Oracle Exadata On-Premises vs Exa@CC vs Exa@CS: Complete Comparison Guide
Oracle Exadata On-Premises vs Exa@CC vs Exa@CS: Differences, Pros & Cons, Technical Benefits, Use Cases, and ROI Analysis
This guide delivers an in-depth comparison of Oracle Exadata's three deployment options — On-Premises (ExaDB-D), Exadata Cloud@Customer (Exa@CC / ExaCC), and Exadata Cloud Service (Exa@CS / ExaCS on OCI). We cover the latest X11M generation (released January 2025), architectural differences, management responsibilities, technical capabilities, pros & cons for each model, and ROI considerations to help DBAs, architects, and IT leaders make the right deployment decision.
1. What is Oracle Exadata?
Oracle Exadata is a purpose-built engineered system combining scale-out x86-64 compute servers, intelligent storage cells, RDMA-over-Converged-Ethernet (RoCE) networking, NVMe flash, RDMA-addressable memory (XRMEM), and Oracle's proprietary database-aware software stack into a single integrated platform. First introduced in 2008 for on-premises deployment, it has grown into Oracle's gold standard for mission-critical Oracle Database workloads — covering OLTP, data warehousing, analytics, and AI vector processing.
What makes Exadata fundamentally different from commodity hardware is its intelligent storage layer. SQL processing is offloaded from database servers directly to storage cell CPUs via Smart Scan, dramatically reducing I/O. Hybrid Columnar Compression (HCC) compresses data 10–50×. Exadata Smart Flash Cache uses database-aware algorithms to transparently accelerate hot data. These capabilities are embedded in the hardware/software stack and unavailable on generic infrastructure.
Performance improvements of Exadata X11M vs X10M — announced January 7, 2025.
2. Exadata X11M — What's New in January 2025
Oracle launched the X11M (13th generation) simultaneously across all deployment models — on-premises, Exa@CC, and OCI — at the same price as the prior X10M generation. All performance gains are pure additive value to existing customers.
AMD EPYC 9J25 — 96 cores/server, up to 25% faster per-core vs X10M. DDR5 up to 33% faster. Up to 3 TB RAM per database server. Full rack: 2,880 cores, 42 TB memory across all DB servers.
2.2× faster flash throughput reaching 100 GB/s per storage server. 1.25 TB DDR5 per storage server configured as XRMEM (RDMA Memory) — a new caching tier below the database buffer cache and above flash.
Storage-layer offloading reduces network traffic 4.7× via Adaptive Top-K Filtering. IVF (persistent) vector index 55% faster. HNSW (in-memory) vector index 43% faster. Binary vector search 32× faster.
X11M can dynamically cap CPU power consumption and turn off unused cores during off-peak periods (nights, weekends). New Database Server-Z and HC-Z form factors for lighter workloads allow mixed rack configurations.
3. The Three Deployment Models
Oracle delivers the identical Exadata engineered system across three commercially distinct models. The hardware and Oracle Database software are the same. What differs is ownership, physical location, management responsibility, and commercial structure.
4. Head-to-Head Comparison Table
| Dimension | On-Premises | Exa@CC | Exa@CS (OCI) |
|---|---|---|---|
| Hardware owner | Customer | Oracle | Oracle |
| Physical location | Customer DC | Customer DC | Oracle / partner cloud region |
| Hardware management | Customer | Oracle (remote) | Oracle |
| OS / GI / DB management | Customer | Customer (or Auto DB) | Customer (or Auto DB) |
| Commercial model | CapEx purchase | 4-year OpEx subscription | Pay-per-use (per-second) |
| Data residency | Fully on-prem | Fully on-prem | Region-configurable |
| Regulatory compliance | Highest control | Strong (Oracle-managed) | Region-dependent |
| OCI cloud services (control plane, Object Storage) | None | Full OCI | Full OCI |
| Autonomous Database | Not available | Available | Available |
| Elastic compute scaling | Procurement required | Add racks (weeks) | On-demand, per-second |
| Hardware refresh | Customer CapEx | Oracle-managed | Automatic |
| Network latency to on-prem apps | Sub-millisecond | Sub-millisecond | 1–5ms cloud egress |
| Multi-region / multi-AZ HA | 2 physical racks | 2 ExaCC deployments | Native (built-in) |
| BYOL support | Full perpetual | BYOL or Lic-Included | BYOL or Lic-Included |
| Minimum contract term | None (owned) | 4 years | 48-hour minimum |
| Internet connectivity required | Not required | Oracle mgmt tunnel | Required |
| Air-gapped deployment | Supported | Not possible | Not possible |
| Multicloud (AWS/Azure/GCP) | Not available | Via OCI integration | Native (Database@Hyperscaler) |
| EE Extreme Performance (all options incl.) | Licensed separately | Licence-Included option | Included by default |
5. Exadata On-Premises — Deep Dive
Traditional on-premises Exadata is the original model that built Oracle's engineered systems reputation over 16+ years. The customer purchases the hardware outright (or finances it), owns the asset, and takes full operational responsibility. Oracle provides Oracle Support for hardware (typically 7 years from last ship date) and software.
Management Responsibility
On-premises customers manage everything from the data center floor up: physical racking, power, cooling, networking, firmware patching, OS patching, Grid Infrastructure updates, database patching, and backup infrastructure. Oracle's Advanced Support Gateway (ASG) enables remote diagnostics and parts replacement, but the operational model is entirely customer-led.
- Complete control over patch timing, change management, and upgrade cadence
- No external network connectivity required — air-gapped / classified deployments possible
- Existing perpetual Oracle Database licenses used directly — no conversion
- No subscription lock-in; own the hardware asset outright after purchase
- Maximum flexibility for ultra-sensitive workloads (defence, intelligence, government)
- Hardware asset depreciates over 5–7 years — long amortization horizon
- X11M-Z and HC-Z variants allow fine-grained server-level customization
- No Oracle ASG tunnel required — fully isolated network
- Massive upfront CapEx — full X11M rack ~$2.96M hardware alone
- Full operational burden on customer DBA and infrastructure teams
- No OCI cloud services: no control plane, no Autonomous Database
- Scaling requires hardware procurement — weeks to months lead time
- Hardware refresh is customer-funded and customer-initiated CapEx cycle
- High DBA skill requirement — Exadata-specific expertise needed for operations
- Cross-site HA requires two full Exadata systems (no extended rack clustering)
- Customer responsible for physical security, power, cooling, and DC space
6. Exadata Cloud@Customer (Exa@CC) — Deep Dive
Exa@CC is Oracle's hybrid cloud answer for organizations that need data residency or ultra-low application latency but want to escape hardware ownership and maintenance overhead. Oracle ships and installs Exadata hardware in the customer's data center, connects it to the OCI control plane via a secure Oracle Advanced Support Gateway (ASG) tunnel, takes ownership of all infrastructure operations, and the customer retains control of the database tier. Crucially, the customer benefits from full OCI cloud services — including Autonomous Database — without data leaving their premises.
What Oracle Manages vs. What You Manage
| Component | Oracle Manages | Customer Manages |
|---|---|---|
| Exadata hardware (compute, storage, network) | ✔ Oracle | |
| Exadata System Software (ESS) & firmware | ✔ Oracle | |
| OCI control plane (cloud console, APIs) | ✔ Oracle | |
| Physical DC space, power, cooling, cabling | ✔ Customer | |
| Oracle Linux OS (on DB VMs) | ✔ Customer | |
| Grid Infrastructure (GI) | ✔ Customer | |
| Oracle Database software | ✔ Customer | |
| Backup & Recovery strategy | ✔ Customer | |
| All of the above (if using Autonomous DB) | ✔ Oracle (Autonomous) |
Commercial Structure — Billing Model
Exa@CC X11M uses a 4-year subscription term. A fixed monthly infrastructure fee (for the rack) is charged regardless of utilization. Compute (ECPUs) is billed per-second with a minimum of 8 ECPUs per active database VM. From May 2025, all new Autonomous VM Clusters on Exa@CC are ECPU-only, replacing the prior OCPU billing metric. BYOL is available for customers with existing perpetual Oracle Database EE licenses.
- Data stays on-premises — satisfies data residency laws and sovereignty regulations
- Oracle manages infrastructure; customer DBA team is significantly freed up
- Full OCI services: cloud control plane, Object Storage, Autonomous DB on-prem
- No hardware refresh CapEx — Oracle's responsibility within the subscription
- Sub-millisecond latency for co-located on-premises applications
- Path to public cloud: identical architecture makes future migration trivial
- BYOL or Licence-Included — flexible for transitioning organizations
- Oracle Operator Access Control lets customers audit every Oracle remote action
- Rolling patch model (Oracle-driven) preserves availability during updates
- Mandatory 4-year commitment with limited exit provisions — high lock-in
- Fixed infrastructure subscription cost regardless of actual utilization
- Scaling beyond one rack requires provisioning additional ExaCC racks (weeks lead time)
- Higher 5-year TCO than Exa@CS for most workloads (fixed infra fee layers on top)
- Requires DC space, power, cooling, and network for Oracle ASG management tunnel
- Multi-region HA requires two separate ExaCC deployments and subscriptions
- 8 ECPU minimum per database VM makes small workloads uneconomical
- Oracle ASG tunnel required — cannot be fully air-gapped
7. Exadata Cloud Service (Exa@CS) — Deep Dive
Exa@CS is Exadata running in Oracle Cloud Infrastructure's public regions — and now also natively available through Oracle Database@Azure, Oracle Database@AWS, and Oracle Database@Google Cloud (with OCI co-located in hyperscaler data centers for low-latency application-to-database connectivity). Oracle owns and manages all infrastructure. Customers access the full Exadata stack through the OCI console, CLI, or APIs, with elastic scaling and per-second billing.
Two Flavors in OCI
ExaDB-D (Dedicated) — A dedicated Exadata system within a cloud region. Customer controls the infrastructure shape (quarter, half, full rack, or custom X11M configuration). Starting configuration is 2 DB servers + 3 storage servers. Scales independently up to 32 DB servers + 64 storage servers in one environment.
ExaDB-XS (Exascale) — Shared multi-tenant infrastructure model with lower entry cost. Pay-per-VM. Now available in AWS, Azure, and Google Cloud. Supports Data Guard between ExaDB-D and ExaDB-XS for flexible migration paths.
- Elastic scaling — add/remove ECPUs and storage on-demand, billed per-second
- No hardware commitment — test, burst, and decommission without penalty
- Native multi-region Data Guard across OCI availability domains
- Deep integration with all OCI services: compute, AI, Object Storage, Functions, API GW
- All Oracle DB options included in EE Extreme Performance — no separate option licensing
- Available in multicloud: AWS, Azure, GCP with low-latency cross-cloud interconnect
- Autonomous Database at full scale — self-patching, self-tuning, self-securing
- ExaDB-XS lowers entry price dramatically for smaller workloads
- No data center footprint — zero physical infrastructure for customer
- Data leaves the customer's physical premises — conflicts with strict residency laws
- Dependent on internet/cloud connectivity for all operations
- Network latency to on-premises applications (1–5ms cloud egress for OLTP)
- Can be more expensive than on-premises for stable, constant, high-utilization workloads
- Less control over infrastructure maintenance windows — Oracle schedules patches
- Data egress costs accumulate for analytics workloads moving large data sets
8. Core Technical Benefits (All Deployment Models)
The following capabilities are properties of the Exadata platform itself and are available across all three deployment options.
SQL query processing offloaded from database servers to storage cell CPUs. Only qualifying rows and columns traverse the network. Full table scans that take minutes on commodity hardware complete in seconds. X11M storage delivers up to 8.5 TB/s of SQL I/O per rack.
Compresses data 10–50× for warehouse and archive workloads. Fully transparent to Oracle Database — queries run against compressed data in storage without CPU decompression overhead. Simultaneously reduces storage cost and I/O bandwidth requirements.
NVMe flash managed by database-aware caching algorithms that outperform generic SSD cache controllers. On X11M EF storage servers, 4× 6.8 TB PCIe 5.0 flash cards deliver 25.2M IOPS read per rack with sub-millisecond latency.
1.25 TB DDR5 per X11M storage server exposed as an RDMA-addressable caching tier between the database buffer cache and flash. 8K SQL read latency dropped from 17μs (X10M) to 14μs (X11M) — a 21% improvement. RDMA transfers bypass the OS and CPU entirely.
Active-active dual-rail 100 Gb/s Ethernet fabric with RDMA. Server-to-storage memory access bypasses CPU. Up to 14 racks interconnectable in a single cluster. The same high-performance fabric as OCI — consistency across all three deployment models.
Automatically maintained in-memory indexes on each storage cell track the min/max values of columns per storage region. Full table scans automatically skip non-qualifying regions entirely — additional I/O elimination on top of Smart Scan without any DBA configuration.
All Exadata models support Oracle RAC — active-active clustering across database nodes for scale-out performance and continuous availability during planned and unplanned downtime. Unavailable on non-Exadata cloud database services; critical differentiator for 24/7 OLTP.
Fully self-managing database: self-patching, self-tuning, self-securing. On Exa@CC it brings autonomous capabilities inside the customer's data center — a unique capability unavailable on any other on-premises database platform from any vendor.
9. ROI & Total Cost of Ownership
Exadata ROI is driven by three primary levers: performance consolidation (fewer machines doing more work), licensing simplification (EE Extreme Performance includes all options on cloud models), and operational savings from offloading DBA toil to Oracle. The right model shifts materially based on workload size, utilization patterns, and organizational constraints.
5-Year TCO Shape — Indicative Framework
| TCO Factor | On-Premises | Exa@CC | Exa@CS |
|---|---|---|---|
| Initial capital outlay | Very High (~$3M+) | None (subscription) | None (pay-as-use) |
| Fixed ongoing infra cost | Support only (~22%/yr) | High (monthly rack subscription) | None (variable only) |
| Variable compute cost | None | ECPU consumption | ECPU / OCPU per-second |
| DBA headcount impact | Highest (full ops) | Medium (DB+OS only) | Lowest (DB tier only) |
| Hardware refresh cost | Major CapEx every 5–7 yrs | None (Oracle-managed) | None (automatic) |
| Best workload fit | Stable, high-utilization, large | Regulated, on-prem, medium-large | Variable, burst, DevTest, new |
10. Decision Matrix — When to Use What
| Scenario / Requirement | On-Premises | Exa@CC | Exa@CS |
|---|---|---|---|
| Strict data residency law (GDPR, MAS, HIPAA on-prem) | Best | Best | Not suitable |
| Air-gapped / no external connectivity allowed | Best | Not suitable | Not suitable |
| Autonomous Database on-premises | Not available | Best | Available |
| Elastic, variable workloads / DevTest / burst | Limited | Moderate | Best |
| Large existing perpetual Oracle license pool | Best | Good (BYOL) | Good (BYOL) |
| Reduce DBA operational burden | Low | High | Highest |
| Multi-region / multi-AZ High Availability | Manual / costly | Possible (2 racks) | Native |
| Sub-millisecond latency to on-prem applications | Best | Best | Not achievable |
| Lowest 5-year TCO (stable, high-utilization) | Best | Moderate | Moderate |
| Lowest entry cost / fastest provisioning | Highest CapEx | Moderate | Best (ExaDB-XS) |
| Multicloud integration (AWS / Azure / GCP apps) | Limited | Moderate | Native |
| Path to full public cloud migration (future) | Complex | Best stepping stone | Already there |
| No long-term contract lock-in | Owned asset | 4-year minimum | 48-hr minimum |
The 3-Condition Test for Exa@CC
Exa@CC is commercially justified when all three of the following conditions are simultaneously true:
- Data residency or regulatory requirement prevents use of Oracle's public cloud data centers.
- Workload size justifies the Base System infrastructure commitment — typically a meaningful ECPU consumption above the 8 ECPU/database minimum.
- Organizational model supports a 4-year contractual commitment with limited exit provisions.
If any one of these conditions is absent, alternative approaches — on-premises with perpetual licensing, OCI public cloud in a compliant region, or non-Oracle database platforms — warrant serious evaluation before accepting the Exa@CC 4-year commitment.
11. Summary & Next Steps
- All three deployment models run the same X11M hardware and Oracle Database — this is a commercial and operating-model decision, not a technology one.
- On-Premises = maximum control, maximum CapEx, maximum DBA burden. Best for stable, large, long-horizon, regulated, air-gapped workloads.
- Exa@CC = cloud operations model inside your data center. Best for regulated organizations that cannot use public cloud but want Oracle to manage infrastructure. 4-year commitment is the key risk.
- Exa@CS = elastic, no-commitment, pure cloud. Best for variable workloads, DevTest, multicloud architectures, and organizations with no on-premises data residency requirements.
- Before committing: build a full 5-year TCO model, engage an independent Oracle licensing advisor, and run a proof-of-concept on real workloads.
Recommended Next Steps
- Inventory your current Oracle Database license estate and assess BYOL eligibility for both ExaCC and ExaCS.
- Classify your workloads by data residency requirement, latency sensitivity, and utilization pattern.
- Run a 5-year TCO model comparing all three models using Oracle's pricing page and an independent advisor.
- Pilot Exa@CS using ExaDB-XS in OCI (no commitment) to validate performance before any long-term decision.
- If evaluating Exa@CC, negotiate the 4-year contract structure carefully — co-term provisions, exit clauses, and upgrade path to ExaDB-XS (Exascale) are important commercial protections.
- Review MOS Notes on Exadata Cloud@Customer (Doc ID 2066000.1) and the Oracle Exadata Database Machine X11M Data Sheet for detailed technical specifications.
Comments