
VEXXHOST at OpenInfra Day North America 2026
VEXXHOST recaps OpenInfra Day North America 2026 in Chicago, highlighting its headline sponsorship, Mohammed Naser's talks, and the future of OpenStack operations.
Read field noteField notes / Latest
Engineering notes from operating open infrastructure: the failures, design decisions, and upstream work that make open infrastructure better.
Browse all field notes
VEXXHOST recaps OpenInfra Day North America 2026 in Chicago, highlighting its headline sponsorship, Mohammed Naser's talks, and the future of OpenStack operations.
Read field note
Ceph vs SAN: Compare performance, scalability, licensing, and operational costs to choose the right enterprise storage modernization strategy.
Read field note
Cloud exit strategy checklist: Review data ownership, exit costs, vendor lock-in, and migration support before signing a cloud contract.
Read field noteA lease end, power limit or end of hardware support forces a data centre exit. Here is how to sort workloads into colocation, hosted or public cloud.
A data centre exit usually begins with a date on a document. A lease comes up for renewal, the utility cannot bring more power into the building, the cooling plant reaches the end of its service life, or the vendor ends support for the servers and storage inside. Each of these triggers forces one decision: where every workload in the facility goes next, by a deadline the lease sets.
Each workload has three possible destinations:
Many data centres are old. In Uptime Institute's Global Data Center Survey 2025, 27.5% of respondents work in facilities 16 years old or older. Operators in the survey also named securing electrical power and timely access to hardware as growing barriers.
If a lease extension, a hardware order and a cloud contract are all on the table at once, you are choosing a home for each workload, an order for the waves and the size of anything left on site. Getting it wrong costs money in predictable places:
Each destination hands a different part of the stack to someone else.
You give up the building and keep everything inside it. The provider supplies power, cooling, physical security and cross-connects. Your team still owns hardware refresh, the hypervisor, storage, backups and every operating system. Physical work happens through the provider's remote hands service or your own staff on site.
Colocation suits hardware bought in the last few years with support left on it, specialized equipment such as hardware security modules, and systems that need low latency to each other.
The provider owns and houses the hardware, and you get a dedicated cloud with VMs, block storage, networks and an API. The hardware line moves from your balance sheet to a recurring service charge. Confirm the pricing unit and the hardware refresh terms in the proposal.
It suits steady VM estates whose hardware is close to end of support, and workloads that must stay in a known jurisdiction. Our hosted private clouds run on Atmosphere, our open source OpenStack platform, in our data centres. Your team can operate them with our 24/7 guidance, or we run them fully, including upgrades, monitoring and incident response.
Public cloud removes hardware ownership and bills for what you use. It suits demand that rises and falls, short-lived projects, and new applications built on the provider's managed databases and queues.
Steady, always-on workloads pay usage rates there for capacity that never scales down. Managed services can create provider dependencies, especially when applications use proprietary APIs.
Ask these of every workload group, in this order.
Remaining vendor support and book value decide whether hardware is worth moving at all. One hyperscaler has already shortened server lives. Its 2024 annual report cut the accounting useful life of a subset of its servers and networking equipment from six years to five, effective January 1, 2025. The report cited the faster pace of technology development, particularly in AI and machine learning.
Rule: if a server leaves vendor support before the first renewal date of its new home, retire it where it stands and move the workload to new capacity.
Applications follow their largest data sets. Splitting a database from the applications that query it across two sites adds latency to each transaction and can incur ongoing transfer charges. Price the actual traffic paths before deciding.
Exit-fee waivers apply to switching providers. Hyperscalers began waiving transfer fees for customers moving data out in early 2024, with prior approval and conditions on how the move is completed. Ongoing traffic from a split estate can incur charges, subject to allowances and qualifying multicloud programs.
Rule: move each large data set in the same wave as the applications that read it most, to the same destination.
Apply any residency rule to backups, replicas and logs as well as primary copies. Leaving your own building also adds a third party to the chain that handles that data. Accountability stays with you, and regulators expect a plan for leaving that provider too:
Our data centres are SOC 2 compliant.
Rule: for regulated workloads, the destination comes out of your risk assessment. It must let you name the jurisdiction of every copy and write the exit plan before you sign.
The exit changes jobs. Electrical, cooling and cabling roles largely disappear with a hosted private cloud or public cloud, while colocation still needs people who replace failed drives and patch hypervisors. Replacing those skills is hard: nearly two-thirds of the operators Uptime Institute surveyed in 2025 reported difficulty retaining staff, finding qualified candidates or both.
These people also hold the undocumented knowledge of which system depends on which.
Rule: colocation needs funded, accountable owners for hardware and platform operations after the exit, in-house or contracted. Without them, it is off the table for that workload. Keep the current team through the last wave either way.
Two dates matter: the renewal notice deadline, when you must tell the landlord whether you are staying, and the handback date, when the space must be empty. The notice deadline comes first. Large estates move in several waves, each with testing and a cutover window, so an exit can need more time than the lease leaves.
Rule: if the waves will not finish before the handback date, decide on a bridge before the notice deadline (see the staging plan below) and make the placement decisions afterwards.
Group first, then apply the constraints, then compare destinations.
A lease date is a contract term, so treat it as a cost you can negotiate. The architecture can then follow the placement list. Price each option against the same short list:
Then stage the move:
Leaving entirely is one option among several. Uptime's respondents put the share of their workloads hosted off premises at 55% in 2025, expecting 58% by 2027. Uptime's analysis adds that few large organizations are comfortable moving every workload out of their own data centres.
Keep a smaller footprint on site when:
A smaller site can run as a private cloud on Atmosphere, and we can run its OpenStack layer for you, from patching to on-call. The trade-offs between managed and self-managed on-premise cloud decide who is on call for it.
The strongest argument against this memo is simplicity. One migration to one provider ends hardware ownership, facilities contracts, facility capacity planning and hardware procurement on a single date. Workload sizing and spend decisions remain. A staged exit pays for two sites during the overlap, every bridge month adds rent, and each destination brings its own contract and monitoring. A team short on infrastructure staff may run one hyperscaler better than a mix.
That argument holds when:
It weakens for large, steady VM estates and large data sets. Those pay usage rates for constant load and can incur transfer charges wherever data and its readers sit apart. Some organizations later bring workloads back from public cloud, at the cost of a second migration.
Re-sort a group whenever a change in power prices, a new residency rule or a hardware refresh changes one of its five answers.
An exit sorted this way usually lands in more than one place. We operate data centres, and the same Atmosphere platform runs our public cloud, our hosted private clouds and clouds on customers' own hardware, on premises or in colocation. A deployment placed on our hardware or yours this year can move between hosted, on premises and colocation later without rebuilding. That matters in the second year of an exit, when a bridge ends or a colocated server reaches end of support.
Once your notice deadline is known, talk to our engineers about your lease dates and workload groups.
Choose from Atmosphere Cloud, Hosted, or On-Premise.
Simplify your cloud operations with our intuitive dashboard.
Run it yourself, tap our expert support, or opt for full remote operations.
Leverage Terraform, Ansible or APIs directly powered by OpenStack & Kubernetes