A Cloud Nobody Owns: Sovereign Infrastructure for the Agent Era with Dominic Williams

0:08Aaron Good morning, good evening, wherever you are, and welcome back to the Enterprise AI Show. This is your host Aaron, and I'm back home after two weeks on the road. And, uh, quick aside, this is my, I think, fourth time trying to actually record this. Had a little bit of technical difficulties. Actually had a tornado go through and had to go to the basement, and a tornado touched down a couple miles from the house.

0:32Aaron But everyone is fine, everything is safe, uh, and we certainly hope everyone out there is safe as well. But moving on to today's show. So what we're looking at today for our topic is how to build a sovereign agent-ready cloud without relying on a single hyperscaler. So Brandon sits down with Dominic Williams, founder and president and chief scientist of the DFINITY Foundation, to talk about the internet computer, private cloud engines, data sovereignty, and an AI-aware replacement for traditional SaaS.

1:10Aaron And of course, all of that is coming up right after this quick break

1:18Brian Today's show is sponsored by Palo Alto Networks. Every time your team deploys a new cloud workload or AI agent, another identity gets permanent access to your critical systems. Legacy tools were built to manage human employees, leaving modern machine and AI access largely unmanaged. That's why Idira by Palo Alto Networks comes in.

1:37Brian Human, machine, AI, one identity platform for all of them. Idira replaces permanent permissions with dynamic access, so you can lock down every identity without slowing down your business. Secure every identity with Idira by Palo Alto Networks. Visit paloaltonetworks.com/idira. Again, that's paloaltonetworks.com/IDIRA.

2:04Aaron What does the outside world already know about your company? Cybercriminals could be seeing leaked credentials, compromised session cookies, exposed infrastructure, and even impersonations of your brand and executives. NordLayer Intelligence by NordStellar gives security teams visibility into those threats across the deep and dark web, data breaches, attack surfaces, and brand impersonation all in one platform.

2:33Aaron Find out what attackers know before they can use it. Visit nordlayer.com/intelligence/enterpriseai and use code ENTERPRISE10 for 10% off your NordStellar plan

2:52Brandon And we're back. Joining me today is Dominic Williams, founder, president, and chief scientist at Definity Foundation and founder and CEO at Upt Lab. Welcome to the show, Dominic

3:02Dominic Thank you for having me, Brandon

3:04Brandon All right, listen, we're gonna have a great conversation about sovereignty, private clouds, maybe a little agents, a little AI, right? We can't-- We have to talk about that as well. But before we jump into all that, we always wanna get a chance to know our guests. So give us a little bit of information about how'd you get into this, what's your background.

3:19Brandon Tell us your origin story, if you will

3:22Dominic Well you probably don't wanna hear my entire origin story. It goes back a long way. I've been what I'd call an engineering entrepreneur for some decades. I've got a particular interest in distributed systems but I've also built things like MMO games for kids. Back in 2010, I launched a game called Fight My Monster, and it grew to three million users.

3:41Dominic And that was exciting 'cause, it, it actually involved a lot of really interesting backend infrastructure that led me to what I'm doing now. But the actual product itself wasn't an infrastructure product

3:52Brandon So have you always been a gamer? What… That's like kind of an interesting story in itself. How did you get into building kids for game, or games for kids rather?

3:58Dominic I had s- some young boys and my eldest used to go to a lot of chess tournaments. And so I wanted to replicate that experience inside an MMO. And the game was conceived as something akin to a giant tournament, and it was designed to have s- to be more like a game of skill than just something where you grind away and level up.

4:21Dominic And you would generate these unique monsters, and you could fight them against other players' monsters and trade them. And it was this kind of hybrid game social network. It was nothing's really ex- existed like it before or since. A lot of kids say that it defined their childhoods, so I'm

4:38Brandon Oh, that's awesome. I know a lot of kids like Roblox and Minecraft.

4:41Dominic that, yeah,

4:42Brandon everybody's kids, I think when they're young they go through those things. So that's fun. That's super fun. That's, but that seems completely different. So now I guess flash forward, tell us about the D'Affinity Foundation. I think I hadn't heard of it as much, so why don't you assume we don't know too much about it.

4:55Brandon What is this thing that you've created here?

4:58Dominic So actually going back to the game quickly I, I built that when I was in the UK, and I later relocated to Palo Alto in California. And in the UK, it's not-- historically it's not been so easy to raise venture capital and things like that. I had previous ventures, and I had some of my own funds, so I'd funded it myself.

5:16Dominic But I didn't have lots of money, and I needed infrastructure that would scale very easily with low effort and low cost. And I pi- was one of the pioneers of NoSQL databases, and Find My Monster was actually the first sort of serious production use of a database called Cassandra. We were using it in beta.

5:36Dominic Actually lost all the data of 880,000 users. Had a very tense 24 hours getting it back. But the beautiful thing about that database is you can just add nodes to add storage capacity and write throughput, and of course, read throughput. So that helped with the scaling, and I also built a kind of custom game server that scaled horizontally.

5:58Dominic Again, you could just add nodes. And that experience, made me think about how decentralization can be used in the context of scaling. And later in California, I became very interested in decentralized networks yeah, initially like Bitcoin that were fully decentralized in the sense that there was no, it wasn't just a decentralized architecture where there was no controller, but also had these wonderful security and resilience properties.

6:31Dominic So I became a pioneer in that field working a lot of the underlying computer science theory that makes those networks run. And eventually in 2015, I started talking about the Dfinity Project, which aimed to create what was essentially a blockchain that could function as a cloud. And I said back then, "Look, we-- it's possible with enough research to create a network that can host code that's tamper-proof, doesn't need to be protected by firewalls, unstoppable, and has various other desirable properties."

7:07Dominic And at the time, people were skeptical, but, I kept persevering. And in 2016 I actually had a share in a decentralized incubator, that, that would fund and incubate projects in Palo Alto called String Labs. And I persuaded my co-founder, "Hey, look, let's incubate Dfinity. Let's try and do this." And by the end of that year, we c- created the Dfinity Foundation in Switzerland.

7:31Dominic And Then in 2017, we ran a fundraiser got some funds into it and it became independent of the, the incubator. And that's how it began. By the end of 2017, we had a pretty good team running. So some very famous people. Actually, our employee number five was the co-inventor of WebAssembly, for example.

7:52Dominic WebAssembly, is this new-- is the Internet's compute format, and that's now present in every single web browser. And a lot of people are using it for backend compute too, like Cloudflare supports, WebAssembly workers and things like that. A guy called Timo Haneke who was someone who'd been prominent in the Bitcoin community, designing algorithms for Bitcoin miners and things like that.

8:12Dominic So a lot of interesting people working on this in the early days. But we scaled it out, and by the time the network launched May '21 we had about 200 people predominantly researchers, very PhD heavy. I think we used to hire mainly from Google. And then over time, we had two research centers one in Palo Alto, another one actually in San Francisco as well, and then another one in Zurich.

8:35Dominic Over time, the center of gravity swung towards Zurich for a bunch of reasons. And in 2021, I relocated from Palo Alto to Zurich, and that's where I am currently, although I still spend a lot of time in California.

8:49Brandon Okay.

8:50Dominic And

8:51Brandon so what kind of, what's the mission? So I mentioned lots

8:53Dominic the mission is

8:55Brandon And I th- I know you're gonna get into a little data and sovereignty a little bit. So what is it like today? Think of it what's the key problem that you've settled on as you've gone through these different, phases of it?

9:03Dominic Yeah. The key problem for me was to really just re-in-- to create a new backend compute stack, an end-to-end, a new end-to-end backend compute stack. And I think that's im- that was important to me because, I've been coding personally for many decades. I started coding when I was, like, 10 years old.

9:21Dominic I've built just about every kind of system you can imagine, and therefore I'm acutely aware of the shortcomings of the traditional backend stack. So I wanted to create a completely new backend stack that solved a, a wide array of problems. And- That included obviously the simplicity of software. I wanted it to be much easier to write backend software and systems and services.

9:47Dominic I wanted to solve for security. I wanted to create a stack where everything you built was secure by default, and you didn't need a security team to protect it. And you didn't need patches and firewalls and configurations and so on. I wanted a stack where software would be guaranteed to run, so you didn't need a sys-sysadmin team tinkering with it.

10:05Dominic And then the other piece of it for me was I wanted this backend stack to be a cloud naturally. And I wanted that to be part of the internet. I wanted to extend the internet and create a cloud the internet way. Essentially I started, coding before it sounds crazy, before, before the web. And so I started out just programming sockets using C a long time ago.

10:31Dominic I was always very interested in the internet and, I rode the whole kind of internet boom, if you like, and the internet connects software. Like right now it's connecting our, video conferencing client software. It connects our web browser to a server. It allows SMTP to relay, relay emails. Y- and that's what the internet does, it just connects software.

10:52Dominic And on top of that, it's created this sort of global open marketplace and afforded us all these social freedoms. And it's a really cool thing, like nobody owns the internet. The governance is a bit centralized through the ICANN, but the technology didn't exist in those days to, to solve that. The internet's a wonderful thing and we much prefer the internet to America Online and CompuServe and I think Bill Gates came out with this idea of the information superhighway.

11:18Dominic He wr- he wrote this book, and initially he was proposing that, big corporations i.e. Microsoft would create this information su- superhighway, and it'd be much better than the internet. And thankfully, that's not what we got, and he then, I think, changed, cha- changed the next edition of the book, he rewrote it.

11:34Dominic And so yeah, we have this, this global internet that nobody owns, and it's a key foundation of mon- modern society and the global economy and everything else, and the way we live. So I'm… My view is can we use a network protocol to extend the internet so that you can build on the internet? Like the internet can actually host your software and run it.

11:56Dominic Wouldn't that be great?

11:57Brandon Okay. And then this is where, is this where the internet computer, as you referred to it, is

12:02Dominic that's why it's called the internet computer. Exactly, yeah

12:04Brandon yeah, why don't you give us a sense, 'cause I think what's going off in my mind and maybe other minds is so we have the traditional hyperscalers, you've mentioned a couple, AWS and Azure, and then I think you have this idea, a slightly different take, like kind of the internet computer. So maybe like contrast like exactly what is the internet computer, and then for all of us that are familiar with hyperscalers why is it different?

12:23Dominic Yeah. So first of all the Internet Computer is the world's only compute stack created by a network. So a mathematical network combines nodes which are run by independent parties professionally. And they can also run on c- on, on hyperscalers too. And it combines these compute nodes to make a single unified serverless cloud platform where code is tamper-proof and always on and so on.

12:51Dominic And that's very different to the way traditional clouds are created. Traditional clouds are created by, taking a machine, installing an operating system, installing virtualization software, then the customer of the cloud installs their own operating system on top of the virtualization software, and then they install their software and so on and so forth, right?

13:09Dominic As it turns out, there's just no way that you can make a stack of software y- a system that, that consists of a stack of software all the way down to the silicon secure. You just can't do it. You can find holes and plug them and just hope that you find holes and plug them before an adversary finds the hole ex- and exploits it.

13:28Dominic And that's what a zero-day exploit is, right? That's a hole in your stack that that nobody knows about yet. And especially in the era of AI, this is a kind of risky way of, creating compute stacks. You're d- really gambling that your AI is more powerful than your adversary's AI, and therefore it's gonna find all the holes, and so you can plug them before they do, but before they exploit them.

13:50Dominic And I always thought that was the, the, the, you know, that was a key thing we need to fix, and it's become increasingly true. Al- also, of course, that's very vulnerable to disruption. If, if you've-- if you're running systems on a on, on machines that are just a stack of software onto, down to the silicon and someone blows up those machines your systems and services and apps and so on are gonna stop working.

14:10Dominic So I wanted to solve that too. Internet Computer is powered by a protocol called ICP, Internet Computer Protocol, and it combines the compute capacity o- of these machines to create a single unified serverless cloud environment. It's different to tradit- traditional serverless clouds in the, in, in the sense that there's data in the serverless environment.

14:29Dominic And we'll get back to how that works. And they can also process web, so they can serve their own user experience. So it's very different to the kind of serverless software you find on Amazon Web Services. And essentially what's happening is that protocol ICP is redundantly replicating your computation and data.

14:48Dominic You don't see it, of course, it just looks like a cloud but inside the protocol, that's what's happening. Your data and computation are being replicated. First of all, it's redundant, so you can blow up some machines and all your systems and services keep working. But also the way those different machines doing the replication synchronize themselves is in- involves something called Byzantine fault tolerance which is a sort of arcane branch of y- computer science, distributed computing that's, been popularized within the-- by blockchain.

15:17Dominic And this essentially ensures that even if some of these nodes start misbehaving the, the cloud itself continues functioning. Let's say, you created what we call a cloud engine by combining four nodes that were dedicated to you, or seven nodes. Yeah, the seven nodes, you could have two of those nodes controlled by Dr.

15:39Dominic Evil. You thought you, you went to a marketplace, and you selected these nodes to be part of your cloud engine. Turns out those nodes are run by Dr. Evil, and the nodes are corrupting your data or corrupting the copies of the data they're responsible for and sending, not following the protocol and sending bad messages to the other nodes.

15:58Dominic What's really cool with Byzantine fault tolerance is that they will be unable to stop your cloud engine functioning. So all of your apps and services will continue running without a hitch, and none of their computation and data will be corrupted. That's a really po- powerful property. In addition to that, there's no infrastructure to hack.

16:18Dominic Like your software and data live within this network, this mathematical network protocol. It's only possible for that network to run your correct software logic against your correct data. There's just no other way it can work. So that's a guarantee. If your app has some-- you've written your app's lo-logic and there's data associated with that, the network guarantees that only the, the logic that you've written can run and against its correct data.

16:45Dominic So you don't need a firewall, you don't need security admins and so on. And the network also is redundant, and so even if some of these nodes fail, even if some of those nel- nodes get taken over by Dr. Evil, the whole thing keeps on running.

16:58Brandon Okay. So I like the the Dr. Evil. That's good. So maybe take us inside, 'cause this all sounds good, but as, I think probably many of us listening would be familiar with you write some application logic, maybe you put it in a container or you package it up in a server sl- service, serverless or, Lambda-type format.

17:14Brandon And then of course, like on the data side, we're all familiar with like S3 and files and obviously databases. So like what am I like, what am I like packaging up here so that you can go run this on kind of all these other nodes that are out there ready for me to take advantage of?

17:31Dominic From you- your perspective as a developer this is just a serverless cloud, and you can just deploy to it. And and, t-today probably you're not writing the code yourself. Probably you're using Claude Code and, Claude Code's building this thing, this app, this service for you locally, and then when you're happy with the latest version, you tell Claude to deploy it to the Internet Computer.

17:51Dominic So really, you're just interacting with this thing thr-through, through your AI prompts. It's all very seamless. And I don't wanna get too technical, but, for those that have a technical background, you can map a domain name to multiple nodes and things like that, right? So although it ha- it might have one internet address, behind that, there can be lots of individual com- compute nodes.

18:11Dominic So one of the cool things, it's serverless, right? And it, you don't need dependencies like databases. So in actual fact the code, the logic runs in persistent memory. So rewinding a bit to what code runs actually you're running WebAssembly in that persistent memory. Any, any language that compiles to WebAssembly can be used within the serverless environment.

18:35Dominic So the two most popular languages are Rust and a domain-specific language called Motoko. But people are also running other languages on the Internet Computer. Now, you might be writing it in Rust, but there's no file system there's no database to connect to and the reason is that your code is running in persistent memory.

18:54Dominic So o-one way to understand that is if the power on your laptop gets low, it goes into hibernate mode And all the memory pages just get written to disk. And then when you put the power back in, it comes back on and it loads the memory pages from disk and your desktop experience continues f-from where you were.

19:11Dominic If you had a bunch of apps open, like sort of Photoshop, you were editing a photo or something, your mem- your power on your laptop ran out, it would hibernate and sync all your memory pages to disk. And then when you put the power back in, it will resume, like bringing those memory pages back off disk, and there will be your Photoshop app with your photo that you're editing where you left off.

19:33Dominic So it's a bit like that, but obviously rather than a sort of clunky hibernate, like these, these memory pages are being persisted constantly. And obviously this is greatly aided by the replication. If you're running some code on the Internet Computer, in actual fact, that identical logic is running against the identical data on several machines at once.

19:55Dominic Ev-even if one of those nodes blow up there's still copies of the data on the other machines and so on and so forth. You can lazily sync that memory to… Those nodes can lazily sync the memory to the, to, to their

20:05Brandon Okay, so that makes sense on the-- So I get I kinda take then a- and I can really write anything that can you know, compile down to WebAssembly. That makes sense. Lots of great solutions for that, and we've actually had some guests on the show talk about that. So that part totally makes sense. I guess I'm a little bit like, okay, so then, I don't know, just pick anything.

20:19Brandon I got a database, like a lookup database of whatever, customer data. Let's say it has 1,000 customers in it, and, the, the code looks it up, displays it, and maybe it makes some updates. That's the part I guess I'm if I give that, how do I give that to you? And then h- and then more importantly, if I give that to you, like, how am I gonna make sure that it isn't, ac-accessible to, intruders, other people

20:41Dominic Oh yeah. That's the second… Yeah, that's the second question. So first of all, with respect to databases memory's much faster. Persistent memory is much faster. So if you spend a lot of money, if on Oracle's solution, it'll give you a symmetric SQL database where, you have a bunch of nodes that all share equally the role of replicating the database.

21:02Dominic Like each, each node has a copy of the database. And the more nodes you add the more read, query capacity you add, but they all do the write. But then, the architecture would evolve. Like you might have like some, an applic- you know, some front end Servers running Node or something orchestrated by Kubernetes, like on Amazon Web Services, and then they are forwarding database requests back to this sme- Oracle Symmetric or Oracle cluster.

21:26Dominic And those application servers are having to marshal data in and out of the, of database requests. All of that's pretty slow, and you've got complex because two different apps might try and update the same field in the database simultaneously, and the, the first one might succeed, and the second one might override it.

21:43Dominic And you've gotta manage all this infrastructure, and it's expensive. On the Internet Computer the memory just lives in… Or the data just lives in memory, excuse me. You don't have to do anything special. In fact the, the data lives in the logic that you write. So you just li- you just write the, logic, the logic that describes the essence of your application or service.

22:06Dominic And that's it. It- it's called orthogonal persistence. You don't have to do anything about persistence because that code is running in memory that's persistent. So if you create a variable or a data structure that's where your data lives. There's no need to copy it into a database. There's no concept of the code crashing and needing restarting, and therefore you need to copy it into a database to keep the, the data safe.

22:28Dominic It, the memory's updated transactionally and that's where your data lives. That massively simplifies the w- what's involved creating the functionality of your application or service. It massively simplifies it, massively reduces the amount of code involved. But it also hugely im- increases the performance because there's no levels of indirection.

22:48Dominic If you run a query the code is just pulling that data directly from memory, and you imagine what would happen with the Oracle database. Guess what? You send a query to your application through some web front end, right? It's got to package that up and put it in an SQL query and put it in a database connection pool, which sends it across the, you know, network to one of the, the, the database servers which unpackages it, runs it against all its indexes and through the query processor and everything else, sends the results back.

23:18Dominic You gotta unpack, un- unpackage that, marshaling stuff in and out. There's a lot going on. So all of that's gone, and you get much better performance. In a sense so the, the new thing that's coming, which we think is what's gonna unlock massive enterprise adoption of the Internet Computer is called cloud engines.

23:34Dominic And cloud engines so this kind of answers- your other q- part of the question, allow the enterprise to select the nodes from a marketplace. So there's about 1,500 nodes in the marketplace now, and you have to select nodes that are the same class. If you select a nano node, then all the other nodes in your cloud engine have to be nano nodes.

23:54Dominic If you select a dense node, all the other nodes have to be dense. The protocol combines these nodes and creates this, private agentic cloud infrastructure for you. And you choose, you choose who, who's providing the node where it is, what data center is it installed in, or maybe it's on a hyperscaler infrastructure.

24:15Brandon so the last

24:16Dominic so you control that

24:17Brandon That makes sense. So as we kinda get everyone comfortable with this. So I think I'm, following along here. So obviously application code, we get that in WebAssembly. Sounds like you're gonna put everything in memory. We'll come back to that in a second 'cause I'm wonder- I wonder a little bit more like what if I have a lot of stuff, but table that for one second.

24:30Brandon But then more importantly, I think what people are gonna wonder is, okay, so if I pick these nodes, like what's the… I can obviously see the blockchain influence here, and I'm thinking like, like the hashing and all this. But like what's the security protection I have so that when I pick these nodes that that my runtime can like interact with th-the data in memory, but, you what'd you say?

24:50Brandon Like Dr. Evil, if I did pick the Dr. Evil node Dr. Evil can't get my get into my memory and read it. So how's that working?

24:58Dominic Internet Computer Protocol is implemented by a p- piece of software called a replica, and it supports TEE environments if you want. The Internet Computer currently does support running apps and services inside TEEs. So this is basically a kind of hardware encapsulation. So for example, if you went and got one of these traditional com- internet computer nodes, about 100 companies run them.

25:23Dominic If you went and got one of these traditional internet computer nodes a- and open it up, you'd just find random bytes on the hard drive and in memory, and that's because it's running a thing called SEV-SNP, which comes from AMD. So there's, you can there are hardware privacy protections.

25:39Brandon Okay, so every, to be on this basically it's a trusted computer. It's basically got all this. To be a node, you're basically saying this node is like running some type of like confidential trusted computing kind of, layer. Is that a, a fair way to say it?

25:53Dominic They don't all run it with cloud engines. You can choose. And some of these nodes will support that and others not. But you can choose nodes from rep-- providers that you trust. Yeah, obviously there's a risk if you choose a, a small- a smaller nodes from a smaller company, and you go to their website, and it all looks a bit shady.

26:10Dominic Maybe they will be poking around in your memory. But yeah, if you want, you can choose nodes from established providers that are clearly trustworthy. With cloud engines, you can also use on-prem nodes, so you don't even have to use someone else's node at all. And the idea very much is that whoever, creates a cloud engine or cloud engines, some enterprise generally has complete control over, who's running the hardware, what the hardware is where it's located.

26:36Dominic And they can change that at any time. What's really cool is, for example, you could create a cloud engine that runs on hyperscalers. So let's imagine, let's say you started off running a cloud engine that where the nodes are in, are running on different Amazon Web Services data centers. So Amazon has a lot of different data centers and, because the nodes are in different Amazon data centers if one of the data centers hosting your node went down and they do actually, there are, pretty regular outages of Amazon Web Services data centers.

27:08Dominic Then it's okay. Your cloud engine would keep on running because it's fault-tolerant, right? You could even do it in the Middle East, which has had a lot of problems with drones fi-flying into Amazon Web Services data centers.

27:16Brandon Yes.

27:17Dominic and again yeah, you could, so long as all of the data centers didn't get hit by drones,

27:21Brandon long as someone, someone's left, someone knows what's going on, right?

27:24Dominic Yeah. Yeah, you-- So let's say you start off like that, and then you realize, I forget how long ago it is, time flies, but I do remember, was it 10 years ago, 15 years ago? There was a time when all the Amazon Web Services data centers went down. They had something, some problem with their message

27:39Brandon Yeah, like a configuration error went crazy.

27:41Dominic Yeah, it actually all went down. Yeah, so you could then think actually, I'm not so sure about that." And so you could you would just… What you'd do is you just add in new nodes from, say, Google and Azure and Oracle Cloud or whatever, IBM Cloud, whatever you want. And and then once you've done that, maybe you delete some of your Amazon nodes.

28:00Dominic And now you've got, a much more re- reliable cloud engine because it's very unlikely that failure's gonna be correlated. If Amazon goes down, it's very unlikely Google's gonna go down.

28:11Brandon Угу

28:12Dominic And so you then, you're very happy with that. And and by the way, when you reconfigured your cloud engine by adding the new nodes and then deleting the old nodes, everything kept working. There, there was all of your enterprise apps and services that you've got running on your cloud engine, all of that kept working.

28:27Dominic There was no interruption whatsoever. Okay, but then you decide you know what? I'm against all this hyperscaler stuff, and I'm worried that, that they're spying on me or whatever." You could then reconfigure it again, and you add in a bunch of nodes run by independent providers in traditional data centers around the world, and then just delete the, the hyperscaler nodes, and it'll again keep on running, no interruption.

28:49Dominic So you know it's complete freedom from vend-vendor lock-in because essentially your cloud engine is a network configuration. And you have to be able to sync the new nodes to the network

29:00Brandon So then on, on the and then how does someone pay for all of this? So is this sort of like someone stands up some nodes and it's like a marketplace? This is like I'll give you a node for X amount of money and I pay you somehow?

29:13Dominic So there's a token economy, but with cloud engines, which I think we're, I think we're beginning to give out coup- coupons next week. It's already in use. So for example,

29:21Brandon Okay

29:22Dominic Pakistan's running a cloud eng- some cloud engines and they're using them to build like a sovereign national chat app. And they also-- they've used AI to build enterprise infrastructure that's actually been used by their government. UN is deploying this across country offices and so the cloud engines are already in use, but they're not in wider release.

29:42Dominic But I think probably next week th- they're gonna be available to all. People can just apply for these coupons and how… It's on opencloud.org for people. So in- internetcomputer.org and then opencloud.org is where you get your cloud engine. And yeah that, that's coming and pe- people will be able to cr- create these cloud engines, and they're just network configurations.

30:03Dominic And so in the payment inside the market, there's a marketplace of nodes. You choose the nodes you wanna combine and then you pay. Now this is the interesting bit. The entire cloud en-engine experience looks like it's run by a company. It looks like it's an Amazon Web Services configurator or something like that.

30:21Dominic In actual fact, it's part of the Internet Computer network. And when you pay via a payment services provider. And there'll be a few of these. Anybody can do it that the network registers. And then the payment services provider takes your money from a credit card, and behind the scenes, they convert it into something called cycles.

30:40Dominic So the Internet Computer, because it's a protocol it can't handle a bank account or something like that. Like there's no a, a protocol's an anonymous thing. It's not legal entity, right? So the protocol moves around these things called cycles and these are used to remunerate the providers, and the providers can turn those cycles back into cash because there's a market for them.

31:02Dominic There's another token called ICP that's in there. But with cloud engines, all of that's hidden. So if you're a enterprise and you go to opencloud.org you'll see a marketplace. You can choose which nodes to combine. You choose a payment services provider on checkout give them your credit card and it just looks like it's-- it looks like a centralized infrastructure.

31:24Dominic In actual fact, it's not. It's a decentralized network.

31:27Brandon And then you mentioned a little bit about, who's using it today, 'cause I think that's always And I think you mentioned Paxos. So is it-- and I know data sovereignty is something that comes up that, mentioned on your website and things like that. Is that… that seems like maybe one of the key use cases is sort of somebody that doesn't wanna be tied to any specific provider.

31:44Brandon Is that what are those kind of the people that are motivated to adopt this, or is it someone else?

31:49Dominic I think it's certainly a large segment. Actually the Internet Computer project, I s- I guess it's 10 years in, and it really, DFINITY was very much a a research institute kind of operation, and we weren't great at productization. But the technology was great. About a year and a half ago a sort of transition happened where we started focusing on productization with the help of partners.

32:12Dominic And for example, there's a product called Caffeine, it's at caffeine.ai, and this thing actually builds on the Internet Computer. You can just chat with AI, and it'll build you something on the Internet Computer. So that, that was actually quite consumer, and I think there's like some hundreds of thousands of apps that have been built by consumers and some enterprises too.

32:30Dominic Umti Labs which is a new company no one's heard of has built something called OpenSaaS, which shows the oth- is the other end of the spectrum, which is f- a replacement for every SaaS service on Earth. And it's free, and it's AI aware, so it's-- gives you it's much more powerful than traditional SaaS too, and you can just install it on a cloud engine.

32:47Dominic You've got everything from consumer through to that. The, the main use so far though, yeah, you're right that around the world there's a lot of people realizing that, tech is one of the key foundations now of modern societies, and the danger is if your tech stack is running in a foreign country or powered by code that's not open source there are hidden kill switches and backdoors.

33:10Dominic For example there's something in the US called the CLOUD Act and this requires any US-based cloud services provider whether they're providing infrastructure, like physical infrastructure, whether they're providing just software it could be SaaS or, it could be SaaS, right? If certain US agencies request backdoor access to data that they, the CLOUD Act requires them to furnish that access e- even if the data's located on servers in, in a foreign country, anywhere in the world.

33:39Dominic There's growing awareness of this kind of thing and during the latest tensions, over Greenland and things like that, people like the French were very worried that their tech was gonna be switched off by America and things like… So I think it's, I think there's just a general awareness that you really need to have around the world you really need to have your own sovereign infrastructure, and that means, the hardware's in your own country and whatever's being used to create the tech stack is open source.

34:03Dominic So you can see there's no backdoors and kill switches in there. And I don't think, I-- by the way, I'm a, my wife's American. I live in America and go back all the time. I'm-- this isn't anti-America at all. I think there'll be huge adoption of this technology in America. But it is, it's, it-- there is a reality there that, if you're a foreign count- if you're a sovereign country and you're building, your critical government and corporate systems on another country's infrastructure, very likely the other country is going to extract the data and put it in a Palantir database.

34:34Dominic Yeah, I'm British or- originally, and I'm sure the UK would do it too. This is the way this thing works. There's a general push around the world toward sovereign tech stacks just as there is for AI. People wanna buy chips from in- NVIDIA and install them in local data centers where they feel they have more control

34:51Brandon Yeah, no, without a doubt. Data sovereignty, I think is, just becomes more and more important by the mo- the moment, right? More and more countries, more and more reasons. I think that's, nothing's gonna change there. S- so as we wrap up, though, I always like to try to give people that are interested a chance to how to get started.

35:04Brandon So you kinda mentioned a little bit of AI at the top and Cloud Code or Codex and things like that. So why don't for someone that is, like all of us out there trying to do something, are there specific do you have an M- MCP server, or is there a good way to get started with this using a coding assistant?

35:19Brandon What do you recommend for someone that k- kinda wants to know

35:21Dominic Yeah. So this is what I'd recommend. First of all, you can develop your app or service locally. Go to skills.internetcomputer.com. Sorry, in-internet com-- skills.internetcomputer.org, and they'll give you like a, a prompt to paste into your coding assistant, whether that's, Claude Code, Codex, Perplexio or whatever.

35:42Dominic And that will give it the root skill. I, I would tell it to code the backend in Motoko, because this has a bunch of really wonderful features that are designed for people building p-predominantly with AI. So f- so it's in fact the first language designed specifically for AI agents rather than humans, and has a lot of very powerful built-for-purpose features.

36:06Dominic When you upgrade an app in production that was written in Motoko it, y- you have to specify, a, a trans- a transformation for memory if it's necessary. And the framework that runs that app, the Motoko language framework, can detect if your… if the AI has made a mistake and some it might result in a lossy update, like some data could be lost during the update.

36:30Dominic And this is actually a real problem with traditional tech. When you update apps and services, there's always this fear that it's gonna go wrong and gonna create a security hole, or it's gonna stop working or that you make a mistake and somehow you, the migration of data has failed and you've lost some data.

36:46Dominic Motoko guarantees that can't happen. So I choose Motoko as the backend language and just see what you build. I use, Claude Code and you get big in-incredible subsidies with Claude Code Max subscription. S-see what you can build. And then you can deploy it to an internet computer. Historically, that's been a bit more involved 'cause you need these tokens and things like that.

37:06Dominic But like what I can tell you is there are changes there coming as well. So you'll be able to just pay with a credit card into a thing that pops up while you're using the coding assistant. You can deploy right now and, many tens of thousands of people have done that. But it's a bit more involved. But then you've got cloud engines coming.

37:22Dominic So in the next week or two get on the list, get yourself a, a coupon, and you'll be able to create yourself a cl-cloud engine for free. And once you've logged into your cloud engine console, again, it'll just give you a prompt that you put into your coding assistant, and again, you'll be able to start building on it.

37:37Dominic And phase one, I think of the rollout is it's just agentic infrastructure. So it's, conceived as the ideal infrastructure for agents to build on. You don't need a security team, you don't need a system, systems admin team. It guarantees if you're using Motoko that you-- when you're upgrading your apps in production, you can't lose data.

37:54Dominic I think the second phase of the rollout Umti Labs will make OpenSaaS available. So for example, like Definity is running on OpenSaaS, and w-we're gonna push that out in alpha soon, and it's an amazing change from traditional SaaS. If you're a big org like Definity, I think we use like Bamboo HR for HR Oracle NetSuite for accounts.

38:15Dominic Or we did anyway. And oh yeah, Lin- no, we don't. We're not using Linear. We're using Jira for project management. All the usual

38:21Brandon many, all, all the SaaS stuff that

38:22Dominic All the, the siloed stuff that costs loads of money. OpenSaaS obviously the, it's one suite that's constantly expanding to cover everything. But it's all integrated, but most importantly, it's AI aware. So there's AI on the in- the inside. And so essentially, subject to the permissions of the user, AI can see the data graph in memory, and it can access the actual APIs inside the code.

38:45Dominic And this means you can ask it the most extraordinary things. So a simple example would be rather than going into Open HR to manage PTO requests, you would say to the AI assistant "Hey please look at all the PTO requests for customer success staff." Approve those that aren't intersecting. Make sure that we always have coverage in all the different time zones we have customers.

39:11Dominic Where there are overlapping PTO requests give preference to our best employees and give me the emails that ask people to choose different dates where necessary. And it'll go away and do this incredibly complex optimization problem for you that would be almost impossible to perform as a human. And, a few minutes later, it'll come back and there are the emails, and you just press approve.

39:33Dominic Done. And this is the extraordinary thing. If you can get all of your enterprise's data into Open SaaS, it can be as good at running a company as it is at writing code 'cause it's got all the context. So that, that's something we're very excited about. And by the way, you can also create… or once Cloud Engines are out or even now, you can already create AI where with Matoko, there's a skill and, the AI can find out how to to do this kind of stuff.

40:01Dominic Yeah.

40:02Brandon All right. I love it. I love the idea of not having to interact with so many SaaS systems. That was honestly that's the the dream of all of us. I don't want any more emails about PTO

40:10Dominic more SaaS systems. And even better than that you'll end up spending in as much as you have to use SaaS, like if you're an organization bigger, a big enough organization, you're gonna need something to manage your HR and you're gonna, and your accounts. But you, instead of spending, all your time like stuck inside the usual user experience screens, you'll instead spend probably 80% of your time just chatting to AI and telling it what to do for you.

40:32Dominic It's, it's-- Once you've tried it you never wanna go back.

40:35Brandon it sounds a lot like coding. Like once you stop writing code and start, telling it to like build stuff, it just, it's a different world

40:40Dominic and m- modern AI is just getting so wonderfully powerful and good at these things, yeah

40:45Brandon All right listen, we have a lot of fun stuff to do. I've learned quite a bit in this conversation. I'm going… You mentioned a bunch of links. I'll get all the links. I'll make sure they're in the show notes and on the website so everyone can find those as well. But if someone has heard you and they wanna get in touch with you personally, any specific way, any good way for them to reach out to you?

41:01Dominic Maybe on social media, something like that. I'm a difficult person to reach. My inbox to be honest, email has just got 50,000 unread emails in it. But there's, there are ways, like you can get… You find other people here, you can get and g- get a hold of me, and we're always open to talk to people.

41:15Brandon That's fantastic. All right. Dominic, thanks a lot for coming on the show today. I really enjoyed the conversation

41:20Dominic Thank you, Brandon. It's been great

41:22Brandon And for everyone listening, we really appreciate it. Of course, if you're watching on YouTube or listening on your podcast player, it always helps if you subscribe or give us a quick we appreciate that. And with that, we'll talk to you next time Thanks for listening. Check us out at theenterpriseaishow.com for past shows, newsletters, and all things enterprise AI