“Not all clouds are created equal” in the AI era: how is OCI different?

During Oracle AI World in Las Vegas, we had the chance to catch up with Richard Smith, EVP and GM, EMEA Cloud Infrastructure at Oracle. Oracle is one of the key players in global AI infrastructure deployments, with massive Capex spend and a complete overhaul of its entire stack geared toward AI. This flagship event underlined that. We discuss the strategic importance of Oracle’s unique approach to AI, data sovereignty, and cloud architecture with Smith.

In a market where every tech giant is scrambling to claim the title of ‘AI Provider,’ Smith is clear on one thing: “Not all clouds are created equal.” He argues that Oracle’s unique Gen2 architecture and consistent cloud performance offer a distinct advantage over competitors. He does not see an AI bubble and has some interesting…

Subscribe to Techzine Talks on Tour (soon to be the Techzine TV Podcast) and watch and/or listen to our other episodes via Spotify, Apple , YouTube or another service of your choice.

Phenomenal growth in cloud

Oracle’s cloud business is experiencing remarkable growth, with OCI (Oracle Cloud Infrastructure) alone generating approximately $7 billion in commitments last quarter. More significantly, the company’s RPO (Remaining Performance Obligations) has reached several hundred billion dollars, with projections to potentially exceed half a trillion dollars in upcoming earnings statements.

Smith emphasizes that these aren’t commitments from a single customer but from many organizations with significant financial resources. “You wouldn’t sign up if you didn’t believe in the inherent value and the capability that it had,” he notes. This highlights that these long-term commitments demonstrate genuine confidence in Oracle’s cloud platform rather than speculative investment.

The AI Data Platform

One of the most significant announcements at Oracle AI World 2024 is the AI Data Platform. This addresses a critical challenge when it comes to AI. How can organizations use private, sensitive data for AI without creating security risks or requiring massive data movement? As we note during our conversation with Smith, if anyone should be able to answer this question, it’s Oracle. It has a 50-year history as a steward of much of the world’s data, so it would be very disappointing if it didn’t. Smith argues that Oracle certainly has an answer.

The traditional approach requires customers to “scrape the application layer or scrape the data layer, and move it so you can do some level of analytics or some level of manipulation from it,” as Smith describes it. According to him, this is how many competitors do it. With Oracle’s AI Data Platform, that is no longer necessary, he argues. This acts as an intermediary layer that allows specialized language models to access private data for specific purposes. The data never leaves secure database environments.

This approach prevents the “big data” problem from repeating itself with AI. During the big data era, companies created massive, unwieldy data lakes or data oceans that ultimately proved ineffective. Smith argues that attempting to consolidate all data into centralized repositories doesn’t work anyway.

Database vectorization

The new Orace AI Database (26ai) includes vectorization as a native, standard feature rather than an add-on. This integration means customers can deploy AI capabilities without standing up separate infrastructure or moving data to specialized vector databases.

The database supports multiple data models (relational, JSON, structured, and unstructured) all within the same Oracle database environment. This flexibility is crucial because “you don’t necessarily need to take data from one place to put it somewhere else to get inherent value from it. In fact, you shouldn’t necessarily have to do that,” according to Smith.

Vectorization is necessary if you want to use data for AI workloads. However, it also has a rather big drawback. It can significantly increase database sizes due to the nature of vector embeddings. Smith acknowledges this challenge: “you do need to think about that. Because you just don’t want petabytes of immutable data storage at the end of the day. That’s very expensive.” Adding vectorization to the database is interesting, but does it also make the database bloat smaller by using additional smart features to constrain it?

Scale-across architectures are part of AI workloads

Beyond traditional scale-up and scale-out approaches, Oracle also sees that scale-across becomes more important. That enables AI workloads and models to span multiple data centers. The Abilene project, with its closely positioned data center modules, exemplifies this approach, where configurations can function as “one single virtual unit, or they can be multiple units”, according to Smith.

Scale-across offers cost benefits by allowing workloads to be optimally distributed based on latency requirements and data locality needs. While spanning workloads across distant data centers (such as across the Atlantic) introduces latency, Smith notes that “if it takes you two or three seconds to get a response on something, but having the data in those two respective locations is the right thing to do because it keeps the data currency correct, then that’s not a bad thing to do.”

Data sovereignty

Data sovereignty remains a highly emotional and complex issue in Europe. Concerns about data location and control often override technical considerations. Smith observes that “from a technical standpoint, I don’t think that where a company has its headquarters is very relevant to the sovereignty discussion, but for some reason, emotionally, it is for many people.”

Oracle’s position on regulation is notably positive: “for the most part, I think regulation is our friend.” Smith argues that appropriate regulatory guardrails are necessary. Europe’s higher degree of regulation, while expensive to comply with, ultimately benefits the industry. Oracle has been identified as a critical vendor under DORA (Digital Operational Resilience Act). This brings significant compliance obligations including potential audits of processes.

The challenge with new regulations like DORA is that “the regulation is new, and the processes are actually not yet explicitly defined”, according to Smith. Oracle is proactively working with regulatory bodies, offering to help identify appropriate compliance processes rather than simply reacting to requirements.

The AI bubble debate

We cannot have a conversation with someone from Oracle without addressing the concerns about an AI bubble. After all, Oracle is a key player in this discussion, with heavy commitments towards OpenAI and significant Capex spend with Nvidia. Smith offers a nuanced perspective on the debate. First and foremost, he argues that AI is here to stay, regardless of iterative changes in specific models like ChatGPT versions. In that sense, AI is not a bubble.

The definition of “bubble” matters significantly too. From a financial standpoint, Smith acknowledges that massive investment is required. He is convinced that returns will absolutely occur, however, and references Oracle’s substantial RPO figures we talked about earlier.

In general, Smith disputes the bubble concept but acknowledges that “every company today calls itself an AI provider,” suggesting that “there will be some marketplace consolidation, as is natural.” In other words, AI can and will have a big impact on the market as a whole. It is also clear from his answers that he does not think Oracle will be on the wrong side of that argument.

Smith’s final point on the AI bubble matter emphasizes Oracle’s unique positioning. “Oracle’s been in the business of data for 50 years. AI would not exist without data”, he argues. This 50-year foundation in data management, combined with a complete stack from hardware architecture through bare metal, GPUs, vectorized databases, middleware, applications, and agentic tools, creates what Smith calls “a pretty compelling value proposition.”

Podcast player

If you don’t want to or can’t watch the video, it’s possible to listen to the audio-only version below.

Techzine Talks on Tour becomes Techzine TV Podcast

Starting 1 January 2026, Techzine Talks on Tour will be rebranded as the Techzine TV Podcast. As we add video to every podcast we record, we think it makes more sense to cluster everything around our newly-launched Techzine.tv platform. If you don’t like watching conversations but rather listen to them without video, that will remain possible. The Techzine TV Podcast will be available on Spotify, Apple and many other podcast platforms. Just search for Techzine Talks on Tour in your favorite podcast app.

Our commitment to our listeners and viewers hasn’t changed. We aim to publish interesting conversations with leaders in the IT space. Tell us how we do by leaving a comment or rate us. We are always looking to improve, but obviously also like to receive a compliment every now and again.

Previous episodes of Techzine Talks on Tour:

Get in touch

We hope you like this podcast series. If so, please let us know. If you have suggestions on how we can improve, we would like to hear those too. We’re also open to suggestions around specific topics, or specific people that want to appear in an episode of Techzine Talks on Tour. You can find both Coen van Eenbergen and Sander Almekinders on LinkedIn, or you can send an email to info@techzine.eu.