — sustained inference and training workloads are far cheaper on owned GPUs at reasonable utilisation than on rented hyperscaler capacity. See Sovereign AI.
  • Greenfield projects — a new platform with no legacy estate to unwind, where modern architecture can be adopted from day one on a private cloud platform.

If two or more of these apply, a structured migration compounds the benefit. If none applies and your current setup is comfortable, “stay and tune” is the honest recommendation — and one we make regularly.


How Ænix engages on a cloud migration

The engagement is deliberately staged so you commit incrementally, with a decision gate before the expensive phase.

Workload-placement framework

The assessment sorts every workload against two axes: how well it fits a private platform technically, and what it costs where it runs today. Four outcomes fall out of that grid — migrate now (clear technical and economic win), migrate later (fit is good but sequencing or contracts dictate timing), re-platform (needs redesign before it moves), and stay (already in the right place). The strategy is the sum of these per-workload decisions, not a top-down target percentage.

What to keep where it is

An honest migration plan leaves workloads alone when moving them adds risk without return. Bursty, unpredictable workloads often belong in public cloud where elasticity is cheap. Managed services with no on-premises equivalent may not be worth rebuilding. Applications mid-rewrite should wait for the new architecture rather than migrate twice. Ænix has no hyperscaler partner economics and no incentive to over-migrate, so “keep it where it is” is a recommendation we make without hesitation when the numbers support it.


How the migration itself runs

Execution follows a cohort-based pattern rather than a single cutover. Workloads are grouped into cohorts by dependency and risk, and each cohort moves to the target platform while the source keeps running. Source and destination run in parallel until the cohort is validated — functionally, on performance, and on data integrity — and only then is the source decommissioned. Nothing is switched off on the promise that the new environment will work.

For estates leaving a legacy virtualisation stack, image conversion is automated: KubeVirt’s containerized data importer ingests virtual-machine images into the destination platform, and Windows guests get their in-guest tooling cleaned up before first boot on the new hypervisor. Networking and storage are redesigned rather than copied — a private platform built on Cilium and LINSTOR behaves differently from NSX and vSAN, and skipping that redesign is one of the most common causes of post-migration fragility.

Sequencing respects what you have already paid for. When contracts or subscriptions still have time on them, the affected cohorts move last, so the plan never forces a write-off of committed spend. A 100-workload estate typically completes in months, not years; larger estates run in cohorts across a longer window while the source environment winds down cohort by cohort.

Source estate
Public cloudLegacy virtualisation stack
moves through
Cohort-based migration
Parallel-run validationAutomated image conversion
lands on
Cozystack
KubeVirtCiliumLINSTOR
completes with
Source decommissioned
Cohort by cohortNo write-off of committed spend

Where cloud migrations commonly stall

Most failed migrations share a small set of causes, and the assessment phase exists to catch them early:

No honest TCO before moving. Hardware refresh, platform-team capacity, and the operational learning curve get left out of the model, and the project stalls when the economics turn out different from the pitch.

Big-bang cutover. A single-weekend “move it all” rarely survives contact with an enterprise estate. Cohort-based migration with validated parallel-run is the pattern that works.

An under-engineered destination. Workloads land on a private platform that was never built for production; operational debt accrues and the team blames the migration when the real issue is destination maturity.

Skipped network and storage redesign. Treating the target’s networking and storage as a copy of the source guarantees fragility. They are engineered fresh for the destination.


Model the numbers before you commit

Migration economics look attractive in the abstract and turn on details in practice — hardware refresh, platform-team capacity, and the operational learning curve all move the result. Before committing, model the delta with the ROI & TCO calculators: VMware-exit savings, DIY-versus-Ænix platform TCO, hosting unit economics, and GPU/AI-inference ROI, each with editable inputs and live results.

For a worked example of a mixed-placement outcome, see the multi-cloud academic GPU case study — where the right answer was a blend of owned GPU capacity and retained cloud, not a wholesale move in either direction.


Ænix is the team behind Cozystack (CNCF Project), and we offer Ænix Platform — our commercial productized offering based on Cozystack.

Frequently asked questions

What are cloud migration services?

Cloud migration services cover the assessment, architecture, and execution work of moving workloads between environments — public cloud, private cloud, or on-premises. Aenix delivers this as a structured engagement: workload classification, TCO modelling, target architecture, then cohort-based execution with parallel-run validation.

Does cloud migration mean moving everything to public cloud?

No. In 2026 the mature approach is workload placement — deciding case by case whether each workload belongs in public cloud, on a private platform, or repatriated on-premises. Many organisations are moving in the opposite direction, repatriating from hyperscalers for cost and sovereignty reasons.

How do you build a cloud migration strategy?

Start with a Platform Readiness Assessment: inventory every workload, classify it (migrate now / later / stay / re-platform), model TCO honestly, and design the target architecture. The strategy falls out of that classification rather than a top-down mandate to move a set percentage.

What is private cloud migration?

Private cloud migration relocates workloads onto infrastructure you control — your own hardware or a dedicated environment — instead of a shared hyperscaler. Aenix typically lands these on Cozystack (KubeVirt, Cilium, LINSTOR), an Apache 2.0 platform with no per-CPU licensing.

How is this different from a VMware migration?

VMware migration is one specific pattern — exiting VCF under Broadcom pricing. This page covers the broader strategy across all cloud migration patterns. If your trigger is specifically VMware, see the dedicated VMware migration hub for cohort sequencing tied to subscription expirations.

Can we keep some workloads where they are?

Yes, and you often should. An honest assessment recommends leaving workloads in place — including in public cloud — when that is the right technical and economic fit. Aenix has no hyperscaler bias and no incentive to over-migrate.

Planning your migration?

Start with a fixed-price Platform Readiness Assessment — architecture review, gap analysis, and a roadmap you can take to the board.