Rise of AI Conference 2026 · Berlin · Applied AI Stage

Beyond Cloud or On-Prem: Rethinking Enterprise AI Sovereignty

At Rise of AI 2026 in Berlin, Daniel Brilz of SVA argues that the usual cloud-versus-on-prem framing of AI sovereignty is wrong, because it forces a binary choice that produces suboptimal architecture. The real question, he says, is which dependencies are acceptable for which workload. He names four forces making sovereignty strategically relevant: regulation moving from theory to enforcement, geopolitics changing how companies view critical workloads, lock-in risk from AI platform choices, and Gen AI becoming core business infrastructure rather than an experiment. Sovereignty, in his definition, means practical control across three layers: technical, operational and contractual, covering access, encryption keys, governance and exit strategy. He presents a spectrum running from fully managed cloud/SaaS to air-gapped systems, argues dependency exists across the whole stack rather than just hosting location, and outlines a two-step decision framework: classify the workload's sensitivity and criticality, then weight architectural options on portability, assurance, resilience and economics. He closes with design patterns and a cost warning that sovereignty is never free.

Key takeaways

  • Framing AI sovereignty as a binary cloud-or-on-prem choice produces suboptimal architecture; the real question is which dependencies are acceptable per workload.
  • Sovereignty rests on three control layers: technical, operational and contractual, covering access, encryption keys, governance and exit strategy.
  • Dependencies exist across the whole stack, not just hosting location; even EU-hosted platforms create their own dependencies.
  • A two-step framework classifies workload sensitivity and criticality first, then weights architecture options on portability, assurance, resilience and economics.
  • Hybrid architecture combining commodity, controlled and sovereign AI clusters is not indecision but the actual right choice for most organizations.

Detailed notes

Reframing cloud versus on-prem as a spectrum, not a binary

Brilz opens by saying the usual question organizations ask, cloud or on-prem, is framed wrong. A binary choice forces suboptimal architecture and suboptimal economics. The real question is which dependencies are acceptable for which workloads. This opens a spectrum of answers rather than a single either/or decision.

Four forces making sovereignty strategically relevant

He lists four drivers: regulation moving from theory into operational enforcement; geopolitics changing how companies think about critical workloads; the lock-in risk of AI platform choices, which exposes organizations to pricing volatility and policy changes with no exit strategy; and generative AI becoming core infrastructure and core business rather than an experiment.

Defining sovereignty as control across three layers

Sovereignty, Brilz says, does not mean nationalism or purity, it means practical control. He breaks this into a technical control layer, an operational control layer and a contractual control layer, covering access, encryption keys, governance, observability and exit strategy. He distinguishes sovereignty (more control) from independence (less lock-in) and resilience (better survivability). The spectrum runs from cloud/SaaS or API (fully managed, maximum convenience) to air-gapped systems (maximum control, maximum operational burden); most organizations need the right level of sovereignty, not the maximum.

Dependencies live across the whole stack

Brilz argues that dependency is not just about hosting location: even hosting in Europe leaves a company dependent on the platform it chooses. He maps dependencies across a business layer (platforms, data), a tech layer (identity, access, encryption keys, model runtime and hosting) and an enablement layer (tool chain and operating skills). He notes that operating on-prem GPU infrastructure requires a specialized, scarce skill pool.

A two-step decision framework and three architecture clusters

The framework has two steps: first classify the workload by data sensitivity, business criticality, regulatory exposure and threat model, and define non-negotiable hard constraints; second, weight architectural options on portability, assurance, resilience and economics to choose a position on the spectrum. This produces three clusters: commodity AI for low-risk production use, controlled AI for internal or enterprise copilots, and sovereign AI for the most critical workloads. Choosing more than one is not indecision but the actual architecture, since most organizations need more than one cluster. On the US-cloud-versus-EU-cloud debate, he says an EU cloud does not automatically mean sovereignty, only a different dependency model; the questions that matter are who can access data and under what conditions, who controls the encryption keys, whether access can be denied technically rather than only contractually, and how fast an organization can exit a platform without rebuilding its stack. He lists design patterns for sovereignty by design: retrieval-augmented generation before fine-tuning so data is not baked into a model, policy-based routing of workloads by criticality, a model abstraction layer that decouples applications from specific models (citing tools like LiteLLM), customer-managed encryption keys, egress control for autonomous AI agents, and a fallback model strategy prepared before it is needed. As a practical example he cites IBM watsonx Orchestrate as a control plane that can deploy on-prem, in the cloud or hybrid, with many connectors to different model providers and databases, and strong built-in governance, security and observability; he places it between the controlled AI and sovereign AI clusters. He closes the main talk stressing sovereignty is not free: it carries operational burden, GPU capacity planning and a talent shortage, while convenience carries its own costs in lock-in and pricing volatility. His three takeaways: sovereignty is a spectrum, not a binary choice, and each workload should be designed for the level it actually needs; start by gaining control of encryption keys, data and access; and hybrid architecture is itself the right choice, not indecision.

Q&A highlights

Asked how to justify time spent on fallback options to management when a fallback may never be used, Brilz says it should not be a separate process requiring its own budget discussion but built into the whole design process from the start. Asked how to identify an organization's non-negotiables, he says organizations usually already know what they do not want, such as not processing certain data in the cloud, and the task is asking the right questions and listening carefully; existing constraints should still be challenged, but companies usually have a reason for them. Asked about skepticism toward hyperscalers' own sovereign cloud offerings, such as AWS's sovereign cloud, he says such offerings still create dependencies, just different ones than EU sovereign solutions, and whether that is acceptable depends on the criticality of the specific workload rather than a general yes-or-no answer. Asked whether people frame sovereignty purely as a moral question, he says he mostly encounters technical and business framings rather than moral ones. Asked whether his framework is just business continuity management under a new name, he says it was not intentionally modeled on BCM, but classifying workloads by these criteria in practice helps organizations identify which workloads need different compute or dependencies.

Quotes

Sovereignty is not a binary choice, but it's a spectrum.

Daniel Brilz

Hybrid is not indecision, hybrid is actually the architecture.

Daniel Brilz

Questions answered in this talk

Is AI sovereignty just about choosing cloud or on-premise hosting?

According to Daniel Brilz at Rise of AI 2026, no: that binary framing is wrong and produces suboptimal architecture. The real question is which dependencies are acceptable for which workload, which spans a spectrum from fully managed cloud to air-gapped systems.

Does hosting AI infrastructure in the EU automatically make a company sovereign?

According to Daniel Brilz at Rise of AI 2026, no. Hosting in Europe still leaves a company dependent on the chosen platform; sovereignty depends on actual control over access, encryption keys and exit options, not on the hosting location label.

How should a company decide how much AI sovereignty it needs?

According to Daniel Brilz at Rise of AI 2026, use a two-step framework: classify each workload by data sensitivity, business criticality, regulatory exposure and threat model to define non-negotiable constraints, then weight architectural options on portability, assurance, resilience and economics to pick a position on the sovereignty spectrum.

Mentioned

AI sovereignty, cloud vs on-prem, retrieval-augmented generation, policy-based routing, model abstraction layer, customer-managed keys, egress control, LiteLLM, commodity AI, controlled AI, sovereign AI

Keep exploring

More from Rise of AI Conference 2026

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 3,750 words

Show the full transcript

Yeah, hello everyone. Um, welcome to you to my talk. Uh, rethinking the enterprise. I don't I think I need the clicker for Uh, welcome you to the uh to the talk uh rethinking the enterprise um, spectrum of AI sovereignty. Thank you. And um, yeah, I'm uh a I um solution architect and AI engineer at uh SVA, which is so short for um System Vertrieb Alexander. And um, yeah, we are we will be talking about um what is uh between the two extremes going to cloud or going on prem um to be sovereign. And uh yeah, usually the um uh the debate of uh getting sovereign um is uh usually framed wrong. So, um this is uh the question that organizations um usually ask is should we go cloud or should we go on prem?

Um, but there uh is a lot a lot between. Um, having such a binary um choice or just getting uh um a binary um choices means uh that you get uh sub optimal architecture and uh sub optimal um economic choices. Um, actually the the real question that you should ask um yourself is um which dependencies are acceptable for which workloads? And that opens a lot more um answers to these questions rather than um just cloud or on prem. And um yeah, this is this will be the only meme in this presentation. So, uh just I that. But, uh why is that? Um, why do you need to be sovereign? Um, There are four forces that actually um yeah, push pushing the or making sovereignty strategically relevant.

This is regulation. Uh first, so that means um regulations going from theory to being implemented and being operational. So, that's why you need to be sovereign. And also, geopolitics change how a companies think about their critical workloads. And um yeah, the lock-in question is um for your AI platform choices. Um that means you get yeah, locked in in pricing volatility and policy exposure. That means that can it can change anytime. And um if you don't have any exit strategy, you are stuck with it. And the fourth fourth is actually um Gen AI and then AI in general is not an experiment anymore. Um it became core infrastructure and core business. So, that's why you have to care about it.

Um and what do we mean by sovereignty? Um sovereignty means um or it doesn't mean nationalism and it doesn't mean purity. Um sovereignty means um you have to get control. You have to get practical control um about um your stack. And that means um you have three layers. This is a technical control layer um and the operational control layer and the contractual control layer. Mm. Your degree of sovereignty um means how much control you have in these layers. So, that is the control about access and this is control about your keys, so encryption and it's control about your operations. And also control about your evidence, so means governance, observability and stuff like that. And also you need control about your exit strategy.

That means if anything changes, just in case, you need to be sure and have a plan how to exit um yeah, your your stack or without rebuilding it. >> [snorts] >> And to make a differentiation serenity means more control in contrast to independence independency which is less lock-in and to resilience which is better survivability. So, you get uh um just in case what happens um you get uh um you're not stuck with it. And yeah, thinking about serenity um in this not in a binary choice um this yeah, a spectrum appears. So, um the spectrum is um from the two extremes which is cloud which means SaaS or API which is fully managed services which gives you speed and inconve- and convenience um to the to the right side which is the other extreme this is air gap so me you're get maximum control but um also get maximum maximal operational burden.

And usually organizations and companies don't need maximum serenity but they need the right level of serenity. And this means thinking in this spectrum. So, um the architectural task is um not to pick one end of these, but uh choose the right place of um uh the right place on the spectrum. Mhm. But, where does um dependency live? So, because um every um choice in the spectrum means different dependencies, and uh this is basically the uh what serenity uh what do you do you need? Uh you need to cut your dependencies to uh be sovereign. And uh where does this live? Um across Spoiler, it does live across the whole stack. So, um usually companies or organizations talk about hosting locations if they talk about uh sovereignty, but uh the truth is uh your dependencies are all over the stack.

Um even if you host in in Europe, you're still dependent on um the the platform uh in Europe that you choose, but uh it's about choosing the the right dependencies. So, you have your um business layer, which is the the orange um colored um planes. So, um you have dependencies in your in your platforms or in your data. And um you have the tech layer, you have dependencies in identity, in access, uh in your keys, so uh encryption again. Um you have dependencies in your models, um runtime hosting, which is still an important, um but it's not the only dependency. Um and you also got um dependencies in your enablement layer. That means uh in your tool chain, and also um uh very important is your operation uh operating capabilities, which are skills.

Um if you want to um uh have a have a rack uh with a GPU on prem, um you need a special skill set to operate this. And the skill pool of people that can do this is very small. So, if you think about if every company wants someone to operate this, this is a very small pool of skilled people. So, but how do you um choose which level of sovereignty do you want to have? Um this is a um you can approach this with a two-step sovereignty decision framework. And in the first place, you have to classify your workload. You get to know your protection needs. So, you have have to understand your data sensitivity, your business criticality, your regulatory exposure, and the threat model.

So, which dependencies you want to secure your business for from and which actors. So, and the most important thing is define your non-negotiable controls, because this will be your hard constraints for your architecture later on. And um if you classify your workload, um you can evaluate your architectural options. This is basically the step one defines weighting for these Yeah, for for these control portability, assurance, and resilience, and economics. So, having a weighting allows you to define the lens of architectural choices you can have. And with this, so you scope basically your architectural choices. And if you do this, you can choose the right architecture choice, which is on the spectrum. And with this, the outcome will be like uh something clustered like commodity AI or controlled AI or sovereign AI depending on your workload.

So, a specific specific tailored um decision or architectural choice. Um and in this case, um the uh US cloud versus EU cloud uh debate um becomes useful again if it's framed correctly. It should not be a uh political discussion, but the important message is even just if you go to an EU cloud, you're not automatically sovereign. Um it's just a different dependency model. And um not the label is is uh important, but the control that you have. And um they they are different for uh EU clouds, but uh um you still have to choose which one or you have to um consider which impacts um does it have if you choose the um EU cloud.

So, which question you should ask um if you want to choose between an US hyperscaler or the um EU cloud model, um you should ask who can access what. So, your data and uh under which conditions. And um how is that evidenced? So, uh in case of govern for for governance, you should ask that. And also we are again at the keys. So, who controls the keys? Um and this should be you. Um And can access be denied technically, not just contractually? So, um if you um think about US clouds, they um say or they they uh contract with a contract, they uh say nobody can access your data, um but um in real, you know, some uh FBI or NSA in in case they can do it.

And also it's important how fast you can exit such a platform without rebuilding your entire stack. And um yeah, thinking about sovereignty independencies and thinking about how you can do this, you can design sovereignty intentionally. And you can do this by design patterns. I'm losing my microphone, so Um so this are example design patterns um to implement sovereignty by design. So you go for wreck before fine-tuning, that means you don't uh yeah, bake in your data in a model, but you give it as a context. So that means you keep control of your data rather than having your data um yeah, baked in in a model. Um and you do policy-based routing, that means you route the workloads depending on criticality or sensitivity.

So less critical workloads can go to uh for example, in a cloud model and more critical workloads you com- compute in in an on-prem model, for example. Also the li- design your uh system with model abstraction layer, that means decoupling the um application from the model. So if anything changes, if you if you use a model as a service and anything changes, um you should be able to switch your model no matter what and the application application which uses the model shouldn't know about it, so and shouldn't notice it. So, um there are tools that you can do uh these like light LLM, for example, um that allow you to to uh policy-based routing or um just like an abstraction layer.

And again, uh customer-managed keys or you you host your own keys uh is very important to gain access uh um or to make sure only you have access um about your data. And um egress control um is important in uh the times of autonomic AI agents, so um you define what is allowed to leave the system, so that not an uh AI agent just can um expose uh anything of your sensitive data, for example. And then uh the fallback model strategy is uh you should think about alternatives um in uh operations and models um before you actually need them, so make sure you can switch um in case you need to. And uh yeah, to make this a bit more practical, um let's have a look at three um architectural blueprints.

And uh these blueprints um are commodity AI, controlled AI, and sovereign AI, that's basically three clusters that build um that you can build around the spectrum. And uh not every organization needs um every architecture, so um but uh most organizations need more than one, so if you want to uh tailor your uh uh your architecture to the to your uh workloads. And um commodity AI is usually for, yeah, low-risk production uses, um low-risk work, and uh controlled AI is more like for internal copilots, enterprise copilots, and uh sovereign AI is for the very critical workloads. Um and choosing more than one, so being hybrid is not indecision, but hybrid is actually the architecture, the architectural choice.

And um let's have a look at the practical implementation of this. So, for tool that can help you with that is, for example, IBM Watson X Orchestrate, which acts basically as a as a control plane. And the control plane allows you to be sovereign. Um what makes Watson X Orchestrate special in this case is that it allows you to fit different sovereignty targets. So, you can deploy on prem, or you can deploy in the cloud, or you can deploy hybrid. So, you can choose whatever you need for your workloads. Um also in this ecosystem you get a lot of a lot of flexibility. Um with many connectors that are available. So, you can connect to different clouds which serve models, or you can connect to your local local models, for example, or local connectors for your databases, and or cloud databases.

And this allows you to be flexible and cut your dependencies. So, if any if anything happens, you can just switch your data to your another platform or another database, and just switch the connector in this tool. Um also governance, security, observability is uh very strong in this platform. So, it is designed from from yeah, from scratch building this this tool. And this allows you to yeah, gain control over your dependencies. And uh if you think about the uh three clusters, um what's in X Orchestrate um yeah, locates around controlled AI and sovereign AI um between these uh depending on uh what you actually um want to do with it. But, uh what are the costs?

Um Actually, there is no no uh free lunch. Everything has a cost. Sovereignty is not free. Um every control has a cost, uh but also every dependency also has a risk. So, um convenience is not free, either. So, in for sovereignty, you get the flat ops bur- burden, uh GPU capacity planning, and uh the uh very important talent secretary. Um But, also for convenience, you get the uh API lock-in pricing volatility. And, uh you get basically the dependency um yeah, uh disadvantages. And, um the goal is not to gain maximum sovereignty, but uh uh you gain the goal is to gain the uh right level of sovereignty. And, um how we do this um consistently, or so how we design this consistently between our workloads.

This is um uh you can follow the decision workflow. This is basically putting in ev- everything together uh in a simple version. Um as I already said, classify your workload um by knowing the sensitivity and the criticality of your workload um and the regulatory exposure, then define your non- neg- negotiables, so your hard constraints. Um and then you're able to choose the right spectrum, uh the right target position on the spectrum. Um then from day one you design for portability and you prepare your exit and your fallback before you need them. And uh following this sovereignty becomes governable. And uh I want to close the talk with uh my three takeaways. Um which is uh yeah, sovereignty is not a binary choice, but it's uh a spectrum um as an important message and you should define design for the level you actually need for every workload.

So, not uh just a um choice for every workload that you have, but um the right choice for every individual workload. Um if you want to gain control, um you should start getting control over your keys first, of your data first, and your access first. So, that means um um getting more control. And third, hybrid is not uh indecision, hybrid is actually the uh architecture. So, you don't have to eliminate every dependency, but uh reduce the critical dependencies um without destroying your speed and your valua- value um designing your uh workloads or your systems. And yeah, that's this is the the the takeaways for um AI sovereignty. Thank you. >> >> Thank you, Daniel. And this is his contact data, but maybe you go back to the takeaways so people can take a picture.

If you take away something, do take away this. We have some time for questions, actually. Are there There is one in the back. And then of course we can go to your contact details so people can say hi. I was just thinking. So it's nice to have like a photo slide. All right, a question. >> Um can you explain how you take time out for fallback options and like say to your higher management because eventually if you're lucky you never actually have to use a fallback. So the first thing is that some kind of manager will ask like why should I allocate time and resources for fallback option that we might not need. So how do you like how do you make the decisions?

How do you sell it inside your company? >> I mean um I wouldn't just uh make it a separate process but include it in the whole process so that it's just part of it and you don't have the discussions about getting allocating time and for designing the solution but No no no include include it. Yes yes it's it should be part of of your whole process. >> There's a question here while the microphone travels I have one. How do you give people advice how they find their non-negotiables? Like how what because you had a nice workflow on the whole thing but the non-negotiables are a key element of that, right? So how would you pick them?

>> Yeah actually it depends the most critical um things that you may not want to change or like usually the the companies have a already a process in place and they know what they don't want to do like they don't want to process this data in cloud for example and yeah this is basically the company or the organizations know best about what they don't want, but it's getting the the right questions and listening carefully to actually find these non-negotiable constraints. >> But you think like what has always been in place should stay in place or should that be challenged at that moment? >> Of course challenge it, but uh I mean usually you can't change this because usually company has a reason for not doing this or not doing that.

>> Got it. I think the microphone has arrived. >> Yes, thank you. Hi, I'm Mark from CIOnet. So, within the CIO community there's some skepticism around the the sovereign solutions of the hyperscalers. So, AWS launched a few months back their their sovereign cloud. I'd just be interested in your your opinion on that. How sovereign are their sovereign solutions so to speak? >> Yeah, that's actually a good question. Um Yeah, it's a mean question. >> Yeah, yeah, yeah. >> There was already a little bit of subtext in your presentation, I feel. >> Yeah, it's even if you go if you choose these sovereign solutions, you still get dependencies, but these dependencies also appear or different dependencies appear if you choose EU sovereign solutions, but the keys to make the right to to get the right dependencies for your workload.

So, if this is fine for for example, a non-critical workload, you can go with that. It's It's not a um that you can say it's good or bad in general. So, it's not a a answer, but um you have to um you have view it in in context of the workload you need it for. So, there's no no yes or no like the binary um choice, but it's it's a spectrum and the spectrum is uh um for your individual workload. >> Right, we have time for even one more question. Somebody has one. I have one. Ah, you you go over here. I have one otherwise. >> I was surprised by your slide title that said that it's not only a moral question.

Like do you encounter many people who only see it like a moral point? What do you tell these people? >> Um I don't know if you if I had some had a moral >> Yeah, it said it's not just a moral question. It's about dependencies. That was the title of your slide. So, I was wondering do you encounter many people that see sovereignty only from like this moral standpoint of we should be sovereign and have control or >> Um no, I I don't. Um but uh it's um you can you can uh have a debate on the moral moral point, but um or the moral aspects of it, but uh usually it's uh more technical or more business oriented, yeah.

>> Okay. >> I am Gunnar is my name. I'm the CISO of an energy company called G Tech. Uh I have a question on few of your details. Uh data sensitivity, uh workload criticality, fallback options. To me that sounds like BCM 2.0. Is it taken out with more success for the stakeholder executive and share shareholder reporting or is it just coincident? So, I'm just asking is that covered with BCM? Uh is there more success in or is it just a technical coincident? >> Um I don't know if I if I get the the question right. Um Can can you rephrase it? >> Yeah, you have taken elements or components of a BCM business continuity management where it's about criticality of your assets and have fallbacks plans.

Usually there is a conversation on what is the impact, what is the return on investment, why do we have to pay for that? A typical executive conversation. Here that sounds that there is more success in doing it specifically in the sovereignty conversation. Is it coincident? Have you balanced it with BCM or is it just it? >> It's not intentionally based on BCM. So it may be coincident, but it's yeah, from from uh the experience that this makes it if if you consider these points, you get more sovereignty. So making classifying your workload um along these points helps you to um define which workload is critical and which workload needs different um execution compute for example or different dependencies than other.

>> All right, any more questions for Daniel? One more maybe? Okay, I'm sure you'll be around. We have your contact details. >> Yeah. >> Thank you so much. Daniel. >> Thank you. >>

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.