The headline AWS migration cost figure most teams arrive at is wrong by a factor of two. Not because the AWS pricing calculator is inaccurate, but because it does not model what cloud migration actually consists of. Compute and storage are the visible cost. Egress, parallel running, retraining, partner consulting, support plans, right-sizing premium, and security rework are the invisible cost. They add up to between 40 and 60 percent of a typical migration budget. This page is a reference for the full number, not just the AWS bill at steady state.
The single biggest determinant of AWS migration cost is which of the AWS 7Rs strategies a workload follows. Lift-and-shift (rehost) is cheapest upfront and most expensive to run. Refactor is the inverse. Most enterprise programmes end up with a 60-30-10 mix across rehost, replatform, and refactor by workload count, which produces a blended per-server cost of roughly $14,000 before egress and dual-running.
Per-workload AWS migration cost by strategy, list-price labour
| Strategy | Per workload (low) | Per workload (typical) | Per workload (high) | Timeline per workload |
|---|---|---|---|---|
| Rehost (lift-and-shift) | $3,000 | $5,500 | $8,000 | 2 to 4 weeks |
| Replatform | $8,000 | $16,000 | $25,000 | 4 to 12 weeks |
| Refactor / re-architect | $25,000 | $52,000 | $80,000+ | 12 to 52 weeks |
| Repurchase (move to SaaS) | $10,000 | $30,000 | $120,000+ | 8 to 26 weeks |
| Relocate (VMware Cloud on AWS) | $4,000 | $7,500 | $12,000 | 2 to 6 weeks |
Workload type changes the per-workload bands meaningfully. A web application is cheaper to rehost than a SQL Server database, because the database brings BYOL licensing decisions, AWS Database Migration Service replication setup, and engine compatibility validation. A monolithic Java application that refactors to microservices on Amazon ECS or AWS Lambda often lands at the top of the refactor band because of the dependency mapping work alone.
You do not pay AWS to bring data into AWS. You pay your existing provider, or the network path, to push it out. AWS only charges egress on the way back out of AWS once data has arrived. For most on-premise to AWS migrations this means the egress cost is paid as either (a) the public internet egress cost of the source environment, (b) AWS Direct Connect port and data transfer fees if a dedicated link is provisioned, or (c) Snowball device rental plus a small per-GB ingest fee.
The arithmetic for a 20 TB migration sets the cost bound. Across a 1 Gbps public internet link, 20 TB at sustained 800 Mbps line rate takes roughly 56 hours of pure transfer, which assumes nothing else uses the link. Most production networks cannot sustain that, so a more realistic plan is to use AWS DataSync for incremental copy plus an initial Snowball for the bulk. DataSync charges $0.0125 per GB transferred. AWS publishes the rate on the DataSync pricing page (as of May 2026). Snowball Edge Storage Optimized devices charge a service fee per device plus a daily extension fee after the included 10-day period. The current device fee is on the AWS Snowball pricing page.
The Snowball break-even
Workloads running Windows Server move two ways into AWS. License Included (LI) instances come bundled with Microsoft licensing, billed per second on EC2 alongside the compute. Bring Your Own License (BYOL) on EC2 Dedicated Hosts lets a customer use existing on-premise Windows Server or SQL Server licences with Software Assurance, but only on dedicated tenancy hardware. The tenancy requirement is the trap. Dedicated Hosts cost significantly more per instance than shared tenancy and limit scaling flexibility.
For SQL Server specifically, AWS RDS for SQL Server (License Included) is often cheaper than BYOL on EC2 Dedicated Hosts once you account for storage, backups, and the dedicated host minimum commitment. The exception is when an on-premise SQL Server estate already has Software Assurance through to 2028 and the cost is fully sunk; in that case Azure Hybrid Benefit on Azure SQL Managed Instance becomes the cheapest destination, not AWS. The licence economics push some Microsoft-heavy estates to Azure for that reason alone, not because Azure compute is cheaper.
AWS MAP is a structured funding programme designed to offset migration cost in exchange for a committed AWS spend ramp. The programme has three phases: Assess (free workshops and tools like Migration Evaluator), Mobilize (small credit to fund landing zone and partner discovery), and Migrate (the large credit, typically 25 to 40 percent of migration cost, tied to wave milestones). MAP is described in the public AWS Migration Acceleration Program page.
Two non-obvious MAP qualification rules drive cost outcomes. First, MAP funding is harder to access once a customer has signed an Enterprise Agreement or has already migrated a material portion of the estate; AWS reserves MAP for new committed spend. Second, MAP credits flow through the partner of record. Self-led migrations are not eligible for the migrate-phase credit, although they can access Assess workshops. The practical implication is that choosing whether to engage a partner is also choosing whether to qualify for $200K to $2M of migration funding, and the partner premium often nets out positive once MAP is applied.
Cutover is rarely instantaneous. For any migration over a handful of servers, source and destination environments run in parallel for between 30 days (best case, a well-rehearsed wave with a tight cutover) and 18 months (worst case, a phased migration with a slow decommission of the source data centre). Every month of parallel running adds 1.5 to 2.5 times the monthly on-premise spend, because the cloud environment is also running.
A 100-server estate at $800 per server per month on-premise spends $80,000 per month on source-side infra during parallel running, plus roughly $60,000 per month on the AWS side as workloads land. A six-month parallel-running window adds about $840,000 to the migration bill on infra alone, before considering staff cost and dual-monitoring tooling. This is the single biggest line item buyers underestimate.
Cutover discipline
A representative cost build for a 100-server mid-market migration to AWS with a 12-month programme, mixed strategy (60 percent rehost, 30 percent replatform, 10 percent retire), 20 TB of data, AWS Direct Connect for steady-state, and Snowball Edge for the bulk seed.
Worked AWS migration cost build, 100 servers, 20 TB, 12 months
| Cost line | Low estimate | Typical estimate | High estimate |
|---|---|---|---|
| Assessment and planning (Migration Evaluator + partner) | $45,000 | $80,000 | $150,000 |
| Migration labour, 90 workloads, blended | $640,000 | $1,260,000 | $2,250,000 |
| Tooling (DataSync, DMS, third-party assessment) | $25,000 | $60,000 | $140,000 |
| Snowball Edge devices (4 devices, 60 days total) | $12,000 | $22,000 | $40,000 |
| Direct Connect, port fees and DTO | $18,000 | $38,000 | $80,000 |
| Parallel running (6 months, blended) | $420,000 | $840,000 | $1,400,000 |
| Cutover and downtime contingency | $60,000 | $120,000 | $350,000 |
| AWS Enterprise Support (12 months) | $120,000 | $180,000 | $240,000 |
| Staff retraining (8 engineers) | $24,000 | $45,000 | $80,000 |
| Security rework (IAM, VPC, encryption) | $50,000 | $120,000 | $280,000 |
| Contingency at 15 percent | $215,000 | $415,000 | $760,000 |
| AWS MAP credit (if qualifying) | ($300,000) | ($450,000) | ($800,000) |
| Net total estimate | $1,329,000 | $2,730,000 | $5,010,000 |
The high estimate includes the kind of overrun that public research consistently documents. McKinsey reported in their 2024 cloud cost research that 35 percent of cloud migration programmes ran more than 25 percent over budget; the typical overrun was concentrated in parallel running, refactor scope creep, and Oracle licensing. Treat the typical column as a target, the high column as the band you should still cash-flow plan to.
The free AWS Pricing Calculator is a superb tool for modelling steady-state cloud spend. It is not a migration cost tool. It does not model partner consulting, MAP credits, parallel running, Snowball device fees, training cost, security rework, contingency, or the licence cost premium for Windows and Oracle on EC2 Dedicated Hosts. Teams that build a migration business case from AWS Pricing Calculator alone typically underestimate cost by 40 to 60 percent. The total cost of a migration is steady-state AWS spend plus a one-time migration cost layer that is often larger than the first year of AWS spend itself.
Six levers consistently produce the largest cost reductions on this site shape:
The cost of an on-prem to AWS migration is not fundamentally a cloud pricing question. It is a programme management question wrapped in a pricing question. The teams that come in on budget are not the ones that found a cheaper EC2 instance class; they are the ones that retired 15 workloads they did not need to migrate, compressed parallel running by six weeks, and qualified for the MAP credit at the assessment stage rather than mid-flight. The discipline is repeatable. The cost discipline is what this calculator and the surrounding reference pages on this site are designed to encode.
A. Mid-market scope (100 servers, 20 TB data, 12 month timeline) typically lands at $1.5M to $3M all-in, including labour, egress, parallel running, support, and contingency. Per-server cost ranges from roughly $3,000 (lift-and-shift) to $52,000 (full refactor). Most enterprise migrations use a mixed strategy that averages $12,000 to $18,000 per workload before egress and dual-running overhead.
A. MAP credits do cover that range for qualifying enterprise migrations, but only when the project is scoped with an AWS Premier or Advanced Consulting Partner before the contract is signed. Self-led migrations without a registered partner usually do not qualify. The credits are tied to assess, mobilize, and migrate phases and are paid against agreed milestones, not as cash upfront.
A. Use DataSync when your network can sustain the throughput and the total transfer window is under two weeks. Use Snowball Edge for datasets above 50 TB or where bandwidth is constrained. A Snowball Edge Storage Optimized device holds 80 TB usable and a typical 10-day rental sits at around $300 plus shipping. DataSync charges $0.0125 per GB transferred; for 80 TB that is roughly $1,024 in data transfer fees plus the cost of the source-side AWS bandwidth.
A. Parallel running is the single largest hidden line item. Most teams run on-premise and AWS in parallel for 3 to 18 months. A workload that costs $800 per month on-prem and $600 per month on AWS during the migration adds about $1,400 per server per month of dual-run cost. For a 100-server programme over six months that is roughly $840,000 in dual-run alone, before egress or partner fees.
A. Under 20 workloads with an in-house team that holds AWS Solutions Architect Associate certifications, yes. Above 50 workloads or with a hard cutover deadline, the partner economics usually win net of MAP credits. Self-led migrations do not access MAP funding and do not benefit from partner credits like RISE for SAP or Microsoft Workload acceleration, so the headline saving evaporates quickly at scale.
A. Small workload migrations of 5 to 20 servers complete in 2 to 6 months. Mid-market migrations of 50 to 200 servers typically run 6 to 18 months from kickoff to full cutover. Enterprise data centre exits of 500 servers or more take 18 to 36 months using a wave-based approach. AWS publishes the wave methodology in their Cloud Adoption Framework which is the reference most enterprise programmes anchor to.
AWS migration cost calculator ->
EC2, MAP credits, DataSync, partner rates
On-prem to Azure ->
Hybrid Benefit, FastTrack, the 3-year bill
On-prem to GCP ->
Sustained-use discount, BigQuery, Migrate to GCE
AWS Snowball cost ->
Per-device, per-day, the egress break-even
AWS MGN cost ->
Free-for-90-days and what happens after
Lift-and-shift cost per server ->
The $3K to $8K band, decomposed
100-server worked scenario ->
Mid-market reference build
10 hidden migration costs ->
Detailed playbook with mitigation
Strategy cost tables ->
Lift vs replatform vs refactor
Updated 2 May 2026