Cloud Migration Cost: What It Really Runs in 2026

What a cloud migration actually costs

Across 2,000+ projects at Digital Heroes, cloud migration quotes cluster into three honest bands. The number moves on workload count and database complexity, not on how impressive the deck looks.

Tier 1: Lift and shift, $20,000 to $50,000, 5 to 9 weeks

Ten to 25 virtual machines moved as-is from a data centre or VMware cluster into AWS, Azure or Google Cloud. You get a light discovery pass, a single-account landing zone, networking and identity wired up, replication via the cloud vendor's own tooling, one cutover rehearsal, and a weekend cutover with a rollback plan. One or two engineers plus a part-time architect.

What falls out of scope at this tier, explicitly: no application refactoring, no database engine change, no zero-downtime cutover (plan for 4 to 12 hours of outage), no multi-account guardrails, no compliance evidence pack, no cost optimization pass, no disaster recovery build, no observability rebuild beyond default cloud metrics. If a vendor quotes $35,000 and any of those words appear in the proposal, the proposal is wrong.

Tier 2: Replatform, $65,000 to $190,000, 3 to 6 months

Thirty to 90 workloads, two to five databases, a handful of applications moved onto managed services instead of VMs. Multi-account landing zone written in Terraform, secrets management, a real network design with hybrid connectivity, waves of migration with rehearsals per wave, monitoring rebuilt, runbooks written, two weeks of hypercare after cutover. Team is typically an architect, two to three cloud engineers, a DBA for part of it, a QA engineer, and a project manager. This is where most companies actually land.

Tier 3: Refactor or regulated, $200,000 to $650,000, 7 to 14 months

Oracle or mainframe-adjacent systems, PCI or HIPAA scope, 100+ workloads, applications rearchitected into containers or serverless, near-zero downtime cutover, full DR with tested failover, and an auditor who will ask for evidence. Add a security engineer, a data engineer, and a compliance lead to the team. Anything past $650,000 is a portfolio programme, not a project, and should be broken into funded waves.

What actually drives the number

1. Workload count and shape ($900 to $1,600 per straightforward VM). At volume, a clean Linux or Windows VM with no exotic dependencies costs $900 to $1,600 to discover, replicate, test and cut over. The first ten cost roughly 3x that because you are building the pipeline, not moving servers. The killers are the outliers: an unsupported OS, hardcoded IP addresses, a licence-locked appliance, a service account nobody can find the password for. Budget $4,000 to $12,000 per offender, and assume 10 to 15 percent of your estate qualifies.

2. Database migration and engine change ($3,000 to $70,000 per database). Same engine to a managed equivalent, SQL Server to RDS SQL Server or Postgres to Cloud SQL, is $3,000 to $8,000 per instance. Changing engines is a rewrite, not a move. Oracle to Aurora PostgreSQL or SQL Server to Aurora runs $25,000 to $70,000 per database because stored procedures, triggers and application SQL all get rewritten and then regression tested. One Oracle instance can cost more than 40 clean VMs.

3. Data volume and the cutover window (adds $15,000 to $60,000). Under 5 TB with an 8 hour window is a non-event. Over 20 TB, or a window under 4 hours, forces continuous replication, delta syncs, physical transfer appliances and two or three full rehearsals. AWS publishes DataSync at $0.0125 per GB transferred, which is small money, but the engineering time to prove a 4 hour cutover is not: expect $15,000 to $60,000 on top depending on how many systems must land together.

4. Compliance and regulated data (adds 18 to 30 percent). HIPAA, SOC 2, PCI DSS or GDPR data residency add encryption design, key management, logging retention, access reviews, network segmentation and an evidence pack. On a $150,000 build that is $27,000 to $45,000 of real work. It cannot be bolted on at the end without redoing the landing zone.

5. Landing zone depth ($12,000 to $45,000). A single account with a default VPC costs $12,000 to stand up. Multi-account with organizational guardrails, centralized logging, transit gateway, hybrid connectivity back to your office or remaining data centre, and the whole thing in Terraform costs $35,000 to $45,000. This line item is the one cheap quotes delete, and it is the one you cannot retrofit cheaply.

6. Rehost versus replatform versus refactor ($1,200 to $60,000 per application). Rehosting an app is roughly the VM price. Replatforming it onto a managed container service is $3,500 to $9,000. Refactoring it into microservices or serverless is $18,000 to $60,000 per application. Refactoring during a migration is the single most common way a $120,000 project becomes a $400,000 one. Move first, refactor after, unless the app physically cannot run on the target.

Worked example: 40 VMs and two databases, VMware to AWS

A mid-market logistics company: 40 VMs on supported operating systems with no licence-locked appliances, six internal web applications, two SQL Server 2016 databases, 9 TB of data, SOC 2 required, 8 hour cutover window acceptable. Five months.

• Discovery, dependency mapping, TCO model: $11,000
• Landing zone: three accounts, VPC, transit gateway, IAM, guardrails, all in Terraform: $24,000
• Rehost 40 clean VMs via AWS Application Migration Service, four waves, $1,150 each: $46,000
• Replatform six web apps onto ECS Fargate, $3,500 each: $21,000
• Two SQL Server databases to RDS, same engine, $6,000 each: $12,000
• Data migration engineering, 9 TB via DataSync plus one Snowball: $7,500
• AWS transfer and appliance charges at list: $2,400
• Two cutover rehearsals plus the weekend cutover: $9,000
• Observability rebuild, CloudWatch plus agents on 40 hosts: $8,000
• SOC 2 evidence pack and security hardening: $14,000
• Runbooks, handover, two weeks hypercare: $9,000
• Project management and QA across five months: $18,000

Total: $181,900. That sits near the top of Tier 2, and the reason is one line: SOC 2. Strip compliance ($14,000) and the observability rebuild ($8,000) and the same estate lands at $159,900. Add one Oracle database coming off the floor at $25,000 to $70,000 and it lands between $206,900 and $251,900.

The ongoing costs nobody quotes

The cloud bill itself. The migration quote is not the cost of running in the cloud. For that estate, 40 compute instances, two RDS instances, storage and load balancers, expect $6,800 to $11,000 per month on-demand. AWS publishes compute Savings Plans discounts of up to 72 percent against on-demand for a three year all-upfront commitment, with one year and no-upfront terms discounted progressively less. If nobody buys commitments by month three, you are paying on-demand for a workload that is not going anywhere.

Egress. AWS publishes data transfer out to the internet at $0.09 per GB for the first 10 TB per month. A reporting system that ships 4 TB a month out of the cloud is $360 a month that did not exist in your data centre. Model it before cutover, not after the first invoice.

Third party services. Datadog publishes Pro host monitoring at $15 per host per month, so 40 hosts is $600 a month. Add log management, a secrets or state backend, and an on-call tool, and $900 to $1,800 a month is normal.

Maintenance at 15 to 20 percent of build per year. On a $181,900 build that is $27,300 to $36,400 annually for patching, Terraform upkeep, alert tuning, DR tests and the small fixes that follow anything real. Vendors who skip this line are hoping you will pay it as ad hoc tickets at a worse rate.

The first year of changes. Every migration surfaces work: an app that needs autoscaling, a report that got slow, an integration that broke. Budget 25 to 40 percent of the build cost for year one changes. On this project, $45,500 to $72,800. It will get spent. Deciding that now is cheaper than discovering it in month four with no budget line.

How to not get burned on price

The cheapest quote is usually cheap because it deleted discovery. Without dependency mapping, the vendor is assuming every VM is a clean VM, and 10 to 15 percent are not. In our experience that assumption produces 20 to 35 percent cost overrun plus at least one failed cutover, and a failed cutover costs more than the discovery would have. Ask any low quote one question: which specific servers did you price, by name? If they cannot produce the list, they priced a guess.

A change request should be priced in engineering days, not adjectives. $1,400 to $2,200 per day is a fair rate for senior cloud engineering, and a change request should say "3.5 days, $6,300, here is what it touches." Anything under half a day should be absorbed without a ticket. If every question generates a change request, the original scope was written to be incomplete.

Four contract terms protect the number. First, fixed scope with a named inventory: an appendix listing every VM, database and application by hostname, with anything not listed priced separately. Second, IP assigns to you on payment, not on project completion, so a dispute does not hold your Terraform hostage. Third, source code and Terraform state live in your repository and your cloud accounts from day one, never the vendor's. Fourth, a rollback plan per wave, written and tested, with acceptance defined as "systems running and verified in the target for 72 hours," not "files copied."

How to brief a vendor so the quotes are comparable

Most quotes are not comparable because every vendor priced a different project. Send all of them the same six things and the spread collapses to something you can actually judge.

1. A workload inventory export. RVTools for VMware, or the assessment export from Azure Migrate or AWS Application Discovery Service. Hostnames, CPU, RAM, disk, OS version.
2. Data volume per system, and where it grows fastest.
3. Maximum acceptable downtime per system, in hours, per system, not one number for everything. This one line moves the price more than any other.
4. Compliance regime and audit date, if any.
5. Target cloud, and whether it is decided. If it is not, say so and ask them to price the decision as a separate assessment.
6. Which applications you would refactor and which you would not touch. Be blunt. Refactor appetite is the difference between $120,000 and $400,000.

Then require every vendor to price the same four phases separately: discovery, landing zone, migration waves, and cutover plus hypercare. Once the phases line up, the comparison takes ten minutes, and the vendor who is $60,000 cheaper will be visibly cheaper in exactly one phase. That is where the risk is hiding.