Mainframe modernisation sits at the extreme end of cloud migration complexity. The programmes are large, long, expensive, and high-risk. The technology stack is fundamentally different from cloud-native architectures, the talent base is shrinking, and the business criticality of mainframe workloads is typically the highest in the enterprise. This page is the 2026 cost reference for mainframe modernisation across the three primary paths: rehost, replatform, and re-engineer.
IBM mainframes still process roughly 70 percent of all financial transactions globally and host core systems for most large banks, insurers, airlines, retailers, and government agencies. The reasons mainframes persist are structural: the workloads are business-critical and well-understood, the cost of replacement is high and risky, the performance and reliability characteristics are exceptional, and the talent base, while shrinking, is sufficient to keep the systems running. Mainframe modernisation programmes exist not because the technology is failing but because the talent base is constraining, the per-MIPS run cost is high, and the integration patterns with modern systems are increasingly difficult.
The economic case for modernisation typically draws on three threads. First, the per-MIPS run cost on mainframe (typically $1,500 to $4,000 per MIPS per year all-in) is materially higher than the equivalent cost on cloud-replatform compute (typically $200 to $600 per equivalent MIPS). Second, the shrinking COBOL talent pool drives wage inflation and operational risk. Third, integration with modern systems (APIs, event streams, real-time analytics) is harder on mainframe than on cloud-native platforms. These three threads combine to a long-term cost reduction case that can justify the substantial one-time modernisation cost.
The mainframe modernisation literature recognises three primary paths. Each has a different cost shape, a different timeline, and a different end-state.
Mainframe modernisation paths, mid-enterprise estate
| Path | What it does | Cost band | Timeline | Risk |
|---|---|---|---|---|
| Rehost (emulator) | Run COBOL as-is on cloud-hosted emulator (AWS MM, Micro Focus) | $5M to $15M | 18 to 30 months | Low |
| Replatform (recompile) | Recompile COBOL on cloud-native runtime; minor refactoring | $8M to $20M | 24 to 36 months | Medium |
| Re-engineer (rewrite) | Rebuild functionality as Java/C# microservices | $20M to $50M+ | 36 to 60 months | High |
| Repurchase (SaaS replacement) | Replace mainframe app with SaaS equivalent | $3M to $20M | 12 to 36 months | Medium |
| Retain (sustain on mainframe) | Continue running on mainframe, modernise integration only | $1M to $5M per year | Ongoing | Low |
| Retire (decommission) | Decommission the workload; preserve data only | $500K to $5M | 12 to 24 months | Low |
AWS launched Mainframe Modernization Service (MM) in 2022 as a managed runtime for COBOL and PL/I applications. MM supports two patterns. First, replatform: recompile COBOL to run on a Micro Focus or Blu Age runtime hosted on EC2-like managed compute. Second, refactor: automated COBOL-to-Java conversion via the Blu Age engine, producing Java code that runs on EC2 or container services. The service is documented on the public AWS Mainframe Modernization page.
MM pricing has two billing models. Single-Tenant Replatform uses dedicated compute units (M2 instance hours) billed similar to EC2; current pricing is roughly $0.20 to $4.00 per hour per compute unit depending on size. Multi-Tenant Refactor uses a different model with managed runtime included. The data layer (Db2 z/OS to managed Db2 LUW or PostgreSQL, VSAM to object storage) carries separate cost.
A typical mid-enterprise mainframe workload (5,000 MIPS effective, 100 GB application footprint, 10 TB data) modernised via AWS MM Replatform typically runs $50K to $400K per month at steady state, compared with $750K to $1.5M per month on the mainframe. The monthly saving is real and compounds over the long-term horizon; the one-time modernisation cost typically pays back in 18 to 36 months at this scale.
Micro Focus Enterprise Server is the dominant commercial COBOL runtime for cloud replatform. It supports IBM CICS, IMS, JCL, and most COBOL dialects with high fidelity. Enterprise Server can run on AWS, Azure, GCP, or on customer-owned infrastructure. The replatform pattern: recompile the mainframe COBOL on Enterprise Server, port the data (Db2 z/OS to Db2 LUW or PostgreSQL, VSAM to managed file storage), redirect interfaces, and cutover.
Micro Focus licensing typically lands at $200K to $800K per year for a mid-enterprise estate, depending on the equivalent MIPS volume. The licence is the primary trade-off for the replatform pattern: customers continue to pay a Micro Focus licence in perpetuity in exchange for keeping the COBOL code intact and minimising rewrite risk. For estates where the long-term plan is to re-engineer eventually, Micro Focus replatform is often used as an intermediate step that removes the mainframe but retains the option to rewrite later at a lower-risk pace.
Full re-engineering rebuilds the mainframe application logic in a modern language (Java, C#, Python) as a microservices architecture. The end-state is a fully cloud-native application that has no mainframe dependency, no Micro Focus licence, and modern maintainability characteristics. The cost is the highest of any path.
Three automated COBOL-to-Java conversion tools dominate the market: Blu Age (acquired by AWS), TmaxSoft OpenFrame, and EvolveWare. Each takes COBOL source code as input and produces Java source code as output, preserving business logic but transforming the structure. Conversion automation typically achieves 60 to 85 percent code coverage; the remaining 15 to 40 percent requires manual conversion or refactoring.
The challenge with automated conversion. The resulting Java code mirrors the COBOL structure: procedural rather than object-oriented, with global state, with COBOL idioms like PERFORM and GO TO translated to Java equivalents that are syntactically valid but stylistically awkward. Long-term maintainability requires human refactoring to bring the code up to standard Java patterns; the refactoring effort can match or exceed the original conversion effort. Most successful re-engineering programmes treat the automated conversion as a starting point and budget separately for the human refactoring follow-on.
The hidden cost of re-engineering
A representative cost build for a mid-enterprise mainframe modernisation programme: 12 mainframe applications (3 to be re-engineered, 5 to be replatformed on AWS MM, 2 to be retired, 2 to be sustained on mainframe), 5,000 MIPS effective workload, 36-month programme, blended onshore-offshore partner team.
Worked mainframe modernisation cost build, 12 applications, 36 months
| Cost line | Low estimate | Typical estimate | High estimate |
|---|---|---|---|
| Assessment and portfolio rationalisation | $800,000 | $1,500,000 | $3,000,000 |
| Re-engineering, 3 applications | $8,000,000 | $15,000,000 | $28,000,000 |
| Replatform on AWS MM, 5 applications | $3,500,000 | $7,000,000 | $13,000,000 |
| Retirement, 2 applications (data preservation) | $800,000 | $1,500,000 | $3,500,000 |
| Sustain plan, 2 applications (modernise integrations only) | $600,000 | $1,200,000 | $2,500,000 |
| Mainframe parallel running (36 months) | $8,000,000 | $15,000,000 | $25,000,000 |
| AWS MM consumption (replatform workloads, 24 months parallel) | $1,200,000 | $2,400,000 | $4,500,000 |
| Data migration (Db2 z/OS to managed Db2/PostgreSQL) | $600,000 | $1,400,000 | $3,200,000 |
| Testing infrastructure and automated test cycles | $1,000,000 | $2,200,000 | $4,800,000 |
| End-user retraining and adoption | $200,000 | $500,000 | $1,200,000 |
| Cutover and hypercare (per workload) | $800,000 | $1,800,000 | $4,500,000 |
| Programme management and PMO (36 months) | $2,400,000 | $3,600,000 | $5,400,000 |
| Contingency at 15 percent | $4,200,000 | $8,200,000 | $15,500,000 |
| Net total estimate | $32,100,000 | $61,300,000 | $114,100,000 |
The typical-column number, $61M over 36 months, illustrates the magnitude of mid-enterprise mainframe modernisation programmes. Three workloads being re-engineered drive $15M of the total. Five replatform workloads drive $7M. Parallel running of the mainframe drives $15M. The cost discipline that brings the typical column toward the low column is portfolio rationalisation: identifying as many retirement and sustain candidates as possible, and choosing replatform over re-engineering wherever the business case allows.
Mainframe modernisation programmes face a talent challenge that other migration types do not. COBOL expertise is concentrated in an ageing workforce; new COBOL developers are rare. Replacing departing COBOL talent at the required scale is increasingly difficult and expensive. This creates time pressure on mainframe modernisation: the longer the programme is deferred, the harder and more expensive it becomes to staff.
The SI partner market for mainframe modernisation has consolidated around a small number of specialist firms (Cognizant, TCS, Infosys, Wipro on the offshore side; Accenture, Deloitte, Capgemini on the onshore side; specialist firms like Modern Systems and Asysco for the pure COBOL-conversion work). Day rates for COBOL-expert consultants typically run $1,400 to $3,000 onshore and $400 to $900 offshore. The blended team structure for mid-enterprise programmes typically uses 70 to 85 percent offshore staffing with onshore architects and programme management.
Mainframe modernisation is the most expensive, longest, and highest-risk shape of cloud migration. The economic case is real but the programme discipline required is unusual: a 3 to 5 year programme structure, deep specialist talent, careful portfolio rationalisation, and a willingness to use the full range of paths (rehost, replatform, re-engineer, retire, sustain) rather than forcing every workload into a single approach. The teams that succeed treat mainframe modernisation as an enterprise transformation programme with a cloud destination, not as a technology migration.
A. Mainframe modernisation programmes typically cost $5M to $50M for mid-enterprise estates over 18 to 48 months. The cost shape depends entirely on the chosen path: rehost (emulator on cloud) is cheapest at $5M to $15M; replatform (refactor COBOL to Java with automated tooling) costs $10M to $30M; full re-engineer (rebuild on modern microservices) costs $20M to $50M+. Most enterprise programmes are 3 to 5 year journeys spanning multiple workloads and multiple paths.
A. AWS Mainframe Modernization (MM) is a managed runtime for COBOL and PL/I applications, supporting two patterns: replatform (recompile COBOL to run on Micro Focus or Blu Age runtime on EC2) and refactor (automated COBOL to Java conversion via Blu Age). MM pricing has both consumption-based (per vCPU-hour) and capacity-based (per-MIPS reservation) models. A typical mid-enterprise mainframe workload runs $50K to $400K per month on MM, materially lower than the equivalent mainframe MIPS cost.
A. Micro Focus Enterprise Server is the dominant commercial COBOL runtime for cloud replatform. The licence cost typically lands at $200K to $800K per year for a mid-enterprise estate, plus AWS or Azure compute on top. The replatform migration cost (recompile COBOL, test, validate, cutover) typically lands at $3M to $12M for a mid-enterprise mainframe over 18 to 30 months. The ongoing licence cost is the trade-off for keeping COBOL code intact.
A. Full re-engineering rebuilds the mainframe application logic in Java (or C#, Python) as a modern microservices architecture. The cost is the highest of any path: typically $15M to $50M+ for mid-enterprise estates. Automated COBOL-to-Java conversion tools (Blu Age, TmaxSoft, EvolveWare) reduce the cost by 30 to 60 percent versus manual rewrite, but the resulting Java code typically needs human refactoring to be maintainable long-term. Most re-engineering programmes run 3 to 5 years.
A. Many mainframe workloads are best served by retirement, not modernisation. Application portfolio rationalisation typically identifies 15 to 30 percent of mainframe workloads that should be retired (the business function no longer exists, the workload is rarely used, or the data is preserved elsewhere). Retirement costs are typically $500K to $5M per workload for data archive, regulatory preservation, and orderly shutdown. The retirement saving compounds: every retired workload removes a permanent run-cost line item.
A. Single workload modernisations of small COBOL applications (under 100K lines of code) complete in 6 to 18 months. Mid-enterprise programmes covering 5 to 20 mainframe applications run 24 to 48 months. Large enterprise programmes covering complete mainframe estates can run 5 to 10 years. The dominant timeline factor is testing: COBOL applications are typically business-critical, business-process-aligned, and have decades of accumulated edge cases that need verification on the modernised platform.
SAP HANA to cloud ->
Another high-cost, multi-year workload
Oracle to AWS ->
Licence economics dominate
Data centre exit ->
Wave planning, multi-workload programme
On-prem to AWS ->
MAP, EC2, Snowball
On-prem to Azure ->
Hybrid Benefit, FastTrack
Refactor vs replatform ->
Strategy cost decision
Strategy cost tables ->
7Rs framework
10 hidden costs ->
Detailed playbook
Updated 2 May 2026