Oracle ships Exadata in three actively-supported generations in 2026 - X11M (the current default), X10M (2024 vintage) and X9M (2022 vintage). The hardware datasheets read like incremental gen-over-gen improvements, but the commercial implications are not incremental: an Exadata X11M vs X10M upgrade decision can swing a 5-year commitment by $4M-$8M, and the X9M-to-X11M decision is even larger because Oracle bundles the refresh into renegotiation of the Database options pool, the support contract and (on cloud variants) the OCPU/ECPU run-rate. This article compares the three Exadata generations on the dimensions that actually drive 2026 procurement decisions - hardware specs, pricing, performance per Processor licence, BYOL math, and the upgrade-or-hold economics for current X9M and X10M customers.
Oracle Exadata is the only database platform where the hardware refresh decision is also a software licensing decision and (on cloud variants) a multi-year commercial commitment. The choice between X11M, X10M and X9M cascades through five separate financial dimensions: hardware purchase or subscription cost, the Oracle Database Processor licence count required to cover activated cores, the per-rack Premier Support obligation, the ExaCC infrastructure subscription rate, and the BYOL ratio applied against existing perpetual entitlements.
The decision matters most acutely for customers whose X9M hardware lands in the maintenance window in 2026-2027. Oracle's standard play on these accounts is to bundle an X11M refresh with a fresh 60-month commitment, a Database options expansion (Partitioning, Diagnostics Pack, Tuning Pack, Advanced Compression), and a multi-year support renewal at full list. The customer who walks into that conversation without independent generation analysis routinely overpays.
The wider Exadata framework sits in the Oracle Database Licensing Guide; the Cloud@Customer perspective is at Cloud@Customer cost; the on-prem vs cloud framework is at Oracle Cloud Licensing Guide; the multi-cloud Exadata variants are covered in Exadata on OCI vs ExaCC vs Database@Azure.
The three generations differ on CPU vintage, memory technology, storage cell capacity and network fabric. The X11M is the 2025 release based on Intel Sapphire Rapids (2nd revision) with DDR5 memory; the X10M is the 2024 release based on Intel Sapphire Rapids (1st revision) with DDR5; the X9M is the 2022 release based on Intel Ice Lake with DDR4.
The headline numbers per Quarter Rack configuration:
| Component | X11M (2025) | X10M (2024) | X9M (2022) |
|---|---|---|---|
| Database server CPU | 2x Sapphire Rapids Refresh, 96 cores/socket = 192 cores/server | 2x Sapphire Rapids, 96 cores/socket = 192 cores/server | 2x Ice Lake, 32 cores/socket = 64 cores/server |
| Database server memory | 3 TB DDR5 (max 6 TB) | 3 TB DDR5 (max 6 TB) | 1.5 TB DDR4 (max 2 TB) |
| HC storage cell capacity | 192 TB raw HDD + 25.6 TB Flash Cache | 192 TB raw HDD + 25.6 TB Flash Cache | 101 TB raw HDD + 25.6 TB Flash Cache |
| EF storage cell capacity | 89.6 TB NVMe Flash | 89.6 TB NVMe Flash | 51.2 TB NVMe Flash |
| RoCE network fabric | 100 Gb/s RoCE | 100 Gb/s RoCE | 100 Gb/s RoCE |
| PMEM (Persistent Memory) | Deprecated | Deprecated | 1.5 TB PMEM available |
The X11M to X10M delta is small on database server compute (same core count, modest IPC and memory bandwidth gains) but meaningful on storage server capacity (no change in 2026 deliveries) and on the platform's Smart Scan throughput where DDR5-refresh bandwidth helps. The X10M to X9M delta is large: X10M database servers offer 3x the core count of X9M (192 vs 64), 2x the memory ceiling, and X10M HC storage cells offer ~90% more raw HDD capacity than X9M HC cells.
The X9M PMEM (Persistent Memory) layer was the differentiating performance feature of that generation; Intel discontinued PMEM in 2022, and Oracle removed it from X10M onward in favour of larger Flash Cache. X9M-resident workloads that depend on PMEM for sub-millisecond commit latency need an alternative design path on X10M/X11M (typically: more Flash Cache, more memory, application-tier write batching).
Oracle Exadata pricing combines four lines: hardware purchase (on-prem), Premier Support (22% of hardware list per year), Oracle Database Enterprise Edition (EE) Processor licences sized to activated cores, and Database options licences. For ExaCC and ExaCS the model shifts to a monthly infrastructure subscription plus per-OCPU/ECPU consumption.
| Line item | X11M Quarter Rack | X10M Quarter Rack | X9M Quarter Rack |
|---|---|---|---|
| Hardware list price (on-prem) | ~$1.20M-$1.40M | ~$1.05M-$1.20M (where available) | ~$850K-$950K (refurbished or remaining stock) |
| Premier Support / year (22%) | ~$264K-$308K | ~$231K-$264K | ~$187K-$209K |
| EE Processor licences (if BYOL, max 24 cores activated) | ~$570K list (24 x $47,500) | ~$570K list (24 x $47,500) | ~$570K list (24 x $47,500) |
| ExaCC infrastructure subscription / month | ~$130K-$155K | ~$110K-$130K | ~$95K-$115K |
| ECPU rate (Database EE BYOL on ExaCC) | $0.67/ECPU/hour | $0.67/ECPU/hour | $0.67/ECPU/hour |
Note that the EE Processor licence cost is identical across generations - Oracle's Core Factor (0.5 for Intel x86) is generation-agnostic. The per-ECPU/OCPU rate on ExaCC and ExaCS is also generation-agnostic at list. The generation premium lives entirely in the hardware line and the resulting Premier Support line.
For ExaCC customers, the X11M infrastructure subscription carries roughly an 18-22% premium over X10M at list. The customer can usually negotiate the X11M premium down to single-digit percentages by referencing X10M as the alternative quote and by committing to longer-term consumption ramp. The negotiation lever set for ExaCC is in Cloud@Customer cost.
We build the full 5-year TCO comparison across X11M, X10M and X9M for your actual workload - including BYOL math, support extension scenarios and the ExaCC subscription rate the X11M premium should land at. Fixed-fee, 2-3 weeks. Typical avoidance on a refresh deal: 12-25 percent of the proposed Oracle quote.
Oracle Database Enterprise Edition is licensed per Processor with the Core Factor (0.5 on Intel x86) applied to physical cores. The number of licences required scales linearly with activated cores - not with the platform's gen-over-gen IPC improvement.
This creates an arithmetic gap that favours newer generations on a per-Processor-licence basis:
| Generation | Cores per DB server | EE Processor licences (per server, full activation) | Indicative DB TPS per Processor (relative) |
|---|---|---|---|
| X11M | 192 | 96 (192 x 0.5) | 1.00 baseline |
| X10M | 192 | 96 (192 x 0.5) | ~0.85 (15% lower IPC, DDR5 v1) |
| X9M | 64 | 32 (64 x 0.5) | ~0.55 (Ice Lake) |
The X9M-to-X11M IPC gain compounds with the per-server core increase: an X11M database server delivers roughly 5.5x the database throughput of an X9M database server, while requiring 3x the EE Processor licences (96 vs 32). The throughput-per-licence ratio improves by ~83% on X11M vs X9M.
The customer running a stable workload that fits in an X9M Quarter Rack does not benefit from this ratio improvement - the X11M throughput excess sits idle, and the larger licence pool is a sunk cost. The customer running an X9M deployment at 75%+ utilisation with growth projection benefits significantly: the X11M absorbs the growth without a second rack, and the per-licence throughput improvement reduces the licence pool needed to cover the same workload.
On ExaCC and ExaCS the perpetual Oracle Database EE Processor licences cover OCPU or ECPU activation at the standard BYOL ratios: 1 EE Processor = 2 OCPUs = 4 ECPUs. The BYOL ratio is identical across X11M, X10M and X9M - the perpetual licence pool covers any generation.
The implication for refresh decisions: a customer with 48 EE Processor licences covering an X9M ExaCC deployment (max 192 cores activated x 0.5 Core Factor = 96 Processors equivalent capacity - constrained by the customer's actual 48-licence pool to 96 ECPUs / 48 OCPUs) carries that same 48-licence pool forward to an X11M ExaCC deployment with identical BYOL coverage.
The customer does NOT need to purchase additional EE licences to move from X9M to X11M ExaCC at the same activation level. Oracle sales will sometimes frame the refresh as triggering a licence true-up - this framing is incorrect for like-for-like activation. The licence pool covers whatever generation the ECPU/OCPU consumption runs on.
Where additional EE licences are required is on the activation expansion: if the customer takes the X11M refresh as an opportunity to activate more cores (because the X11M can accommodate them and the workload needs them), the incremental ECPU consumption above the BYOL coverage pool defaults to License-Included pricing. The Database options pool (Partitioning, Diagnostics Pack, Tuning Pack, Advanced Compression, etc.) follows the same logic: same generation-agnostic ratio, same expansion mechanics. The full BYOL framework is in Multi-cloud BYOL rules.
For an X9M-resident customer, the X11M refresh business case rests on five drivers. The refresh pays back when at least two are true, and is hard to justify when none apply.
The customer whose X9M deployment is well-sized, well-supported through 2028, with no EA renewal, no ExaCC migration plan and no storage pressure has no compelling refresh case. The right answer is to extend support, defer the refresh, and revisit at the next natural breakpoint (typically Premier Support expiration in 2028-2029).
On ExaCC, the generation choice affects three contract dimensions: the infrastructure subscription rate (X11M premium of ~18-22% over X10M at list), the available shape catalogue (Quarter, Half, Full Rack on each generation; Eighth Rack on X9M only), and the available storage cell mix.
For ExaCS on public OCI, the generation choice is largely invisible to the customer - Oracle controls the underlying hardware refresh and the customer consumes ECPUs at the published rate regardless of which generation the back-end runs on. The X11M ExaCS shapes appear in the OCI console as expanded shape options; existing ExaCS X10M/X9M VM clusters continue to run on the generation they were provisioned on.
On Oracle Database@Azure, @AWS and @Google Cloud, the underlying Exadata generation is also abstracted - the customer sees ECPU and storage capacity, not the hardware generation. New 2026 Database@Hyperscaler regions land on X11M; regions that opened earlier remain on X10M and X9M with no customer impact on rate. The full multi-cloud framework is in Exadata on OCI vs ExaCC vs Database@Azure and Database@AWS guide.
The wider Oracle negotiation framework is in the Oracle Negotiation Guide; the support-cost-reduction framework is in Oracle Support Cost Reduction Guide; the optimisation framework is in Licence Optimisation Guide.
Rarely on hardware grounds alone. The X9M database servers carry Premier Support through 2028 and Extended Support beyond. An X11M upgrade returns value only when (a) you are growing compute capacity 40%+ and the per-OCPU improvement avoids buying additional racks, (b) you are negotiating an EA renewal where the X11M refresh is bundled as a discount lever, or (c) you are migrating to ExaCC or ExaCS and want X11M-class performance under cloud commercials. Pure refresh of working X9M hardware before 2027 is hard to justify.
X11M database server: 192 Intel Sapphire Rapids cores (2 sockets x 96 cores). X10M database server: 192 cores (2 sockets x 96 cores, prior Sapphire Rapids generation). X9M database server: 64 Ice Lake cores (2 sockets x 32 cores). The X11M and X10M look identical on raw core count - the gen-over-gen gain is roughly 18-22% on IPC and significantly higher memory bandwidth (DDR5 vs DDR5 v1).
No. The perpetual Oracle Database EE Processor licences move with the customer, not the hardware. The same 48 Processor pool that covered an X9M deployment covers an X11M deployment - subject to the Core Factor rules (Intel x86 = 0.5). The hardware refresh does not trigger a new commercial event on BYOL. The licence count required scales with the activated cores on the new platform, not the platform name.
X11M HC (High Capacity) storage server: 192 TB raw HDD plus 25.6 TB Flash Cache. X11M EF (Extreme Flash) storage server: 89.6 TB raw NVMe Flash. X11M XT (Extended) storage server: 252 TB raw HDD with 8 TB Flash Cache for archive/colder data. Compared to X9M HC (101 TB HDD + 25.6 TB Flash), the X11M HC nearly doubles capacity per cell. Storage server capacity is the dimension where X11M differentiates most from X10M and X9M.
Yes. Oracle continues to ship X9M-based ExaCC racks where customers have standardised on that generation and the X9M Premier Support window remains open. New ExaCC orders in 2026 default to X11M, but X10M and X9M remain orderable for compatibility with existing fleets. Mixed-generation ExaCC fleets are operationally supported - the OCI control plane abstracts the underlying hardware, and OCPU/ECPU consumption is normalised across generations.
Twice a month. Oracle cloud, DRCC, ExaCC contract patterns, audit-defence tactics and BYOL maths. Written by former Oracle insiders.
No spam. Unsubscribe any time. Independent - not affiliated with Oracle Corporation.