Would it also be conceivable, for example, for AWS to set up its data center machines in the Oracle data centers and operate them there for its customers?
Not an uncharming thought, but that would be purely speculative. This would definitely not be decided in Germany, but would clearly be decided by our central engineering teams. Various aspects would have to be assessed, for example security, etc.
In general, there has been quite a cultural change at Oracle in recent years—a move towards openness. When I think back to how Larry Ellison used to berate AWS and Microsoft just a few years ago and today they’re cooperating.
And vice versa, of course. I can still remember from my Microsoft days when we tried to replace Oracle database environments with Postgres SQL. That wasn’t so easy. This shows that the level of customer satisfaction with our database is high, particularly in terms of stability and innovation, especially now in the 23ai environment.
When I made my trips to get to know the customers, there were certainly some discussions, but almost nobody confronted me with a discussion about instabilities, technological shortcomings or the like. That’s why this combination is so important: now I can also use the database stack that I have learned to value as a customer in the cloud and, above all, in the cloud of my choice. In other words, real added value. Everyone welcomes that equally.
However, I currently have the feeling that many people are trying to draw up boundaries again and give preference to their own stack along the lines of: “Dear customer, if you are already in my cloud and in my stack anyway, then why not use my tools, my features and my services instead of something else external?” But regulation demands more openness from cloud providers.
Exactly. Regulation is one aspect. But the other issue of modularity also plays an important role. We have always had this in the application environment. In the on-premises world, we have heard users complain about monolithic blocks versus modular systems and best-of-breed. The issue doesn’t disappear when I’m suddenly in the cloud. Of course, every provider wants you to use as much of your own stack as possible.
But having the option of creating these possibilities via standard interfaces, standard data exchange and certain standard formats and thus supporting change should be the case.
We keep hearing from many users that we are now in the cloud and are finding that the whole thing doesn’t work so well from an economic perspective if everything is only in one place. Here, too, we are to some extent the challenger in the market with a completely different price-performance ratio, which is also due to the architecture. With us, you can consume a fully-fledged cloud, i.e. its full functionality, in just four racks. With our competitors, you need many times that amount to get a fully-fledged cloud. And the smaller solutions are always a subset of the functions.
But you can’t get very far with four racks, can you?
Of course, four racks don’t have the same compute power and storage capacity as 100, but there’s no lack of functionality. That’s why we can also offer particularly attractive conditions and place this infrastructure and platforms in market segments where people were previously asking: “Well, this is actually too expensive, and can we even afford it?”
But that is basically the crucial question. Do I focus on a technical migration to the cloud and simply move my IT from my data center to the cloud, which ultimately doesn’t bring any particular added value? Basically, you’re giving away all the opportunities for modernization and transformation.
Hardly anyone does pure lift and shift these days. Many are moving into new application development with the cloud, or at least into a certain degree of cloudification, containerization, etc. Then there are perhaps a few topics that are no longer strategic in terms of the time horizon, but which will be needed for the next two or three years. Of course, the effort involved should be kept to a minimum, so encapsulate it or keep it on-premises. Or if you really want to empty the data center, then simply move it over to the cloud and continue to operate it there.
Good, but that still has a very technical focus. I would go one step further in the direction of process and organizational modernization. You mentioned it: Oracle wants to focus on certain industries. Are you also going into a real business consultancy that you offer yourself or is this done via partners?
Well, in those areas where we have sufficient know-how ourselves—and I would, first and foremost, mention the healthcare sector with Oracle Health, formerly Cerner—we have built up a lot of capacity in the retail environment or in the hotel industry. Otherwise, we pursue the approach of strategic partnerships, i.e. addressing these topics together with the relevant consulting firms.
I would not see it as an obvious goal to build up corresponding capacities ourselves. There are excellent industry-specific specialist consultants and the large consulting firms such as PwC, Deloitte, Accenture etc., with whom we naturally talk intensively about partnerships and then also address certain industries. That is why we have also organized ourselves internally by industry in order to improve this connectivity.
Oracle cooperates with various providers of large language models (LLMs). Their own bots and AI agent technology then build on this. How do AI agents from different platforms understand each other and how do they exchange information?
PwC is building interesting platforms for this. At the end of the day, you need the right platforms, because there are LLM developers and there are also many customers who want to train their LLMs enriched with their enterprise data, and this requires powerful infrastructures.
We operate the entire stack in the AI environment. We started integrating AI into our own applications very early on, both in the industry-specific applications and in the Fusion Stack. We have probably overtaken SAP worldwide in the ERP environment. This is also due to the fact that a lot of agentic AI is already integrated into our applications, whether it’s human capital management (HCM) systems or supply chain topics.
We also offer managed platforms with various LLMs, whether it’s Cohere or Meta or others. And we also offer entire infrastructures, from very small, very modular systems that can be easily integrated into your own operations, through to superclusters that can also be used to train LLMs.
You are sitting on a veritable treasure trove of data with your database technology, which is used by customers.
Well, our customers are sitting on a treasure trove of data.
Other providers get customers’ consent and use their data to train AI features. Their argument is that this also benefits you because it makes the AI more intelligent.
We explicitly don’t train models with customer data. We are very strict about this and say that the data belongs exclusively to the customers and we help the customers, sometimes together with partners, on special projects. But always in the context [that] the customer is in control, the customer decides.
That will not change. If only because, as a database provider, we receive a great deal of trust from our customers. This is where the most critical data, which often contains the central value chain, is stored. There are large car manufacturers where not a single car would roll off the production line if these databases were to come to a standstill. This is the heart of a company. And this data belongs to the customer.
So you provide the basic technology. But if the customer says, “I would like to enrich and train this or that AI model with my business data….”
They can do this via 23ai, for example. We would also make all the cloud features available, and we would logically also work together with the customer in the project groups. We would make our engineers available, and we would contribute the expertise of our partner companies. Everything that is necessary for the customer to achieve an optimal result. But we would never use this data or metadata to train our own agents.
Here in Germany in particular, we often talk about topics such as the data economy and data spaces, some of which are also being pushed by politicians—especially in view of the new possibilities offered by AI. Industries should set up data rooms and share data in order to generate new ideas. But it’s not really getting off the ground.
Take a look at predictive maintenance models. Does machine manufacturer A share its data with machine manufacturer B? Is it really in their interest? Do you really want a generally trained model to exploit the advantage that you may have found in the algorithm based on vibration analysis or number of revolutions, whatever, and thus have longer running times and less wear. Is that what you really want?
Basically, I think it’s always interesting to improve the quality of data, but you have to look at the specific cases and see how the interests are distributed. It would be quite conceivable to create a platform for a certain subset of toolmakers or machine builders who all want to complete a specific use case. It is then conceivable to run this simulation in a controlled environment with a controlled user group and their consent.
But not necessarily to create an agent that can then be resold to all sorts of other companies. Then we quickly get into the discussion: Who is actually driving how much innovation? What about IP theft? Another area that needs to be closely monitored. Not everything that’s feasible is always desirable or permissible.
This article originally appeared on Computerwoche. It was translated into English using DeepL and edited for clarity.
More Oracle news and insights:
