IBM’s Power processor line has always sold itself on a different pitch than x86 commodity servers: fewer, more reliable systems for the workloads a business genuinely cannot afford to lose — core banking, ERP, retail transaction processing. Power11, announced in July 2025 and now the generation enterprises are actively evaluating and deploying through 2026, pushes that pitch further by wrapping in built-in quantum-safe cryptography, automated ransomware detection, and on-chip AI inferencing. This explainer breaks down what actually changed in Power11 versus Power10, based on IBM’s own announcements, and separates vendor performance claims from independently confirmed specifications.
At a Glance
| Attribute | Power11 detail |
|---|---|
| Announced / GA | July 8, 2025 / July 25, 2025 (most systems); PowerVS cloud July 15, 2025 |
| Process node | Refined 7nm (Samsung), not a new 5nm node |
| Cores per chip | 16 per die; up to 15 active for workloads, 1 automatic hot-spare core |
| Top clock speed | Up to 4.4 GHz (Power E1180); up to 4.15 GHz (Power S1122) |
| Memory | High-speed DDR5; some systems also support DDR4 at reduced bandwidth |
| Uptime claim | 99.9999% availability (single Power E1180, IBM-measured, unplanned downtime) |
| AI acceleration | On-chip inferencing; first Power server to support the IBM Spyre Accelerator (Q4 2025) |
| Security | Built-in NIST-approved quantum-safe cryptography; Power Cyber Vault ransomware alerting in under 1 minute |
The Chip: Incremental Silicon, Not a New Node
Power11 is not a ground-up redesign. Per IBM’s own briefings, it stays on a refined 7-nanometer Samsung process rather than jumping to 5nm, and keeps the same 16-core-per-die layout as Power10. What’s new is how IBM uses that 16th core: up to 15 cores are available for AIX, IBM i, or Linux workloads, and the 16th is a spare that automatically activates if an active core fails, with no manual intervention or downtime, according to IBM. Clock speeds move up modestly — the two-socket Power S1122 reaches up to 4.15 GHz versus a flat 4 GHz on the Power10-based S1022, and the high-end Power E1180 reaches 4.4 GHz versus 4.15 GHz on the E1080. IBM also credits 2.5D integrated stacked-capacitor packaging and other thermal engineering for the efficiency gains, and adds a new Energy Efficient Mode that trades 5–10% of core performance for up to 28% lower energy use, a figure IBM states relative to its own Maximum Performance Mode.
IBM’s headline performance claims — up to 45% more capacity than Power10 in entry and midrange models, up to 55% more core performance than Power9, and roughly 2x the performance per watt of “comparable” x86 servers — are IBM’s own vendor-reported figures, measured against specific IBM and HPE server configurations rather than an independent third-party benchmark suite. They’re directionally useful for understanding IBM’s positioning, but procurement teams evaluating a specific workload should ask IBM for configuration-matched benchmark data rather than taking the headline percentages at face value.
Reliability and Security: The Real Differentiator
Where Power11 most clearly tries to separate itself from x86 alternatives is uptime and security guarantees. IBM states Power11 delivers 99.9999% availability, based on unplanned downtime measured on a single Power E1180 system, and claims zero planned downtime for maintenance — though IBM’s own footnotes note that some maintenance activity may still require Live Partition Mobility to move workloads between systems, so “zero planned downtime” describes the workload’s experience, not the complete absence of any maintenance operation.
On security, Power11 ships with built-in, NIST-approved quantum-safe cryptography, which IBM positions as protection against “harvest-now, decrypt-later” attacks where encrypted data is stolen today to be decrypted once quantum computers mature, as well as against firmware-integrity attacks. The IBM Power Cyber Vault feature guarantees a ransomware alert within one minute of detected anomalous activity; it’s worth noting precisely what that guarantee covers, per IBM’s own terms — the guarantee is for the alert itself, with remediation limited to drive replacement up to the covered product’s cost and subject to IBM’s stated terms, not a broader breach-recovery guarantee. Other security features include immutable snapshots, autonomous patching and workload migration that can occur without IT intervention, and integration with IBM Concert, which uses generative AI to help identify risks and automate security patching.
AI: On-Chip Inferencing and the Spyre Accelerator
Power11 supports on-chip inferencing, letting enterprises run AI models directly against data that lives on-premises or in a hybrid-cloud Power environment rather than moving it elsewhere first. IBM states Power11 is the first Power server generation to support the IBM Spyre Accelerator, a dedicated AI system-on-a-chip that IBM also designed to be compatible with its IBM z17 mainframe and LinuxONE 5 platforms, with availability targeted for Q4 2025. Power11 also supports IBM’s watsonx AI platform and Red Hat OpenShift AI for running and managing models. It’s worth being precise about what “AI chip” means here: Power11’s own silicon provides on-chip inferencing acceleration, while the heavier AI acceleration workload is intended to run through the separate Spyre accelerator attached to the system — the two are complementary, not identical.
The Server Lineup
- Power S1122 / L1122: Two-socket systems; the “L” variant is lower-cost and Linux-only.
- Power S1124 / L1124: Larger scale-out systems in the same family.
- Power E1150: Midrange enterprise system.
- Power E1180: High-end flagship, the system IBM’s uptime and top clock-speed figures are measured on.
- Power S1112: Entry-level, half-width, single-socket system; reported to be planned for 2026 as the successor to the Power10-based S1012.
IBM did not disclose pricing at launch, and Power11 is explicitly not IBM’s mainframe line — it’s a separate product family from IBM Z, despite sharing some underlying design philosophy and, via Spyre, some AI accelerator compatibility.
Bottom Line
Power11 is best understood as a disciplined, reliability-and-security-focused refresh rather than a dramatic architectural leap: the same 16-core, 7nm-class silicon as Power10, with meaningful but incremental clock-speed gains, wrapped in genuinely new quantum-safe cryptography, faster ransomware detection, and a path to dedicated AI acceleration via Spyre. For enterprises already standardized on Power for mission-critical workloads, it’s a straightforward, lower-risk upgrade. For organizations evaluating Power against x86 or cloud-native alternatives for new workloads, the decision will hinge less on raw clock speed than on how much IBM’s uptime, security, and AI-readiness claims are worth relative to the total cost of the platform — a number IBM has not published and that varies by configuration.
FAQ
Is Power11 built on a new manufacturing process?
No. It uses a refined 7nm Samsung process, the same node family as Power10, rather than moving to 5nm.
How many cores does a Power11 chip actually have?
16 cores per die, but IBM reserves one as an automatic hot spare; up to 15 are available to run workloads, and IBM has stated there is no way for customers to manually activate all 16.
Is Power11 the same as IBM’s mainframe (Z) line?
No. Power and IBM Z are separate processor families with different target workloads; Power11 is IBM’s Unix/Linux/IBM i enterprise server line, not a mainframe.
What is the IBM Spyre Accelerator and does Power11 include it?
Spyre is a separate AI accelerator chip, targeted for availability around Q4 2025, that Power11 is the first Power generation to support; it is an add-on accelerator rather than part of the base Power11 chip itself.
Has IBM published Power11 pricing?
Not at launch. IBM confirmed general availability dates for most systems (July 25, 2025) and PowerVS cloud (July 15, 2025) without publishing list pricing in its initial announcements.
