RAISE Summit '26 | Philip Rathle, Neo4j & Amit Govrin, Agentcy Labs

Clips
More from RAISE Summit '26

Philip Rathle

CTO

Neo4j

Amit Govrin

Co-founder and CEO

Kubiya.ai

play_circle_outline AI-Native Software: Agentic Infrastructure and Context as the Durable Moat

play_circle_outline Centralized data lakes don't equal centralized knowledge or connected signal

play_circle_outline From LLM Hype to Semantic Rigor: Ontologies, Knowledge Graphs and Neo4j for Explainable AI

play_circle_outline Economic Incentives Driving AI Sovereignty and Hybrid Deployment Choices for Data Residency, Control, and Critical Assets

play_circle_outline Deterministic reasoning, neurosymbolic approaches, multi-hop graph reasoning for high-stakes decisions

Info

Philip Rathle, Neo4j & Amit Govrin, Agentcy Labs

Philip Rathle

CTO Neo4j

Amit Govrin

Co-founder and CEO Kubiya.ai

search

>> connection Silicon Valley and Wall Street.

>> I'm John Furrier, hosting theCUBE here with Gabor Latvay, my co-host.

>> Welcome to The RAISE Summit coverage brought to you by Solidigm with support from Tensordyne, d-Matrix, and Argentum AI.

>> Welcome back to theCUBE's livestream here in Paris, France. I'm John Furrier, your host of theCUBE. We're here for the RAISE Summit. Day 2, wall-to-wall coverage of theCUBE, live streaming again from Paris. Of course, the top story is AI infrastructure that's enabling massive change at the agentic layer, and of course, AI-native software layer emerging. This session also is talking about sovereignty, because we're in Europe, they care about sovereignty, but it's everywhere. And that's the topic of this one, is that Philip Rathle's here, CTO of Neo4j, the leading graph database. Phil, great to see you again.

>> Pleasure to be here, John.

>> As always, going to be great. And Amit Govrin, who's the CEO of Agentcy Labs, Agent with a T, CY. New venture. Congratulations on your startup, which is based in Palo Alto.

>> Appreciate it.

>> We'll unpack the hell out of sovereignty.

>> Sovereignty is a big topic, as we all know.

>> Philip, you're going to be on the main stage giving a talk on not just graph database, but in general, the stack of AI. And we haven't seen this movie before. We've seen comparables in cloud native, digital online, a lot of graph databases, but now graph databases and all the underlying context is going to be the big topic.

>> What's the main talk track?

>> Context is a huge topic here. And so if you look at the AI stack and what part is being commoditized, the applications and code are clearly getting more and more commoditized. And then at the lower end like everyone's got access to the same models, same GPUs. Okay, there's some scarcity issues with that, but that's most likely temporary. So then what ends up being durable is your knowledge, your context, your meaning, your ability to connect up all the different intelligence and data that's sitting inside all those different silos of your business and then making that available to agents. So the panel I'm on a bit later today is on the, just talking about the AI stack. And obviously I'm representing what I believe and many of our customers believe is the the most durable moat, which is context.

>> On the data side, this is a huge topic and it comes up all the time.

>> You got to harness— got to get those harnesses in place. I just saw a blog post from Databricks and the cost is really around harness selection.

>> Which points to the context. And talk about what your customers are doing because now you're starting to see the data piece come into it.

>> You guys are exposed to a lot of the top enterprises who are using graphs. They understand their data.

>> What is the current state of the data relationship to the stack and what's that look like?

>> So what a lot of customers are finding and the way they're bumping into graphs is you can have even if you have all your data centralized into a Databricks or a Snowflake or the like, that doesn't necessarily mean that your knowledge is centralized or you know how to connect things. A trade-off of these technologies where you can just dump effectively everything and consolidate in one place is that you actually give up the ability to connect the signal amongst the noise and differentiate the two. So what the harnesses that we're seeing that are being successful are the ones that have access to a system of context, knowledge, meaning. Words like semantics and ontology, which were very kind of niche graph terms for many years, have come back into vogue. And it's because models can be trained on your data. And then if you unleash them on all the data in the entire company, pull back massive amounts of context. A context window, I guess this word means many things, but if you send millions and millions of tokens, that's not only expensive, but you're praying that your model is going to be able to sort out all the noise on the fly and do it in a short period of time or cheaply get the right answer, that doesn't work, nor is it going to work anytime soon. So synthesizing your data into a system of knowledge and then having this left brain, right brain behavior where the LLMs can go into a system where your knowledge is organized.

>> it's interesting, last year when we were here at RAISE, you and I on camera and outside when we were having coffee at the cafes, we were talking a lot about graphs. We love graphs.

>> Everyone knows I love graphs, so it's like my bias is already out there. But a lot's changed this year.

>> What's been the big change for graphs, and why is graphs the perfect scenario for AI?

>> Because I think everyone understands context and that that's important, but what, in what way does context show up? What you really need is information about the actor and the things being acted upon and the system so you can understand multiple levels out what that whole system looks like, whether it's connecting your supply chain data back to inventory, back to customer, or whether it's connecting inside of a payment graph to understand multiple intermediaries so you can have AI that will catch fraud. It comes down to pulling your— giving your AI system quick access to what's the data that really matters, the signal, how that signal is connected to basically preprocess it. What doesn't work, and there have been some meta studies on this looking at thousands of research papers on what makes AI more or less accurate is you can't just throw infinite amounts of text into a context window and hope to get a better answer. So it turns out knowledge graphs solve for hallucinations. They solve for the black box problem because they can give you explainability and they importantly solve for governance insofar as you can have a knowledge layer that filters out what's appropriate according to security, privacy rules and corporate firewalls.

>> It's interesting, too the success of some of these terms like ontology is in vogue.

>> And we see Alex Karp from Palantir. Everyone sees the success of Palantir, whether you like their position on the defense stuff that they do.

>> Put that aside.

>> They're successful because they're thinking about things that way. And that's what customers are doing right now.

>> They're like, okay, we got to look at the ontology and start mapping our knowledge in order to get the intelligence.

>> That's a nice cross. And we're going to do a lot of— I'm going to be hosting two GraphTalks with Neo4j, your company, in San Francisco this month, and then September in New York City. So if you're interested, check that out.

>> But this is the topic of conversation for practitioners. We're talking about—

>> this is serious players looking at it in the enterprise. They're going down that road. They're building ontologies.

>> They're building the data modeling, not in the old way, in the new way with graphs.

>> And our customers have been doing this for a long time. So ontology, which has been popularized by Palantir, is not a new concept. It's a way to organize data. Graphs are the natural structure for organizing it because, knowledge is deeply interconnected. And the good news is that AI tooling has made it much easier to create and maintain ontology. So it used to take multiple years for customers to go and have people manually sort through and make sense of the meaning in their data. Now we've been able to build tooling thanks to AI where you can point at a database, click a button, it'll come up with a graph model, click another button, it'll bring the data in and then do likewise with entity and relationship extraction from unstructured text and ontologize that. So AI is— it's a virtuous cycle.

>> Great win for the practitioners.

>> 100%.

>> Yeah.

>> Let's talk about sovereignty because this maps also to the benefits. By doing that, you get a lot of benefits.

>> One of them here in Paris we're talking about is sovereignty because European countries, all part of the EU and around the world, want to have sovereignty of not just the data, there's all the data issues and we've seen that in the cloud, but the AI economy is money.

>> the outcomes from AI tooling is better software,, but also driving revenue.

>> You're seeing real revenue. So you start to see everyone talk about sovereignty in two contexts. Sovereignty for the data, keep it in the country or the region, but the money being made, the e-commerce that comes from AI is real

>> money.

>> Yeah.This is huge. What's your guys' take? Let's demystify sovereignty. What does that mean? Amit, you wrote a great couple of posts on LinkedIn and on your newsletter around the nuances.

>> Sovereignty has kind of been kicked around and has multiple meanings too.

>> Talk about multiple meanings.

>> What's your take on sovereignty?

>> 100%. In an encapsulation, sovereignty is exerting agency and control over your AI. And then how do you go and decipher that into— you have to be free and clear of state, economic, and threat actors overtaking any level of control over your stack. So if it's data and data residency as an example. That's only one aspect, but how do you go and assert that control to make sure that nobody— you're not paying rent to somebody else, in other words, and the unit economics around that, because they hijacked your business because now they own a certain pillar, within that region.

>> Yeah, and there's so much validation on that too. just the question I've been having earlier this morning about whether the weights on open source models are open. Again, that points to control.

>> Sovereignty for enterprises is another definition. So you start to see this become mainstream, Philip.

>> okay, what am I missing?

>> Is it ready for prime time?

>> And that's why I like what Philip and his team are doing. I think there's a lot to talk about beyond the residency part of it. So maybe I'll give you the stage for a second and talk about where your concept of sovereignty falls in, because we have kind of these talking points.

>> Yeah, so first question is sovereignty with respect to what? So there's nation-state level of sovereignty, which gets a lot of the press, but also companies have their own sovereignty as well, ability to carry out AI operations, but also ability to secure their data. And then finally, individuals do too, and individual data is stored by these other categories, but many nation-states guarantee rights to people vis-à-vis their own data. So these three are all interrelated, but let's talk about it from a standpoint of a company. If a company's primary moat is their context and their knowledge, meaning not just what's in all the data in any one silo, but pulling out the signal, connecting it so that any AI agent can use any data as appropriate to make a smarter decision based on the entirety of the company's knowledge, then that thing becomes ultra important. Meaning you want to make sure no one else can shut it off, and you want to make sure that no one else can access it. And it's up to each— I've seen different companies with different approaches vis-à-vis the logical or physical sovereignty. So there's a big debate and move of the pendulum swing from everything's in the cloud to now we're going to go back to self-managed. I think those two will coexist for a while. It's obviously a huge amount of effort to rebuild all of your infrastructure, fabs and chips all the way up. Yeah, but so you need to prioritize what you care about most. But I'd say in the range of priorities, the most important is owning your own data, making sure that you only can access it, and that also you're accumulating that knowledge because you have data network effects that the more different data sets you bring together, the more the accumulated body of knowledge becomes valuable.

>> Yeah, we've had conferences on chips, cloud, on-prem.

>> Data and knowledge is the lifeblood of an organization.

>> That's the brain.

>> Yeah.

>> that's why I love graphs so much. Having that brain needs arms and legs. And right now, only about 1% of enterprise data is actually out in the open.

>> So that's going to unlock.

>> Yeah. And the brain's a perfect analogy because the graph is just storing and processing data the way the brain does. It's—you have neurons and synapses, nodes and relationships. So it's— I often describe it as the left brain to the LLM right brain. LLMs are spontaneous, creative. They make mistakes. You don't know why. They're usually right. But sometimes they're not. And so having the capacity to do AI with a full brain, both hemispheres, is ultra important. And so that's really at the core where graphs fit in.

>> the big conversation around sovereignty and intelligence, we're all talking about general intelligence, AGI, and all that stuff.

>> General intelligence is like the LLMs.

>> They have all the general knowledge of the internet. You want to know how to do something or basic stuff, but specialized intelligence is super hot. This is the new frontier, the frontier that we're seeing emerge super fast here in RAISE and certainly in Silicon Valley, in New York and around the world is that the specialization of domain-specific things, departments and companies or countries, that's the knowledge, that's the entity, that's the real holy grail we're seeing. Thoughts?

>> We call it the crown jewels. You have the hardware, you have the intelligence layer, you have the data layer, You obviously have the FinOps and the whole unit economics layer and then the tech stack and everything has to be copacetic. Lose one, you lose it all.

>> What is the big challenge of sovereignty?

>> Is it definition?

>> Is it technical?

>> Is it change management? Is it understanding?

>> What's the take on this, guys?

>> Because this is something I'm trying to figure out where the best leverage is.

>> we all know the 5-layer cake.

>> There's no data in there.

>> Okay.

>> okay, well, of course it's from NVIDIA, but they should care about that because data, they do a lot of data processing.

>> It's a startling omission. And actually, this is one of the things I'm planning to cover later is, okay, chips and infrastructure. you could— I throw infrastructure and chips together into the same bucket. And above that, you have your data layer and your data can be used It's what's used to train models. It's what's used alongside models and used to create the intelligence for sure. But coming back to your sovereignty question, first and foremost is understanding how much do you care about what different aspect of sovereignty? Because the more you pull the thread, the more it can get very expensive and very time-consuming. And at the end of the day, what risk are you trying to mitigate for? And so, on the spectrum of things to care about. Economic value chain is something a lot of companies here in Europe care about, that some economic value accrue to European companies. But at what point are you— at some point someone's got to rely on chips made in other places. And these are global supply chains. So really what's the most important is making sure that your data is intact. And making sure that no one can last week was a big rallying cry for people here at the conference of all of a sudden this model was yanked and no one had access to it. And that sort of brings up another sovereignty issue is access to models. So that's pushing a lot of the conversation here towards open source models as well.

>> Yeah, me and Philip share the same thought here. It's not binary, it's not zero or one. Sovereignty could have different layers to it, different elements to it. And economic value, in my opinion, is the biggest one. We're also selling to enterprises, not just to governments. And it doesn't have to be military grade and air-gapped environments for it to be sovereign. It could also have different aspects. And the data is the biggest crown jewel of them all.

>> Okay.

>> So pretend I'm like an architect for a country or a company. What do I have to do to be sovereign?

>> Is there a technical playbook? Is there like a cloud native, do I stand up sovereignty? Like you mentioned pulling the thread, it gets complex and costly.

>> Where do I get over my skis?

>> What's the—

>> how do you guys view the deployment or just laying out the architecture?

>> What's your take on what's the best practice? What's the best approach?

>> Start us off.

>> So we came out with a 5-layer approach. The first one is territorial, that's the obvious one, right? Where's the data handled and so forth? But then you have operational. Who's actually touching my processes? If I'm Telefónica and I have Huawei as an example setting up these data centers, who does Huawei answer to? Is it to the European Union or to Beijing? So that's another example of where sovereignty may lie. Then you have the stack itself. Is it open source? Is it forkable? Are you able to actually go ahead and see what's inside? Open Weights model comes into that place. Then you have the fourth layer, which is around legal, right, where jurisdiction lies. And then the fifth one is unit economics. Okay, is it feasible? Is my favorite model that may have been included in my all-in subscription now part of the API and now it's 10x more expensive because the vendor decided so. So there's different litmus tests that you have to run through in order for it to be encapsulated.

>> So it sounds like a cloud-native, DevOps conversation to me. Things you're rattling off, it's like, okay, we didn't—didn't we do that before with cloud,

>> or is it the same kind of thing.

>> How should people think about that?

>> Yeah, at some level this is another platform shift, and with each platform shift you need to totally rethink everything. And particularly in a world that's tightly coupled and where people are trying to repatriate concerns, you also need to rethink everything. I want to throw out maybe an alternate way to look at sovereignty and tie it back to how reasoning happens. Because in some way you could say, to the degree that I'm depending on a model to do my highest stakes reasoning where a lot of comments I hear from CIOs is that certain decisions are so high stakes, you need 100% accuracy. And then for either health and human safety, regulatory, or other reasons, you need explainability. That models are not necessarily suited for all of your reasoning. And this isn't surprising in some ways. There's this movement back in the '90s of neurosymbolic AI, that you have certain kinds of AI that's deterministic. And I'm also seeing the beginnings of a pendulum swing back to this. And so from a sovereignty standpoint, you could say having optionality to be able to run certain kinds of decisions deterministically versus non-deterministically in a model is also a form of exerting one's agency. So if I look at sovereignty from a standpoint of if I'm just going to throw it to a model, it'll do its own thing, and I'm not involved in the decision. Having an alternative to that where you actually exert agency over what those business rules are and they run exactly the same every time is something else you get in a graph with multi-hop reasoning that's become increasingly popular once people adopt it as a knowledge graph.

>> All right, final question for you guys is where are customers at right now?

>> How would you peg the progress adoption, awareness. Where are they in this progression? Because obviously sovereignty speaks to everything we're talking about—

>> control, open, closed, weights, security.

>> These are technical conversations. There's mechanisms involved, there's databases.

>> Where are the customers in this journey?

>> Where are they? Down the road? Are they kicking the tires? Are they just jumping in more advanced verticals?

>> What's this?

>> What's your take?

>> On a scale of 1 to 10, I would say we're at around 1 or 2 in terms of the migration path or the thought process behind it. You have some advanced users who are already running production workloads. Many are still experimenting. They're still getting their harnesses configured. They're trying to figure out which models should be whitelisted or otherwise in a company, and are we going open source and open weights, or are we gonna leverage the models from the AI labs. So I think that process is— we're still in this

>> phase—they're in formulation mode.

>> Correct.

>> Pretty much. And what's your take?

>> Yes. And they're also moving very fast. So, but the buying criteria are rapidly evolving. And so having optionality around open weights and transparency and choice of data residency, being able to run in your own data center or choice of clouds, being able to encrypt your data such that you only hold the key. These are increasingly becoming table stakes.

>> You jumped ahead on my next question which is the velocity, because I wanted to know on the speed, what's going to be the keys to make that go to 8?

>> What's the driver for forcing that speed?

>> Is it just adoption because they see the value.

>> So startups are at 11, which is interesting right there. And they're just they have nothing to lose and everything to gain. And then to the degree enterprises are using those startups' technology, that accelerates things. Enterprises, obviously, the larger they are, the more they do have to lose and the more regulation and process they have. So a lot of the conversations in yesterday's CXO Summit were executives sharing where they are in their journey. But there's a huge appetite for training up their workforce. Retaining people, upskilling them, and also using the technology for top-line value, not just bottom-line savings.

>> Yeah, I would agree.

>> I think the AI adoption is going to be super fast because they see the value.

>> Yeah.

>> And it's going to come down to weighing the gain-to-loss ratio on the risk side.

>> Decisions have already been handed from the board level down, so it's just about pulling everybody up with them, training up on skills. Governance guardrails in order for them to consume it responsibly.

>> Well, we're trying to pull people up here on theCUBE. Guys, thanks for coming on and sharing. Good, we'll see the Neo4j customer base in San Francisco in July and New York in September. Amit, always great work you're doing at the Labs.

>> there's a lot of technology implementation to think through, reference implementation, but also the business model piece.

>> So a lot of great content there.

>> Thanks so much.

>> Thank you very much.

>> All right, I'm John Furrier, theCUBE. Sovereignty is a technical and business model opportunity for an organization, country, or region. It's going to be super important because you got to know who's touching what.

>> Of course, having control and not getting blackboxed by a model or having something pull away from you.

>> Of course, controlling the knowledge and the intellect is the new moat. The data drives intellect, and that's what we're seeing here at RAISE.

>> Thanks for watching.

Philip Rathle, Neo4j & Amit Govrin, Agentcy Labs

search

>> connection Silicon Valley and Wall Street.

>> I'm John Furrier, hosting theCUBE here with Gabor Latvay, my co-host.

>> Welcome to The RAISE Summit coverage brought to you by Solidigm with support from Tensordyne, d-Matrix, and Argentum AI.

>> Welcome back to theCUBE's livestream here in Paris, France. I'm John Furrier, your host of theCUBE. We're here for the RAISE Summit. Day 2, wall-to-wall coverage of theCUBE, live streaming again from Paris. Of course, the top story is AI infrastructure that's enabling massive change at the agentic layer, and of course, AI-native software layer emerging. This session also is talking about sovereignty, because we're in Europe, they care about sovereignty, but it's everywhere. And that's the topic of this one, is that Philip Rathle's here, CTO of Neo4j, the leading graph database. Phil, great to see you again.

>> Pleasure to be here, John.

>> As always, going to be great. And Amit Govrin, who's the CEO of Agentcy Labs, Agent with a T, CY. New venture. Congratulations on your startup, which is based in Palo Alto.

>> Appreciate it.

>> We'll unpack the hell out of sovereignty.

>> Sovereignty is a big topic, as we all know.

>> Philip, you're going to be on the main stage giving a talk on not just graph database, but in general, the stack of AI. And we haven't seen this movie before. We've seen comparables in cloud native, digital online, a lot of graph databases, but now graph databases and all the underlying context is going to be the big topic.

>> What's the main talk track?

>> Context is a huge topic here. And so if you look at the AI stack and what part is being commoditized, the applications and code are clearly getting more and more commoditized. And then at the lower end like everyone's got access to the same models, same GPUs. Okay, there's some scarcity issues with that, but that's most likely temporary. So then what ends up being durable is your knowledge, your context, your meaning, your ability to connect up all the different intelligence and data that's sitting inside all those different silos of your business and then making that available to agents. So the panel I'm on a bit later today is on the, just talking about the AI stack. And obviously I'm representing what I believe and many of our customers believe is the the most durable moat, which is context.

>> On the data side, this is a huge topic and it comes up all the time.

>> You got to harness— got to get those harnesses in place. I just saw a blog post from Databricks and the cost is really around harness selection.

>> Which points to the context. And talk about what your customers are doing because now you're starting to see the data piece come into it.

>> You guys are exposed to a lot of the top enterprises who are using graphs. They understand their data.

>> What is the current state of the data relationship to the stack and what's that look like?

>> So what a lot of customers are finding and the way they're bumping into graphs is you can have even if you have all your data centralized into a Databricks or a Snowflake or the like, that doesn't necessarily mean that your knowledge is centralized or you know how to connect things. A trade-off of these technologies where you can just dump effectively everything and consolidate in one place is that you actually give up the ability to connect the signal amongst the noise and differentiate the two. So what the harnesses that we're seeing that are being successful are the ones that have access to a system of context, knowledge, meaning. Words like semantics and ontology, which were very kind of niche graph terms for many years, have come back into vogue. And it's because models can be trained on your data. And then if you unleash them on all the data in the entire company, pull back massive amounts of context. A context window, I guess this word means many things, but if you send millions and millions of tokens, that's not only expensive, but you're praying that your model is going to be able to sort out all the noise on the fly and do it in a short period of time or cheaply get the right answer, that doesn't work, nor is it going to work anytime soon. So synthesizing your data into a system of knowledge and then having this left brain, right brain behavior where the LLMs can go into a system where your knowledge is organized.

>> it's interesting, last year when we were here at RAISE, you and I on camera and outside when we were having coffee at the cafes, we were talking a lot about graphs. We love graphs.

>> Everyone knows I love graphs, so it's like my bias is already out there. But a lot's changed this year.

>> What's been the big change for graphs, and why is graphs the perfect scenario for AI?

>> Because I think everyone understands context and that that's important, but what, in what way does context show up? What you really need is information about the actor and the things being acted upon and the system so you can understand multiple levels out what that whole system looks like, whether it's connecting your supply chain data back to inventory, back to customer, or whether it's connecting inside of a payment graph to understand multiple intermediaries so you can have AI that will catch fraud. It comes down to pulling your— giving your AI system quick access to what's the data that really matters, the signal, how that signal is connected to basically preprocess it. What doesn't work, and there have been some meta studies on this looking at thousands of research papers on what makes AI more or less accurate is you can't just throw infinite amounts of text into a context window and hope to get a better answer. So it turns out knowledge graphs solve for hallucinations. They solve for the black box problem because they can give you explainability and they importantly solve for governance insofar as you can have a knowledge layer that filters out what's appropriate according to security, privacy rules and corporate firewalls.

>> It's interesting, too the success of some of these terms like ontology is in vogue.

>> And we see Alex Karp from Palantir. Everyone sees the success of Palantir, whether you like their position on the defense stuff that they do.

>> Put that aside.

>> They're successful because they're thinking about things that way. And that's what customers are doing right now.

>> They're like, okay, we got to look at the ontology and start mapping our knowledge in order to get the intelligence.

>> That's a nice cross. And we're going to do a lot of— I'm going to be hosting two GraphTalks with Neo4j, your company, in San Francisco this month, and then September in New York City. So if you're interested, check that out.

>> But this is the topic of conversation for practitioners. We're talking about—

>> this is serious players looking at it in the enterprise. They're going down that road. They're building ontologies.

>> They're building the data modeling, not in the old way, in the new way with graphs.

>> And our customers have been doing this for a long time. So ontology, which has been popularized by Palantir, is not a new concept. It's a way to organize data. Graphs are the natural structure for organizing it because, knowledge is deeply interconnected. And the good news is that AI tooling has made it much easier to create and maintain ontology. So it used to take multiple years for customers to go and have people manually sort through and make sense of the meaning in their data. Now we've been able to build tooling thanks to AI where you can point at a database, click a button, it'll come up with a graph model, click another button, it'll bring the data in and then do likewise with entity and relationship extraction from unstructured text and ontologize that. So AI is— it's a virtuous cycle.

>> Great win for the practitioners.

>> 100%.

>> Yeah.

>> Let's talk about sovereignty because this maps also to the benefits. By doing that, you get a lot of benefits.

>> One of them here in Paris we're talking about is sovereignty because European countries, all part of the EU and around the world, want to have sovereignty of not just the data, there's all the data issues and we've seen that in the cloud, but the AI economy is money.

>> the outcomes from AI tooling is better software,, but also driving revenue.

>> You're seeing real revenue. So you start to see everyone talk about sovereignty in two contexts. Sovereignty for the data, keep it in the country or the region, but the money being made, the e-commerce that comes from AI is real

>> money.

>> Yeah.This is huge. What's your guys' take? Let's demystify sovereignty. What does that mean? Amit, you wrote a great couple of posts on LinkedIn and on your newsletter around the nuances.

>> Sovereignty has kind of been kicked around and has multiple meanings too.

>> Talk about multiple meanings.

>> What's your take on sovereignty?

>> 100%. In an encapsulation, sovereignty is exerting agency and control over your AI. And then how do you go and decipher that into— you have to be free and clear of state, economic, and threat actors overtaking any level of control over your stack. So if it's data and data residency as an example. That's only one aspect, but how do you go and assert that control to make sure that nobody— you're not paying rent to somebody else, in other words, and the unit economics around that, because they hijacked your business because now they own a certain pillar, within that region.

>> Yeah, and there's so much validation on that too. just the question I've been having earlier this morning about whether the weights on open source models are open. Again, that points to control.

>> Sovereignty for enterprises is another definition. So you start to see this become mainstream, Philip.

>> okay, what am I missing?

>> Is it ready for prime time?

>> And that's why I like what Philip and his team are doing. I think there's a lot to talk about beyond the residency part of it. So maybe I'll give you the stage for a second and talk about where your concept of sovereignty falls in, because we have kind of these talking points.

>> Yeah, so first question is sovereignty with respect to what? So there's nation-state level of sovereignty, which gets a lot of the press, but also companies have their own sovereignty as well, ability to carry out AI operations, but also ability to secure their data. And then finally, individuals do too, and individual data is stored by these other categories, but many nation-states guarantee rights to people vis-à-vis their own data. So these three are all interrelated, but let's talk about it from a standpoint of a company. If a company's primary moat is their context and their knowledge, meaning not just what's in all the data in any one silo, but pulling out the signal, connecting it so that any AI agent can use any data as appropriate to make a smarter decision based on the entirety of the company's knowledge, then that thing becomes ultra important. Meaning you want to make sure no one else can shut it off, and you want to make sure that no one else can access it. And it's up to each— I've seen different companies with different approaches vis-à-vis the logical or physical sovereignty. So there's a big debate and move of the pendulum swing from everything's in the cloud to now we're going to go back to self-managed. I think those two will coexist for a while. It's obviously a huge amount of effort to rebuild all of your infrastructure, fabs and chips all the way up. Yeah, but so you need to prioritize what you care about most. But I'd say in the range of priorities, the most important is owning your own data, making sure that you only can access it, and that also you're accumulating that knowledge because you have data network effects that the more different data sets you bring together, the more the accumulated body of knowledge becomes valuable.

>> Yeah, we've had conferences on chips, cloud, on-prem.

>> Data and knowledge is the lifeblood of an organization.

>> That's the brain.

>> Yeah.

>> that's why I love graphs so much. Having that brain needs arms and legs. And right now, only about 1% of enterprise data is actually out in the open.

>> So that's going to unlock.

>> Yeah. And the brain's a perfect analogy because the graph is just storing and processing data the way the brain does. It's—you have neurons and synapses, nodes and relationships. So it's— I often describe it as the left brain to the LLM right brain. LLMs are spontaneous, creative. They make mistakes. You don't know why. They're usually right. But sometimes they're not. And so having the capacity to do AI with a full brain, both hemispheres, is ultra important. And so that's really at the core where graphs fit in.

>> the big conversation around sovereignty and intelligence, we're all talking about general intelligence, AGI, and all that stuff.

>> General intelligence is like the LLMs.

>> They have all the general knowledge of the internet. You want to know how to do something or basic stuff, but specialized intelligence is super hot. This is the new frontier, the frontier that we're seeing emerge super fast here in RAISE and certainly in Silicon Valley, in New York and around the world is that the specialization of domain-specific things, departments and companies or countries, that's the knowledge, that's the entity, that's the real holy grail we're seeing. Thoughts?

>> We call it the crown jewels. You have the hardware, you have the intelligence layer, you have the data layer, You obviously have the FinOps and the whole unit economics layer and then the tech stack and everything has to be copacetic. Lose one, you lose it all.

>> What is the big challenge of sovereignty?

>> Is it definition?

>> Is it technical?

>> Is it change management? Is it understanding?

>> What's the take on this, guys?

>> Because this is something I'm trying to figure out where the best leverage is.

>> we all know the 5-layer cake.

>> There's no data in there.

>> Okay.

>> okay, well, of course it's from NVIDIA, but they should care about that because data, they do a lot of data processing.

>> It's a startling omission. And actually, this is one of the things I'm planning to cover later is, okay, chips and infrastructure. you could— I throw infrastructure and chips together into the same bucket. And above that, you have your data layer and your data can be used It's what's used to train models. It's what's used alongside models and used to create the intelligence for sure. But coming back to your sovereignty question, first and foremost is understanding how much do you care about what different aspect of sovereignty? Because the more you pull the thread, the more it can get very expensive and very time-consuming. And at the end of the day, what risk are you trying to mitigate for? And so, on the spectrum of things to care about. Economic value chain is something a lot of companies here in Europe care about, that some economic value accrue to European companies. But at what point are you— at some point someone's got to rely on chips made in other places. And these are global supply chains. So really what's the most important is making sure that your data is intact. And making sure that no one can last week was a big rallying cry for people here at the conference of all of a sudden this model was yanked and no one had access to it. And that sort of brings up another sovereignty issue is access to models. So that's pushing a lot of the conversation here towards open source models as well.

>> Yeah, me and Philip share the same thought here. It's not binary, it's not zero or one. Sovereignty could have different layers to it, different elements to it. And economic value, in my opinion, is the biggest one. We're also selling to enterprises, not just to governments. And it doesn't have to be military grade and air-gapped environments for it to be sovereign. It could also have different aspects. And the data is the biggest crown jewel of them all.

>> Okay.

>> So pretend I'm like an architect for a country or a company. What do I have to do to be sovereign?

>> Is there a technical playbook? Is there like a cloud native, do I stand up sovereignty? Like you mentioned pulling the thread, it gets complex and costly.

>> Where do I get over my skis?

>> What's the—

>> how do you guys view the deployment or just laying out the architecture?

>> What's your take on what's the best practice? What's the best approach?

>> Start us off.

>> So we came out with a 5-layer approach. The first one is territorial, that's the obvious one, right? Where's the data handled and so forth? But then you have operational. Who's actually touching my processes? If I'm Telefónica and I have Huawei as an example setting up these data centers, who does Huawei answer to? Is it to the European Union or to Beijing? So that's another example of where sovereignty may lie. Then you have the stack itself. Is it open source? Is it forkable? Are you able to actually go ahead and see what's inside? Open Weights model comes into that place. Then you have the fourth layer, which is around legal, right, where jurisdiction lies. And then the fifth one is unit economics. Okay, is it feasible? Is my favorite model that may have been included in my all-in subscription now part of the API and now it's 10x more expensive because the vendor decided so. So there's different litmus tests that you have to run through in order for it to be encapsulated.

>> So it sounds like a cloud-native, DevOps conversation to me. Things you're rattling off, it's like, okay, we didn't—didn't we do that before with cloud,

>> or is it the same kind of thing.

>> How should people think about that?

>> Yeah, at some level this is another platform shift, and with each platform shift you need to totally rethink everything. And particularly in a world that's tightly coupled and where people are trying to repatriate concerns, you also need to rethink everything. I want to throw out maybe an alternate way to look at sovereignty and tie it back to how reasoning happens. Because in some way you could say, to the degree that I'm depending on a model to do my highest stakes reasoning where a lot of comments I hear from CIOs is that certain decisions are so high stakes, you need 100% accuracy. And then for either health and human safety, regulatory, or other reasons, you need explainability. That models are not necessarily suited for all of your reasoning. And this isn't surprising in some ways. There's this movement back in the '90s of neurosymbolic AI, that you have certain kinds of AI that's deterministic. And I'm also seeing the beginnings of a pendulum swing back to this. And so from a sovereignty standpoint, you could say having optionality to be able to run certain kinds of decisions deterministically versus non-deterministically in a model is also a form of exerting one's agency. So if I look at sovereignty from a standpoint of if I'm just going to throw it to a model, it'll do its own thing, and I'm not involved in the decision. Having an alternative to that where you actually exert agency over what those business rules are and they run exactly the same every time is something else you get in a graph with multi-hop reasoning that's become increasingly popular once people adopt it as a knowledge graph.

>> All right, final question for you guys is where are customers at right now?

>> How would you peg the progress adoption, awareness. Where are they in this progression? Because obviously sovereignty speaks to everything we're talking about—

>> control, open, closed, weights, security.

>> These are technical conversations. There's mechanisms involved, there's databases.

>> Where are the customers in this journey?

>> Where are they? Down the road? Are they kicking the tires? Are they just jumping in more advanced verticals?

>> What's this?

>> What's your take?

>> On a scale of 1 to 10, I would say we're at around 1 or 2 in terms of the migration path or the thought process behind it. You have some advanced users who are already running production workloads. Many are still experimenting. They're still getting their harnesses configured. They're trying to figure out which models should be whitelisted or otherwise in a company, and are we going open source and open weights, or are we gonna leverage the models from the AI labs. So I think that process is— we're still in this

>> phase—they're in formulation mode.

>> Correct.

>> Pretty much. And what's your take?

>> Yes. And they're also moving very fast. So, but the buying criteria are rapidly evolving. And so having optionality around open weights and transparency and choice of data residency, being able to run in your own data center or choice of clouds, being able to encrypt your data such that you only hold the key. These are increasingly becoming table stakes.

>> You jumped ahead on my next question which is the velocity, because I wanted to know on the speed, what's going to be the keys to make that go to 8?

>> What's the driver for forcing that speed?

>> Is it just adoption because they see the value.

>> So startups are at 11, which is interesting right there. And they're just they have nothing to lose and everything to gain. And then to the degree enterprises are using those startups' technology, that accelerates things. Enterprises, obviously, the larger they are, the more they do have to lose and the more regulation and process they have. So a lot of the conversations in yesterday's CXO Summit were executives sharing where they are in their journey. But there's a huge appetite for training up their workforce. Retaining people, upskilling them, and also using the technology for top-line value, not just bottom-line savings.

>> Yeah, I would agree.

>> I think the AI adoption is going to be super fast because they see the value.

>> Yeah.

>> And it's going to come down to weighing the gain-to-loss ratio on the risk side.

>> Decisions have already been handed from the board level down, so it's just about pulling everybody up with them, training up on skills. Governance guardrails in order for them to consume it responsibly.

>> Well, we're trying to pull people up here on theCUBE. Guys, thanks for coming on and sharing. Good, we'll see the Neo4j customer base in San Francisco in July and New York in September. Amit, always great work you're doing at the Labs.

>> there's a lot of technology implementation to think through, reference implementation, but also the business model piece.

>> So a lot of great content there.

>> Thanks so much.

>> Thank you very much.

>> All right, I'm John Furrier, theCUBE. Sovereignty is a technical and business model opportunity for an organization, country, or region. It's going to be super important because you got to know who's touching what.

>> Of course, having control and not getting blackboxed by a model or having something pull away from you.

>> Of course, controlling the knowledge and the intellect is the new moat. The data drives intellect, and that's what we're seeing here at RAISE.

>> Thanks for watching.