In this interview from theCUBE + NYSE Wired: AI Factories - Data Centers of the Future, Amit Eyal Govrin, chief executive officer of Agentcy Labs, joins theCUBE's John Furrier to discuss why sovereign AI is widely misunderstood — and why getting the definition right is essential for enterprises and governments building AI infrastructure at scale. Govrin frames sovereign AI not as a simple question of data residency, but as a prescriptive litmus test spanning five pillars: territorial, operational, technological, legal and financial. He illustrates the gaps with concrete examples — GDPR-compliant data centers can still be exposed by U.S. federal production orders, and Telefónica's 5G infrastructure in Spain, operated by Huawei, carried jurisdictional risk well outside EU boundaries. On the technology pillar, Govrin notes that code enterprises cannot inspect or fork leaves them exposed to forced migration, citing Google's abrupt end-of-life of the Gemini Python SDK as a real-world case in point.
The conversation also explores Agentcy Labs' core thesis: that sovereign AI is fundamentally a deployment problem, not a compute or codebase problem. Govrin details how the firm is developing reference architectures to help enterprises and sovereign governments future-proof their AI stacks against vendor lock-in, pricing model shifts and geopolitical disruption. Drawing a direct parallel to the early days of cloud FinOps, he explains why token economics are following the same patterns as cloud unit economics — and why enterprises need the same TCO discipline to weigh managed API access against self-hosted, open-weight model deployments on their own infrastructure. From building modular, multi-vendor stacks to enabling on-premises inference behind enterprise firewalls, Govrin outlines a roadmap for how organizations can achieve genuine AI sovereignty in an era where compute is becoming a matter of national and commercial strategy.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. If you don’t think you received an email check your
spam folder.
Sign in to AI Factories - Data Centers of the Future.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for AI Factories - Data Centers of the Future
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for AI Factories - Data Centers of the Future.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. If you don’t think you received an email check your
spam folder.
Sign in to AI Factories - Data Centers of the Future.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to theCUBE + NYSE Wired: AI Factories - Data Centers of the Future
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Amit Eyal Govrin, Agentcy Labs
In this interview from theCUBE + NYSE Wired: AI Factories - Data Centers of the
Future, Amit Eyal Govrin, chief executive officer of Agentcy Labs, joins
theCUBE's John Furrier to discuss why sovereign AI is widely misunderstood — and
why getting the definition right is essential for enterprises and governments
building AI infrastructure at scale. Govrin frames sovereign AI not as a simple
question of data residency, but as a prescriptive litmus test spanning five
pillars: territorial, operational, technological, legal and financial. He
illustrates the gaps with concrete examples — GDPR-compliant data centers can
still be exposed by U.S. federal production orders, and Telefónica's 5G
infrastructure in Spain, operated by Huawei, carried jurisdictional risk well
outside EU boundaries. On the technology pillar, Govrin notes that code
enterprises cannot inspect or fork leaves them exposed to forced migration,
citing Google's abrupt end-of-life of the Gemini Python SDK as a real-world case
in point. The conversation also explores Agentcy Labs' core thesis: that
sovereign AI is fundamentally a deployment problem, not a compute or codebase
problem. Govrin details how the firm is developing reference architectures to
help enterprises and sovereign governments future-proof their AI stacks against
vendor lock-in, pricing model shifts and geopolitical disruption. Drawing a
direct parallel to the early days of cloud FinOps, he explains why token
economics are following the same patterns as cloud unit economics — and why
enterprises need the same TCO discipline to weigh managed API access against
self-hosted, open-weight model deployments on their own infrastructure. From
building modular, multi-vendor stacks to enabling on-premises inference behind
enterprise firewalls, Govrin outlines a roadmap for how organizations can
achieve genuine AI sovereignty in an era where compute is becoming a matter of
national and commercial strategy.
In this interview from theCUBE + NYSE Wired: AI Factories - Data Centers of the Future, Amit Eyal Govrin, chief executive officer of Agentcy Labs, joins theCUBE's John Furrier to discuss why sovereign AI is widely misunderstood — and why getting the definition right is essential for enterprises and governments building AI infrastructure at scale. Govrin frames sovereign AI not as a simple question of data residency, but as a prescriptive litmus test spanning five pillars: territorial, operational, technological, legal and financial. He illustrates the gaps wi...Read more
exploreKeep Exploring
What are the five pillars of sovereignty, and how do the territorial and operational pillars create risks for enterprises using cloud and AI services (for example, EU data residency vs. US production orders and operator jurisdictional exposure like Telefónica/Huawei)?add
How can organizations future‑proof their AI stacks and businesses against vendor lock‑in, changes in pricing or licensing models, and geopolitical shifts?add
How do you see compute becoming national infrastructure and the rise of environment-specific full-stack (rack-scale) AI solutions shaping the future of sovereign AI and labs like Agency Labs—what is your vision for this, why does it matter, and why should people care?add
What are the main challenges when deploying agentic AI at scale?add
What will sovereign AI infrastructure look like, and how should organizations evaluate the operational and financial trade‑offs (deployment patterns such as edge vs cloud, token usage, unit economics and TCO, and choices between hosted APIs vs self‑managed models)?add
>> Palo Alto, Studio Connection, Silicon Valley, and Wall Street. I'm John Furrier, host of theCUBE, here with Dave Vellante, my co-host. Hello, I'm John Furrier with theCUBE here at the New York Stock Exchange CUBE Studios, part of the NYSE Wired program and open community. This is our AI Factories series and also kind of a preview for the RAISE Summit in Paris where all the AI leaders, and more importantly, the AI infrastructure leaders are gathering to discuss the global AI infrastructure build-out. You're starting to see things like SpaceX, a whole nother generation of technology. A lot of misunderstandings around it, but the clear tech leaders understand that people are going to be vertically integrating. It's a whole nother set of horizontal scale, obviously global distributed computing. This next guest is a new member of theCUBE Research Collective, and also the CEO of Agentcy Labs. It's a research and software architecture firm, Agentcy, the T, so Agentcy. Amit Govrin, former CEO of cloud-native startups. You've been on theCUBE many times, but congratulations for one, being part of theCUBE Research and our CUBE Collective, but also being the CEO of Agentcy Labs. Very relevant AI sovereignty firm. Thanks for coming on. Appreciate it.
Amit Eyal Govrin
>> Thank you, John. I appreciate the addition to the family. It's very exciting to be part of this collective.
John Furrier
>> So explain Agentcy Labs, because I think this is a very compelling direction, and I know you got some traction. Obviously sovereignty, sovereign cloud, certainly on the international scale, people are digging in and there's not just sovereign cloud, it's sovereign AI, which is an economy-driven feature. We saw that last year at RAISE. So talk about what the motivation is and some of the work that you're doing.
Amit Eyal Govrin
>> Absolutely. I think, John, there's a misnomer a bit about what is sovereign AI, and I think it's a definitional problem, not so much a demand-side problem. I think everybody is trajectoring towards the same goals and have the same north star in mind. However, when it all comes down to it, I liken this to the definition of champagne for those wine connoisseurs out there. There's cava, there's sparkling wine out of Napa, and there's champagne. If you want to achieve champagne, you need to follow a certain prescriptive approach. You need to be able to go and to follow certain guidelines and a litmus test, if you may, in order to qualify for that. Anything short of that will keep you, to some degree, exposed and it can make the whole tree fall on itself. So one of the things that we have looked at at the Agentcy Labs, having the unique position of building agentic systems behind complex enterprise environments, firewalls, VPCs, all the edge cases, architecting systems to scale at very, very high volumes, we have seen where a lot of these different definitional problems come into play, especially when sovereign states are now looking to go and to create this their own. This is much more than just a national security conversation, because some sovereign initiatives are around that point. It's also a commercial conversation for enterprises looking to go and to gain autonomy, control over their own stack. How do you go and build that stack in a way that's fundamentally future-proofing their own business and following some of these prescriptive guidelines of what sovereign AI would look like?
John Furrier
>> So what is sovereign AI definition-wise? Words take on many meanings. It depends on what view you have into the market. It's technical. There's operational and obviously governments talk about sovereignty all the time, national here in the US, Europe, but there's a technical angle here. And what we're seeing with the neoclouds and the AI factories is they can be distributed in countries, US, Europe, Asia Pacific, everyone can have a boundary geography, but it's still connected to hyperscalers. They're still distributed computing nodes, if you will. So there's a technical cloud native kind of architecture hybrid cloud while overlaying AI. So I know that sovereignty meant data protection privacy years ago in the cloud, but in my conversation, Amit, people are talking about keeping the economy because we're seeing revenue being generated by AI. So I bring that up because that's just data points that we have. When you are out there looking at the word sovereignty, why is it broken or why is it misunderstood? Could you clarify how you see AI sovereignty as a definition?
Amit Eyal Govrin
>> Absolutely. One of the things that we could do is we can go down point by point. We're going to have a slide up here for your reference, how we view the definition in the litmus test around sovereign AI. But in a definitional standpoint, this is where it gets interesting, and we can go point by point and give you an antidote next to each. But sovereign AI goes well beyond just where does the data sit. I think this is where the biggest confusion lies. Sovereign AI asks something much bigger. For example, how is intelligence created? Where is it trained? How's it governed and deployed? And then who can reach into any of those layers? It goes beyond just, "Is my data hosted on US-Central-1 region of AWS in Frankfurt? And does a data center reside there, and is it GDPR-compliant?" Because that's only one aspect of AI sovereignty. It's much, much larger than that. This is where we can go into those definitions. So there's a slide up, I can walk you through some of these.
John Furrier
>> Yeah, let's put up the five pillars of sovereignty. So you've got territorial, operational, technological, legal, and financial. It makes sense. It's complex and we see this with telcos all the time, but with AI specifically, governments are leaning in. Geography is about data. Talk about the territorial piece, Amit, first.
Amit Eyal Govrin
>> Correct. So it's a classic residency. You have EU-Central-1 treating it as a legal firewall, if you may. But is it the geographic presence in and of itself an issue if you're a European enterprise and there's a US federal production order that lands on AWS desk asking for an audit trail of all of that data? So at that point, even though you initially signed up for a sovereign AI or a sovereign data residency within the European Union, now you have a US federal production order that allows them to go and to encroach and do that. And that's not exactly what you signed up for as a European enterprise looking to protect your data. So that's one exposing issue that lives there. Where does the data live? But also where else are you potentially exposed if that's the only definition you're relying on?
John Furrier
>> Just about the operational piece now, because cloud and AI converging, take us through the operational point, pillar two.
Amit Eyal Govrin
>> Yes. So let's take an antidote around Telefónica, who's a Spanish national telco carrier. They had hired Huawei to deploy and manage our 5G infrastructure in Spain. So while the Spanish soil was where the servers were hosted, the company who was operating this had other jurisdictional policy out of Beijing, namely China. So while you still had the sovereignty of the data residing within Spain, you had other risk exposures that lay outside of the jurisdiction of the European Union.
John Furrier
>> So on the operational side, we got that one. Now the technology, this is where it gets really interesting because now you've got the cloud native and the work you've done in the past is significant. So unpack the technological angle around the intellectual property, the stacks, the AI factories are booming. We think the intelligence injected into these sovereign areas, sovereign networks, call it whatever you want, sovereign countries, sovereign networks, sovereign data, unpack the tech piece.
Amit Eyal Govrin
>> This question, John, in my opinion, goes down to the core of the open source initiative around you don't own what you can't fork and you can't see. And this is really the bigger picture about the technology stack. If you don't have control over the code and have the ability to inspect it, then you're exposing yourself to vendor lock-in and to other type of behaviors that fall outside of your control and lack of control will cause for lack of sovereign initiatives. So this goes into stories we can talk about. For example, companies who standardized off of Google Gemini's coding assistant found themselves having a forced migration because Google end of life their Gemini Python SDK back in November '25. Now that wasn't planned by the customers. This was essentially a company decision by a hyperscaler to say, "We're now going into a different direction. Anybody standardizing off of our previous technology stack needs to go and to own that part as well." So you could see where if you don't have full control over the stack and have the ability to at least control this with a fork or control it with your own kind of... If you don't have full control over the stack, then you clearly are exposing yourself to additional risks.
John Furrier
>> Take us through the vision for Agentcy Labs around putting this all together. We're going to do a lot of content. We're going to also do some research. You have a plan with the labs. Explain what's going on with the labs, because the technology's lining up. You got the legal, you got the ops, you got the sovereignty, territorial. It's very clear that we're going to have boundaries and they're going to span a full stack of issues. What is your vision? What are you guys working on? What are you scratching out? You don't have to go into detail. I know you got a lot of confidential things going on, but what is the core thesis and what are you going to be unpacking?
Amit Eyal Govrin
>> So John, one of the core thesis that we have at Agentcy Labs really building in this space for a few years now and seeing some of the pitfalls, some of the rough edges around the corner that a lot of enterprises and even sovereign governments are going to be encountering as they build in this domain. We have put together reference architecture that will go and serve hopefully as a north star for people building sovereign AI stacks in this domain to go into reference, to go in to build around this so that they can go ahead and avoid some of those same pitfalls. We're working on sovereign AI initiatives for some pretty high-profile initiatives that I can't obviously discuss. How are they going to go and future-proof their business against vendor lock-in if there's a AI lab that goes and changes their pricing model on them? How are they going to go and future-proof themselves if somebody goes and changes their licensing model on them? How are they going to go and build as they go and see some of these geopolitical shifts come across? So how do you go and future-proof your business so that you're not single... focally focused on a certain vendor, a certain pricing model or a certain technology stack? You could go have a modular approach and always go and build with the next best model with the next best standards.
John Furrier
>> One of the things we're seeing, I want to get your thoughts on this because the compute is becoming a national infrastructure in the US, you're starting to see a full stack of solutions. And even if you look at the AI factory market, NVIDIA, Dell Technologies, they're winning with rack scale systems. CoreWeave has the Vera Rubin workloads now pumping out in production, cooling. So you have the infrastructure, but the stacks are different by environment. I think this plays well into Agentcy Labs because there's no one general purpose architecture. Yeah, you can have a stack, but it's going to be dependent upon what's in the environment. Now again, it's distributed computing, global distributed computing and hybrid cloud, but each node or geography or whatever you want to call it has to be different. And Dell learned that. And since they learned that, their sales went through the roof, and I think that's a template and a predictor for what sovereign AI will look like on the big picture. Your lab is going to take a stab at that. I like that. What's your mindset around this? Obviously still early, there's a lot of unknowns, but like NVIDIA, five years, they didn't know they were going to be this far along. SpaceX, that's the same market we're seeing with them. People see value in what yet isn't invented. So I think this direction of sovereign AI from an infrastructure of the five pillars is a nice frame. Fill in that vision. What do you think? Why is this important? Why should people pay attention to this?
Amit Eyal Govrin
>> I think, John, you hit the nail on the head, which is this is not a codebase problem. It's not even a GPU problem or a compute problem. It's a deployment problem. When you start looking into some of the inner workings of deploying agentic AI at scale, now you're looking at areas that go far, far beyond how is this going to be running on a local computer, on a local host? As an example, you can now go ahead and look at how's this going to behave behind your own network firewalls? Does this scale infinitely? Can I go ahead and use on-prem inference engine? Can I use open-weight models? Can I go ahead and use my own stack, add my own rules under my own control? And that's where some of these edge cases really start speaking louder than just the codebase itself.
John Furrier
>> Well, I'm really looking forward to working with you because I published a hyperconverged edge piece at Mobile World Congress in February. We've been continuing to cover in the AI factories. We'll be covering the CoreWeave launch on June 30th. That's going to be big. You're going to start to see that Vera Rubin workloads come in. You're already hearing about tokenomics. The FinOps Foundation is being renamed Tokenomics. They're going to have two different foundations. You're kind of seeing that cloud wave happen in the AI wave again. It's different, but it's connected. So you have a whole nother connective tissue in this. And I think the work you're doing, we're going to publish a joint paper. You got the five pillars. You guys got a lot of lab work going on over there with customers doing reference implementations, scoping out sovereign AI infrastructure. And again, this is a perfect storm because we've seen companies that do this in other areas become embedded into the neoclouds, become embedded into the stack and there's many places to play. So I'm sure there'll be a lot of discussion we'll be having. Thoughts on what they will look like? Shoot the arrow forward and give us a feel for what you're thinking. I'm clear on what I see. It's going to be edge. You're going to have really robust, cohesive, these network environments, their nodes, whether it's a country, region, local, all happening at the same time. What do you see?
Amit Eyal Govrin
>> Actually, I think you hit on a very interesting point. It's around the FinOps. So I actually got my software career started in the world of FinOps. Prior to it even being considered FinOps, it was called Cloud Analytics back in 2013. However, I'm seeing a lot of these same themes repeat themselves in the early days of cloud. In the past you had to educate the early enterprise users how to go and think about monolith applications different than distributed applications in the cloud. You can't just lift and shift them and hope for the best because the unit economics won't work out. You're seeing the same type of patterns and behaviors work themselves into token maxing, which is pretty bad behavior if we want to go and be honest about it because it has to be-
John Furrier
>> Yeah, it's bad behavior. It's basically the wrong metric because it's like saying the more lines of code I write, the better the code. Old school coding, when I was growing up, it was tight code, short code. Because remember the memory constraints back in the '80s and '90s, Amit, when it was really early computer science, computer industry, that was the key. We see constraints at larger scales. And I think the complexity, this is why I like this direction, is it has huge impact because the scale and the supercomputing. So again, you're going to have to have tighter code. Now tokens, how you use the tokens against the outcomes, I think this is where your lab work and some of the narratives we're covering with the NYSE Wired community and theCUBE, because deep tech meets outcome. So that's money, that's societal benefits, et cetera. So really, really great stuff.
Amit Eyal Govrin
>> Absolutely, John. And this is where we can really draw the FinOps and the financial perspective of sovereign AI into it if now you have a way to measure not just outcome and ROI, but also pure unit economics. Is this better for me to stay on this AI lab going and hitting their APIs at a certain rate, knowing that this is my budget for the year? Or does it make sense to maybe use open-weight models to prune it down and to have a GPU cluster on Hetzner out of Germany rather than using one of these cloud service providers with their own local inference? And having that apples to oranges type of discussion, "Does it make sense? Does it make sense for me to deploy this way? Am I able to go in to get the best possible outcome? And where's that breakeven point?" And I think this is the TCO calculations of the cloud back in 2012 or 2014.
John Furrier
>> Well, you can't have a TCO calculation unless you get the cost now. And this is why I like the FinOps crossover. I think this is a really great example, Amit, your new opportunity Agentcy Labs that falls in line with the collaboration. We have theCUBE with Brian Baumann with the NYSE and the NYSE Wired collective that's formed around that is a win-win. It's a co-designed model. This is extreme co-design in practice. NVIDIA, and many others in the early days of cloud, cracked the code on this. So we're looking forward to co-producing, co-developing content data with you and the team. Really thank you for the time explaining all this, and we'll see more content flowing from you, me, together. Thanks for your time.
Amit Eyal Govrin
>> Looking forward to it, John. Thank you for your time as well. And looking forward to having future guests on to help narrow down the scope of this, and really to dive deep into some of these conversations. Who are the innovators in this domain? Who are the ones who hit the litmus test to know how to go and to innovate and to bring this whole industry forward? I'm very excited for this one.
John Furrier
>> Yeah. We're going to interview the leaders. We're going to get their data, what they're working on. Again, like we did with AI Factories, this is a whole nother level. It's a really big dimension. Agentcy Labs, Agent C-Y, Agentcy. Agentcy are going to be big and then they're going to be running, creating value. They're going to be digital workers, they're going to be in countries. That's going to create an AI economy. And this is why it moves so fast. Of course, we'll do our part to keep it pumping to you and all the content and the data. I'm John Furrier, host of theCUBE here at theCUBE's NYSE Wired Studio, New York City. Thanks for watching.
John Furrier