The conversation around cloud hosting has shifted. For a decade, moving to the public cloud was treated as a one-way journey, the thing responsible businesses simply did. Today the traffic is running in both directions, and the numbers behind it are hard to ignore.
In 2026, moving workloads out of the public cloud is no longer a contrarian position. It is the mainstream plan.
The numbers behind the rebellion
Surveys of CIOs in 2026 report that 86% plan to move at least some workloads back from public cloud, the highest figure ever recorded. This is not a fringe of grumpy sysadmins. It is a substantial majority of the people who signed the cloud contracts in the first place.
The savings are not marginal either. The same 2026 research finds the average repatriated workload saves around 32% of its annual cloud cost, and that figure already accounts for hardware and operational overheads. Broader analyses, including Broadcom’s, put modern private infrastructure at 40-50% lower total cost of ownership for steady-state workloads.
Sit with that for a moment. For workloads that just run, day in, day out, the hyperscaler premium is not a rounding error. It can be close to half the bill.
The famous escapes
The headline cases are well known by now. 37signals, the company behind Basecamp, projects over $10 million saved across five years after leaving AWS. Dropbox saved $75 million over two years by building its own infrastructure. GEICO cut its compute costs by 50% per core.
The instinct for most small businesses is to file these stories under “not relevant to us”. These are companies with enormous bills, in-house platform expertise and negotiating leverage.
That instinct gets the scale right and the lesson wrong. The reason those companies saved money was not their size. It was the shape of their workloads. Steady, predictable demand is cheap to run on infrastructure you control and expensive to run on infrastructure priced for elasticity. That shape describes most SME systems far better than it describes Netflix.
What hyperscaler pricing actually buys
Hyperscaler cloud is a genuinely brilliant product for a specific problem: demand you cannot predict. It sells you the ability to scale to millions of users overnight, deploy in thirty regions, and spin up a thousand machines for an experiment on Tuesday and delete them on Wednesday.
That optionality is real, and you pay for it whether you use it or not.
Hyperscaler cloud is priced for elasticity that most small businesses never use.
Now look at what a typical SME actually runs. A website. A CMS. A database. Perhaps a booking system or a customer portal. Traffic that moves within a predictable band, with a busy season you can see coming a year out.
A predictable website, CMS and database workload is exactly the steady-state case where right-sized infrastructure wins on cost, simplicity and control. A dedicated server or a modest VPS, sized for your real peak with sensible headroom, plus an edge and CDN layer to absorb traffic spikes and serve static assets fast around the world, covers that shape of demand for a fraction of the price. The CDN handles the burst. The server handles the business.
There is a simplicity dividend too. One or two machines you understand fully are easier to reason about, secure and debug than a sprawl of managed services with their own billing meters, failure modes and permission systems.
The sovereignty question
There is a second force behind the repatriation wave that has nothing to do with cost. Surveys of IT leaders in 2026 find that 57% want their infrastructure located within a single country.
For UK businesses this is increasingly practical rather than ideological. Clients ask where their data lives. Contracts and insurers ask. Regulated customers ask twice. Those conversations are considerably shorter when the answer is a specific data centre in a specific country, rather than a region name and a paragraph about cross-border transfer mechanisms.
An honest caveat
None of this means the cloud is a mistake. It means the cloud is a tool, and tools have a shape.
If your demand is spiky, global or experiment-heavy, the hyperscalers remain the right answer. A startup that might have ten users or ten million next quarter should absolutely rent elasticity. So should anyone running short, intense bursts of compute, or serving latency-sensitive users on four continents.
And repatriation done badly is worse than an expensive bill. Moving off the cloud without monitoring, tested backups and someone accountable for the machines simply trades an invoice for an outage. The hyperscalers bundle a great deal of quiet operational competence into their prices: hardware replacement, patching, redundancy, someone awake at 3am. If you leave, that competence has to come from somewhere. It does not appear because you bought a server.
The honest framing is not cloud versus servers. It is rented elasticity versus owned predictability, with operations done properly in either case.
What to do about it
Pull the last twelve months of cloud invoices. Separate the spend into steady workloads that run constantly and elastic workloads that genuinely scale up and down. Most SMEs find the steady portion dominates.
Compare what you provision with what you use. Look at actual CPU, memory and bandwidth consumption against what you are paying for. Persistent low utilisation on a big bill is the signature of a workload paying for elasticity it never touches.
Price the alternative properly. Cost a right-sized dedicated server or VPS with an edge and CDN layer in front, and include the honest extras: monitoring, backups, patching, and the time of whoever is accountable. If the saving survives those overheads, and at 32% averages it usually does, it is real.
Insist on the boring parts. No migration should proceed without monitoring and alerting, automated and restore-tested backups, a patching routine and a named person who answers when something breaks. If nobody can name that person, you are not ready.
Move one workload first. Pick something meaningful but not existential, run it on the new footing for a quarter, measure cost and reliability, then decide about the rest. Repatriation is a series of small reversible decisions, not a leap.
The default is not neutral
Here is the uncomfortable truth about hyperscaler billing: it grows by default. Services accumulate, instances stay on, tiers creep upward, and shrinking the bill is nobody’s job. Doing nothing is itself a decision, and it is the most expensive one available.
This is the model we run for our own clients: managed, monitored, right-sized infrastructure, sized to the workload and watched by someone accountable, instead of an invoice that quietly compounds. The 86% are not rebelling against the cloud. They are rebelling against paying for a capability they never use. For most SMEs, that rebellion is overdue.
Flux Dynamics hosts and manages client systems on right-sized, monitored infrastructure rather than passing on a hyperscaler invoice that grows by default. Start a project and we will price what your workload actually needs.