It depends heavily on whether the buyer is looking at new, refurbished, or upgraded existing hardware. An entry Power server typically starts in the high four figures to low five figures, while multi-core enterprise configurations and full new deployments can run well into six figures once storage, networking, and infrastructure are included.
Refurbished and used systems usually land 40 to 70 percent below comparable new hardware pricing, depending on generation, condition, and warranty terms. The most useful first step is sizing actual workload requirements, since price comparisons only make sense against a specific configuration.
Open question page
The honest answer depends on remaining IBM support runway, budget, and how critical the workload is. New hardware carries the longest support and warranty window but at the highest price. Refurbished hardware can deliver most of the same capacity at a meaningfully lower price, provided the seller tests and warranties the unit properly.
Buyers should compare total cost of ownership over the realistic hardware lifespan, not just sticker price, and should always confirm the refurbished seller's testing process before assuming price alone makes the decision.
Open question page
Start with current CPW utilization, memory headroom, and partition count on the existing server rather than shopping by model number. A hardware partner can translate that data into processor core, memory, and capacity-on-demand recommendations for a specific target model.
Growth planning matters too. Sizing only for today's workload often means another hardware conversation within a year or two, so most buyers size with some deliberate headroom built in.
Open question page
No. PowerVS extends the Power servers you already run into IBM Cloud without changing the operating system, the Db2 for i database, or the RPG and COBOL applications running on it. watsonx connects to that environment as a separate AI layer through APIs and data connectors, so the ERP and the business logic underneath it stay exactly where they are.
The realistic path is incremental: extend infrastructure into PowerVS where it makes sense, connect watsonx to a specific, narrow use case, and prove it against real data before expanding scope. A full platform replacement is a much higher-risk move than extending what already works.
Open question page
PowerVS is IBM Power infrastructure hosted in IBM Cloud, running the same IBM i, AIX, or Linux on Power operating environments and the same Db2 for i engine as an on-premises server. It is an extension of the platform, not a replacement for it.
Migrating off Power means moving a workload to a different architecture entirely, which usually means rewriting or re-platforming the applications running on it. PowerVS avoids that by letting a workload move to the cloud, or simply extend into it for capacity, disaster recovery, or a bounded AI project, while the applications, security configuration, and high availability design carry over unchanged.
Open question page
Power9 has passed two separate IBM lifecycle milestones, not one. IBM withdrew Power9 from marketing on October 20, 2023, meaning it can no longer be purchased new. Its End of Standard Service date, when IBM stopped developing new firmware, PTFs, and fixes for the platform, was January 31, 2026.
A Power9 system does not stop working on either date. What changes is that new hardware is no longer available and IBM is no longer issuing new code for the platform. Shops still running Power9 in production should have a third-party maintenance contract, a migration plan, or both.
Open question page
Most shops move from Power9 directly to Power10 or Power11 rather than upgrading generation by generation, since IBM i, RPG, COBOL, and Db2 for i all carry forward across Power generations without an application rewrite. The real project work is confirming your IBM i release is supported on the target generation, checking HMC version compatibility, and planning a tested data migration and cutover window.
Downtime is required for the actual cutover, but with proper planning that window can be scheduled during off-hours instead of forcing an unplanned outage. Going straight to Power11 is typically simpler than a two-step move through Power10 first, since it avoids migrating the same workload twice.
Open question page
P Series language usually points to AIX or Unix workloads, while iSeries language usually points to IBM i and AS/400 workloads. Modern IBM Power hardware can be relevant to both, but the operating system, firmware level, licensing, and application support decide the correct path.
For a real buying decision, confirm whether the workload runs AIX, IBM i, Linux on Power, or a mix before comparing model numbers.
Open question page
Choose Power 11 when the project is strategic, the workload needs the longest runway, or the business is already spending enough on migration that it should avoid shortening the next lifecycle window.
Power 10 can still be rational when pricing, availability, existing standards, or known workload fit make it the better near-term decision. Buyers should compare hardware price, software licensing, support horizon, and migration cost together rather than treating generation alone as the answer.
Open question page
Gather the evidence that describes workload behavior, not only the current machine label.
- Partition-level processor use
- Memory allocation and pressure
- Disk busy
- I/O latency
- Storage layout
- Backup window
- Network and adapter needs
- HMC level
- Firmware
- Operating system releases
- Named peak events such as month-end or inventory close
The useful goal is to separate current average demand, current peak demand, and expected growth. That lets the quote solve the workload instead of only replacing the current model with a newer label.
Open question page
AIX and IBM i can run in separate logical partitions on IBM Power hardware when the shared design is planned correctly.
- System fit
- Licensing
- Releases
- HMC
- VIOS
- Storage
- Resource allocation
The safe design documents the ownership and recovery model before the shared hardware is purchased.
- Partition role
- Processor and memory allocation
- I/O path
- Storage dependency
- Backup scope
- Monitoring owner
- Recovery expectation
Open question page