Most small companies looking at cloud migration in 2026 are not doing it because someone read a whitepaper about agility. They are doing it because the VMware renewal quote came in and it did not look like last year's.
Broadcom ended VMware perpetual licensing in December 2023 and moved everyone to per-core subscriptions with a 16-core minimum per CPU, whether or not the CPU has 16 cores. For three ESXi hosts with a pair of aging Xeons apiece, that arithmetic changes fast. So the server room gets re-examined for the first time in six years, which is healthy. Just do not let a renewal quote pick your architecture.
Next slide, the 6 Rs: rehost, replatform, repurchase, refactor, retire, retain. Useful words. Most decks stop there, as if sorting eighteen workloads into six buckets were a plan. The labels say nothing about sequence, cutover windows, or what breaks when the accounting server changes IP. Sequence and blast radius are the plan. The taxonomy is just how you talk about it in the meeting.
Rehost is the most maligned and most correct choice
"Lift and shift just moves your mess into someone else's data center" is the standard objection. It is true. It is also the right call for most SMBs. Refactoring on the way to the cloud means debugging two problems at once. When something misbehaves during cutover weekend, you want exactly one variable in play: did the platform change, or did the code change? Not both.
Move the VM. Confirm it behaves for a month. Then, with a working rollback and no deadline attached, decide whether that file server should have been SharePoint all along. Migration and modernization are two projects. Run them as one and you finish neither on time.
What actually kills these projects
Never the compute. Compute is a form you fill out. It is these two:
- The file server. Four terabytes of shared drives, fifteen years of NTFS permissions inherited from an admin who left in 2018, and a
P:\path hardcoded into Excel workbooks whose author has retired. Copying the bytes is trivial. Reproducing the permission model, and the drive letter, eats the weekend. - The line-of-business app the vendor stopped supporting in 2019. It wants Server 2012 R2, an old SQL compatibility level, and a license key tied to hardware. The vendor will tell you to buy their new cloud product. Sometimes that is the honest answer. Either way, price it before you commit to a date, because that one app decides whether this is a six-week project or a nine-month one.
Cost efficiency is not a property of the cloud
On-prem, you buy for peak once. In the cloud, idle capacity bills by the hour. A lift-and-shift of an over-provisioned estate, where every VM got 8 vCPU and 32 GB because that was the template, routinely costs more per month than the amortized hardware it replaced. Right-size before you move, not after the first invoice.
Then there is egress. AWS, Azure, and Google Cloud all charge per GB of data leaving. Cloudflare R2 charges nothing for it, and Backblaze B2 publishes free egress up to three times your average monthly stored data, which is why both show up in backup designs. Archive tiers are worse than they look: cheap to store, expensive to retrieve, and subject to minimum durations, so deleting a Glacier Deep Archive object after a week still bills you for the full 180 days. All three now waive transfer-out fees if you leave outright, but that is a one-time claim with a deadline and does nothing for a routine restore.
Start with email and files, then stop for a year
For most SMBs: move mail and files to Microsoft 365 Business Premium and leave everything else alone. That retires the two workloads with the ugliest on-prem failure modes, the migration paths are well worn, and it buys a year of usage data before anyone touches the ERP.
Sometimes the answer is no migration: one refreshed host running Hyper-V or Proxmox instead of a VMware subscription, plus immutable offsite backup with Object Lock. Cheaper, faster, and boring. We say so when it is true. The alternative is billing a client for a year of work they did not need. Our core IT services page covers what an assessment involves. But the first useful step costs nothing: find out what the file server and that legacy app will really take. Price those two and the schedule writes itself.
