Oracle: Sovereignty promise between claim and reality

Contents

At Oracle AI World in Frankfurt, leading Oracle managers presented their plans for a distributed, sovereign, and AI-capable cloud infrastructure. The focus was on how companies can remain operational in the face of growing data protection requirements and an increasingly fragmented cloud landscape. Andrew Bond, CTO of Oracle EMEA, set the stage: Oracle is not just another hyperscaler, but architecturally distinct from the competition. Whether this claim holds up remains to be seen in practice – but the arguments were certainly substantial.

According to Oracle, its EU Sovereign Cloud has been running in Frankfurt and Madrid for almost three years. It is operated exclusively by EU personnel and is physically separated from the public Oracle Cloud network. Technically, it is based on the same OCI building blocks as the public cloud, but runs in its own isolated realm. This is Oracle's response to a key concern of European customers: data should not leave the EU, and operational responsibility is demonstrably within the EU.

Antonio Freitas, responsible for Sovereignty Strategy at Oracle EMEA, explained the development of the offering in a roundtable session: His team worked closely with EU institutions such as the European Commission, ENISA, and national regulatory authorities. The goal: not only to understand the requirements for sovereign cloud infrastructure but also to incorporate them into product development early on. However, it remained open at the event to what extent Oracle is closer to an EUCS certification (EU Cloud Services Scheme) than, for example, AWS or Azure.

A customer example was intended to demonstrate that the concept works in practice: Stefan Moch, VP of Arvato Systems, reported on the operation of drug verification systems in accordance with the EU's Falsified Medicines Directive (FMD). The system processes five to six million pharmacy transactions daily in Germany alone and around 35 to 40 million pack verifications across Europe. High availability and low latency were crucial. These performance metrics could not be independently verified at the event.

Unified Pricing: A Promise with Question Marks

One argument that Bond repeatedly emphasized in his keynote: Oracle charges the same price for cloud services in Europe as in the US. Since data center operations are significantly more expensive in Germany, this means a considerable cost advantage for European customers. Oracle estimates savings of 30 to 50 percent compared to competitors – according to its own figures, not independently verified.

Indeed, publicly available price lists show that AWS and Azure often charge higher list prices for the Frankfurt region than for US regions. Whether Oracle's calculation holds up at this level, taking into account competitors' enterprise discounts, reserved instances, and committed-use models, remains to be seen. Customers should conduct their own TCO comparisons here.

Dedicated Region and Alloy: The Cloud in Your Own Data Center

In addition to the public cloud and the EU Sovereign Cloud, Oracle presented its Dedicated Region and the Alloy model. The Dedicated Region allows companies to operate the complete OCI infrastructure, including SaaS, PaaS, and IaaS, in their own data center, starting from three racks. According to Oracle, Vodafone has consolidated 40 data center locations in six Dedicated Regions, two of them in Germany, using this solution.

The Alloy model takes a different approach: partners can build their own cloud offerings based on OCI. Oracle cited Fujitsu's Japanese Sovereign Cloud and AIS's Thai Cloud as references. Conceptually, Alloy is reminiscent of competitors' offerings such as AWS Outposts, Azure Stack, or Google's Distributed Cloud. However, Oracle differentiates itself through the complete SaaS availability. Oracle emphasizes that all variants are technically the exact same platform.

Multi-Cloud Portability: Bridges to AWS, Azure, and Google

Another focus was Oracle's multi-cloud strategy. The company has integrated parts of its OCI infrastructure directly into the data centers of Microsoft Azure, Google Cloud, and AWS. Oracle database services such as the Autonomous Database can be used natively there. For example, those who want to combine Amazon Bedrock with an Oracle database do not need to migrate data to another environment.

In line with this, Oracle announced Multi-Cloud Universal Credits: a unified consumption model that is intended to make Oracle database services flexibly transferable between OCI, AWS, Azure, and Google Cloud – without new contract negotiations each time. The idea sounds plausible in an era of rapidly changing AI strategies, which have recently led to strong revenue growth in the cloud business, but it also raises questions: How portable are workloads in practice when they rely on Oracle-proprietary features like the Autonomous Database? True multi-cloud portability, even with Oracle, remains tied to its own ecosystem.

Architectural Decisions: Bare Metal and RoCE Instead of Virtualization and InfiniBand

Oracle bases its performance and cost arguments on fundamental architectural decisions. Unlike AWS and Azure, Oracle consistently relies on bare-metal instances: workloads run directly on dedicated hardware without hypervisor overhead – this is intended to particularly benefit database-intensive applications. For network connectivity, Oracle uses RoCE (RDMA over Converged Ethernet) instead of InfiniBand and combines it with its in-house Acceleron architecture to reduce cabling effort for large GPU clusters. However, both decisions also have downsides: without a hypervisor, flexible resource distribution between workloads is lost, and whether RoCE can keep up with the congestion management of InfiniBand in very large cluster topologies is a subject of debate in the industry.

AI Strategy: Data Platform Instead of Model Race

The event concluded with Oracle's AI strategy. With the AI Data Platform, Oracle aims to securely connect private corporate data with generative AI models, rather than developing its own foundation models. The core element is the Autonomous Database, which supports not only relational data but also vector, JSON, graph, and spatial data, and can connect to external data stores via open formats like Apache Iceberg.

The approach of positioning itself as a data platform for AI, rather than competing in the race for its own models, is understandable: most companies struggle less with the availability of models than with the integration and governance of their own data. However, Oracle failed to provide concrete reference customers from German-speaking regions.

Conclusion: Strong Arguments, Open Questions

Oracle presented itself at AI World as a relevant alternative for companies with high demands on data sovereignty and database performance. The Sovereign Cloud architecture, unified pricing, and multi-cloud integration are understandable differentiating factors compared to the three major hyperscalers. Naturally, the central promises remained unverified at such a marketing event. Customers should validate cost advantages based on their own workloads and question how portable their applications truly are if they have invested deeply in Oracle-proprietary features.

(olb)