Latest / Agentic Conversations (formally mlops.community) / The Rise of Sovereign AI and Global AI Innovation in a World of US Protectionism // Frank Meehan // MLOps Podcast #331

Transcript

The Rise of Sovereign AI and Global AI Innovation in a World of US Protectionism // Frank Meehan // MLOps Podcast #331 · Agentic Conversations (formally mlops.community)

• 0:00
  We just saw the the power outage in Spain, right?
• 0:03
  I mean, what happens if AI is in the banking sector and it just
• 0:06
  suddenly, you know, you know, collapses or takes over or all
• 0:15
  right. Hi, I'm Franklin.
• 0:16
  I'm Co founder at Frontier 1A leading sovereign AI specialist
• 0:22
  working to bring AI to sovereign in frontier markets around the
• 0:28
  world, Asia, Latin America, Middle East, Africa and also
• 0:34
  working very closely with Nebius, which is one of the if
• 0:40
  not the leading AI cloud infrastructure providers in the
• 0:44
  world. And I'm very excited to be
• 0:47
  working with both of those companies and opportunities.
• 0:51
  And and secondly, how do I take my coffee?
• 0:58
  I take my coffee. I have 3 cups in front of me.
• 1:01
  All of them are Americanos with oat milk usually in them, but
• 1:08
  it's occasionally A cappuccino or a flat white may sneak in to
• 1:14
  that as well. It's a fairly flexible option.
• 1:19
  I wanted to talk about sovereign AI, so I brought you here.
• 1:21
  I think you know a thing or two about it.
• 1:23
  You've been having a lot of conversations.
• 1:26
  Let's just get into it. Why sovereign AI?
• 1:31
  Sure. Well, I think it's, let's start
• 1:34
  with the market itself in AI, right.
• 1:36
  So I've had a, you know, a long history in AI right back from
• 1:41
  being on the board of Siri in 2009 when no one in Silicon
• 1:45
  Valley cared less about AII mean really didn't.
• 1:48
  And it was so interesting. And then, you know, when I, I
• 1:51
  was working for a, a major family office out of Hong Kong
• 1:54
  Horizon Ventures and, and we invested early into deep mind as
• 1:58
  well, which was 2000 and 2011, I think, when we did that so
• 2:02
  extremely early. And you know, it was like, how
• 2:06
  do you, how do you pick those opportunities?
• 2:08
  How do you see those trends happening?
• 2:09
  And I think over the years, just working globally in so many
• 2:12
  different markets, I've been in Hong Kong and Singapore and
• 2:15
  Silicon Valley and London everywhere, you get a feel for
• 2:17
  things and you get a feel for how things are happening, the
• 2:20
  way trends are moving. And, and I think the way the
• 2:24
  market is, is super interesting. So you have the US and the US is
• 2:29
  from an AI perspective, it's, it's all about the hyperscalers,
• 2:33
  right? So you have the big guys who are
• 2:35
  providing the GPU cloud infrastructure like Nabius, for
• 2:41
  instance, and Corwave. And then of course, they're
• 2:44
  providing that to their major customers are open AI, you know,
• 2:49
  the hyperscalers, right? And also very, very heavily
• 2:52
  funded AI startups. Right.
• 2:55
  So in the US, customers of all these massive AI data centres
• 3:00
  and, and, and are effectively the big hyperscalers, right?
• 3:04
  Because that's where the money is.
• 3:05
  Like there's a trillion dollars that's gone into AI in the US In
• 3:08
  terms of investments outside of the US, it is a totally
• 3:11
  different picture, right? In Europe, it's like, how much
• 3:14
  is that, what, $10 billion I think last year was invested
• 3:17
  into AI startups. It's a fraction of the spend,
• 3:20
  right? So, so where is the AI compute?
• 3:25
  Where are the AI opportunities? And you know, I've been looking
• 3:28
  at this for, you know, a couple of years and really NVIDIA, to
• 3:32
  be honest, picked this early. They saw that Jensen has been
• 3:36
  extremely, extremely keen on the sovereign AI element because he
• 3:41
  saw that effectively, if I'm a government and I want to deploy
• 3:46
  AI into a model, right, I want to put my, my data, my defense
• 3:51
  data, my finance data, my agriculture data, my health data
• 3:54
  into a model. That data cannot go to the US,
• 3:57
  It cannot go to China, it cannot go to my neighbor.
• 4:00
  It has to be in my country. It has to be localized.
• 4:04
  And no one was prepared for this.
• 4:06
  AWS is not prepared to be a local player.
• 4:09
  AWS is a regional player. I have a hub.
• 4:12
  I use the AWS. It's either I use the German hub
• 4:15
  or I use the, it's, you know, the, the US, the US or the Irish
• 4:20
  hub. But it's nothing has been this
• 4:21
  localized ever. And suddenly localization and
• 4:24
  suddenly as well the, the recent developments in the US around
• 4:30
  from a political viewpoint have forced countries to go.
• 4:34
  We cannot rely on the US. We can obviously must be wary of
• 4:39
  China. We must be wary of all these big
• 4:41
  players. So we have to look after
• 4:42
  ourselves and, and so, but what do we do?
• 4:46
  And that's where it's become extreme national priority to
• 4:51
  have AI implementations because I, I have to do it local, then I
• 4:55
  have to put my, my data in there and I have to make it secure.
• 4:58
  I have to have localized models, localized applications,
• 5:01
  etcetera. And so sovereign AI has gone
• 5:04
  from a small thing to suddenly be one of the biggest
• 5:07
  implementations, I think of AI from a rest of the world market
• 5:11
  outside of the US. Yeah, it feels like it is
• 5:13
  something though that is like you're steering a gigantic cargo
• 5:20
  ship in a way, because it takes so much to stand up these data
• 5:26
  centers. But I don't know, maybe you're
• 5:28
  you're the guy on the inside. How long term are we thinking
• 5:32
  here? And I know everybody's
• 5:33
  scrambling, but it's not like you can just turn a data center
• 5:36
  on from nothing. Well, yes and no.
• 5:40
  It depends on the size and depends on what it's going to
• 5:43
  do. So if I'm going to have, so if
• 5:47
  I'm like, you know, a market like France or something and I
• 5:50
  want to have the latest Blackwell chips from NVIDIA.
• 5:55
  So I'm going to need state-of-the-art, liquid cooled,
• 5:59
  state-of-the-art data center, right.
• 6:01
  Then that needs to be from the ground up.
• 6:03
  There are very few data centers, if any, in Europe who are ready
• 6:06
  for that. Just a few, obviously, Nabius
• 6:09
  being one of them, right, who are by far like probably leading
• 6:14
  in the ability to do that. Two can even get hold of those
• 6:17
  chips, right? So even if I'm going to build
• 6:20
  that state-of-the-art data center, can I get hold of them?
• 6:23
  And again, I'm going to need a partner, you know, like a Nabius
• 6:28
  or call either someone to do that who has NVIDIA backing.
• 6:31
  Very strong. And to just rack them and to
• 6:34
  figure out how to plug them all together.
• 6:37
  All the wires, where do they go? What did the box look like?
• 6:40
  That right and and the sophisticated sort of, you know,
• 6:44
  cloud management, Kubernetes, Docker type management that they
• 6:47
  have on top from managing all those different chips because
• 6:49
  they don't just have Blackwell, they go all the way down
• 6:51
  different types of chips or different usage cases.
• 6:54
  What's interesting though is in other markets, so let's say
• 6:57
  Indonesia. So Indonesia where like, OK,
• 7:00
  they were very smart. I think they worked with Nervius
• 7:02
  and they were like, OK, look, we don't need Blackwell chips.
• 7:06
  We're probably going to need the age 2 hundreds, H1 hundreds are
• 7:08
  fine, right? And what they did is they got
• 7:12
  they took a they so the government there said, OK, we're
• 7:16
  going to work with Hutch Indo, which is the major telecom
• 7:19
  provider. They're going to take one of
• 7:21
  their existing data centers and we're going to repurpose that.
• 7:25
  So we're going to upgrade that so it can take the chips that
• 7:28
  it's capable of taking. And NVIDIA did that work and
• 7:31
  we'll get it up and running. We'll put a local model on top
• 7:34
  applications in space. And it took them four months to
• 7:36
  put it up and running. Wow.
• 7:38
  Right. So I'm looking at an 18 to 2
• 7:40
  year period if I'm going to try a massive liquid cooled hundred
• 7:45
  200 MW facility. But if I'm just putting up a 5
• 7:48
  to 10 MW facility with like chips that are suitable for the
• 7:53
  needs that I have and and to be honest, like probably mostly
• 7:58
  inference, not so much the modelling.
• 8:01
  I was going to say that then. Yeah.
• 8:02
  I mean there's really different ways of doing things.
• 8:05
  So it's yes and no depends on the on the size and scale of
• 8:10
  what you're trying to do. What are different countries
• 8:14
  ways of approaching this and how do they differ?
• 8:18
  Yeah. So you know Indonesia and you
• 8:20
  know, the Philippines or you know, Morocco, etcetera, all
• 8:24
  these markets that we at Frontier one are working with.
• 8:28
  So we're we're you know, we're, we're basically what we call
• 8:30
  sovereign AI specialists, right. So we will work with governments
• 8:34
  on a local level to figure out what their needs are.
• 8:37
  And and it's it's very it's very local because you have to have,
• 8:41
  you'll have to have a local company.
• 8:42
  You'll be using local talent, you have to bring in the
• 8:45
  partners, but still they will have to go through a local
• 8:48
  company that is and, and very close in tune with the
• 8:52
  government and the corporates on the ground, right.
• 8:55
  So each, each market says, OK, what are we going to do?
• 8:59
  We're going to have a three, the biggest data center in the
• 9:02
  world, like the three gigawatts that India is trying to do.
• 9:05
  Are we going to do 1 GW like France?
• 9:08
  We're going to do 100 megawatts like some of the other markets
• 9:12
  in in Europe. And how are we going to do that?
• 9:15
  Well, we're going to need to establish a local company.
• 9:18
  And let's say the government's going to go, someone's going to
• 9:20
  have to say, where's that customer coming from?
• 9:22
  So the government says I will take up 25% of the compute,
• 9:26
  right? The other 75 or 30 or 80% or
• 9:29
  whatever has to come from somewhere else, right?
• 9:32
  And that's going to be government.
• 9:34
  That's going to be corporates and research facilities,
• 9:36
  etcetera. And then someone's going to have
• 9:38
  to figure out the financing. So either the government says
• 9:40
  I'm going to finance all of it or I'm going to find as part of
• 9:43
  it and a partner needs to come in and finance the rest of it.
• 9:46
  So for, for the markets that, you know, I, you know, that we
• 9:50
  work with at Frontier 1, like as we said, across Asia and Middle
• 9:53
  East and Latin America, all of these markets, it's, it's maybe
• 9:56
  a 5/10/10 mega mod facility, right?
• 9:58
  It's 4000 to 67000 GPUs. It's a hundred $200 million, you
• 10:06
  know, that type of thing is, is very doable in those markets,
• 10:09
  right? So if you, you put that together
• 10:12
  for them, you find who the right partner is.
• 10:14
  You know, we, we really like working with Nabius, for
• 10:17
  instance, because they're very flexible and and international,
• 10:21
  but like we find that right partner and who's the and then
• 10:26
  there's the other elements, right.
• 10:27
  So you know, you introduced us to those great guys in Italy who
• 10:33
  were looking at sovereign AI and rails, right?
• 10:35
  What's the pre Prem AI, Prem AI, right?
• 10:38
  Great team, right? Spoke to them earlier today
• 10:41
  really smart, right and they've cottoned on to yeah, how do we
• 10:45
  do this for these markets at the smart way in the lower in the
• 10:49
  lower level. I spoke to another team earlier
• 10:52
  today who were building edge solutions.
• 10:57
  So so like competing with Nabius, but for remote market
• 11:01
  and for remote locations, right, where power is low sun is hot,
• 11:06
  you know, dust, you know, edge locations, right?
• 11:10
  So you got to think, OK, where is this thing going to be?
• 11:13
  What's the market? Who are the user cases?
• 11:16
  Who are the customers? And for governments, they've got
• 11:19
  like, it's really complicated to try and figure all that out.
• 11:21
  And so and you, you go, you're not going to go to McKinsey and
• 11:25
  EY, they're not going to have a clue either.
• 11:26
  So you know, you've got to go to the specialists and you know, so
• 11:30
  people like us use like no software, no infrastructure,
• 11:34
  know like how to do the financing.
• 11:36
  You know that that type of approach is I think the right
• 11:39
  way to go about it. But we will see.
• 11:42
  Yeah, it feels like the different governments are
• 11:46
  interested in AI because they have to be.
• 11:50
  We kind of established why it's not all right to just now all of
• 11:55
  a sudden keep doing things with the US as such a close partner.
• 12:02
  And folks are thinking, well, if we're going to be building out
• 12:08
  our AI capabilities, we should invest in our own country and
• 12:15
  our own AI capabilities as opposed to shipping off these
• 12:19
  different workloads or inference to a faraway land.
• 12:23
  And even a lot of these examples that you're talking about, if
• 12:26
  it's inference, it's going to be much better the closer it is to
• 12:30
  you. So it makes a ton of sense that
• 12:33
  you would want to put up different data centers in
• 12:37
  different countries so that things can hopefully be faster
• 12:42
  and more enjoyable for the end user.
• 12:46
  I'm wondering about how you feel governments will change when
• 12:54
  they now bring online all of this AI capability.
• 12:59
  That's really interesting. I mean, regulation and data
• 13:04
  protection and sovereignty and everything comes into that
• 13:07
  question for sure. The question is, look, I think
• 13:12
  governments know that they could be more efficient deploying AI.
• 13:18
  They're not sure how often, but they know they can, right?
• 13:21
  And so let's take a very simple example of I want to put AI
• 13:25
  models or whatever in or applications into my health
• 13:30
  system and make it more efficient to do that.
• 13:34
  I got to be sure that that health data can go into a local
• 13:37
  model that's super secure. But at the end of the day, I
• 13:42
  want to be able to provide that access to that data to companies
• 13:46
  that are going to be in startups.
• 13:48
  They're going to be building stuff on my to make my health
• 13:53
  system more efficient. So for instance, you know,
• 13:55
  there's probably a local player that's working on digitizing
• 13:58
  health records, a regional player.
• 14:00
  So that player needs to have access to that data, but it
• 14:05
  needs to be secure, right? And it needs to be.
• 14:09
  And the government needs to feel that I can provide that access
• 14:13
  on a secure basis. Now, it's not like I'm going to
• 14:16
  do everything that way. I'm going to be obviously using
• 14:18
  open AI and I'm doing others. But I need to, I need to be very
• 14:23
  sure that that data can be locked down and secure.
• 14:26
  And I also need to have a local vector database, a local
• 14:29
  cybersecurity platform, a local local model, right, that
• 14:35
  provides all that. I think governments trying to
• 14:38
  figure all that out. But effectively what they really
• 14:41
  at the very simplistic level want to do is can I provide, can
• 14:44
  I put AI securely into my, you know, into government and
• 14:49
  provide access to companies that want to do things in health or
• 14:55
  agriculture or whatever it is securely and don't feel that my
• 14:59
  data is going to leak. Because if it does leak, that's
• 15:01
  going to be on the front page, the news, and it's not going to
• 15:04
  be good for us. It's funny that you mentioned,
• 15:08
  hey, governments want to do this but don't necessarily know how,
• 15:12
  and not having sovereign AI to me feels like 100% a blocker on
• 15:20
  them iterating and trying things.
• 15:23
  So they can't figure out how until they start iterating.
• 15:26
  They try and put it in here, put it in there, realize maybe
• 15:30
  that's not really where the value is, so let's try again.
• 15:33
  You can't do any of that experimentation until you cover
• 15:37
  your bases, and until you get that sovereign AI up and
• 15:41
  running, you're kind of Sol. Totally, and I think that's that
• 15:45
  that's absolutely right. It's you know, they know that
• 15:51
  that's a star package and and and let's just look at research,
• 15:54
  right? Universities and everything that
• 15:56
  need to do research again, that's a lot of government
• 16:00
  sponsored research. And again, you want to keep that
• 16:02
  secure, right? I need to be sure that that data
• 16:06
  hasn't gone into someone else's model in say Silicon Valley
• 16:10
  somewhere, right? And I want to you know, have
• 16:14
  that aspect to it. I don't mean a lot of people
• 16:17
  have seen sovereign AI is like government control and it's not
• 16:20
  it's it's government. It's it's trying to figure out
• 16:23
  how do you like look after the data in the right way that is
• 16:28
  very important to you. I mean, you know, your citizens
• 16:31
  health records are very, very important to you.
• 16:35
  You cannot let that data go to another nation and neither your
• 16:41
  defense like data, right? You're that's also extremely
• 16:45
  important to you. You cannot let that leak say
• 16:48
  it's, it's tricky, but it it needs to be done.
• 16:51
  But then at the same time, you can't do everything.
• 16:53
  So you're going to have these great startups and companies
• 16:56
  that probably local or regional that are going to want to work
• 16:59
  on the on the system. And you have to put that in
• 17:02
  place for them. You have to have the rails for
• 17:04
  them to do that, which is what pre Mii, yeah, which is what
• 17:11
  those guys are doing, which is really cool.
• 17:13
  And so I think, you know, it's our job at Frontier 1 is to
• 17:18
  bring all that together and be the master builder right in
• 17:22
  these markets and help them understand all those aspects and
• 17:26
  also always keep on top of it and make sure best of breed is
• 17:29
  coming through the door. But yeah.
• 17:33
  Yeah, recognizing if you need a Blackwell or if you're good with
• 17:36
  just an A-100 fleet of A1 hundreds is already so valuable.
• 17:40
  I can only imagine how that type of expertise can be in a way put
• 17:51
  to good work. And then on top of it, all the
• 17:54
  other stuff that you bring is is incredible.
• 17:57
  I wanted to just zoom in and almost like push back on the
• 18:03
  medical data one for a second. The defense one, I think
• 18:08
  everyone 100% agrees you want to be doing your defense data in
• 18:13
  your country and that is without a doubt number one priority.
• 18:20
  The medical one, I often wonder if it's not better to get the
• 18:26
  more medical data the better because you're going to find
• 18:28
  these random very long tail type of diseases and you get to see,
• 18:33
  well, where else is that happening?
• 18:36
  And it could be highly valuable for researchers.
• 18:40
  I it's also funny that you mentioned this, because I was
• 18:43
  just listening to the acquired podcast on my chart, which is a
• 18:48
  company, I think it's epic the, the company and they created
• 18:52
  this thing called the Cosmos and it anonymizes hundreds of
• 18:58
  thousands, if not millions of data, patient data records.
• 19:03
  And then researchers are able to go and look at that and even
• 19:06
  doctors are able to go and look at that.
• 19:08
  So in that regard, I don't know. Yeah.
• 19:14
  I mean, I will go with that say, look, it's it's two aspects to
• 19:18
  it, right? Anonymized data for research
• 19:20
  purposes, 100%, no question. But the question is when, you
• 19:24
  know, if I'm going down to the NHS here and they're pulling up
• 19:28
  my record and they're trying to figure out like what's that's an
• 19:32
  internal NHS system which is accessing my record.
• 19:37
  And that absolutely needs AI because it's a disaster of a
• 19:40
  system clearly, right? Because nobody, I mean, they sit
• 19:44
  there and go, you know, and it's like, whatever, right?
• 19:47
  That's a different problem, right?
• 19:50
  So, so there's two parts to that problem.
• 19:52
  One is the research on anonymized data.
• 19:54
  No question, I need as much data as I can.
• 19:56
  But of course the other question is the localized NHS equivalent
• 20:01
  is accessing highly sensitive private data.
• 20:04
  And. That's a different problem.
• 20:07
  So there are two problems to that.
• 20:09
  I see what you're saying. So it's more of this when the
• 20:14
  inference side of things or the execution of the workloads are
• 20:19
  happening, you don't want them getting even if it is fully
• 20:24
  anonymized. It's better to keep that close
• 20:27
  to home. Yeah, and especially if I'm
• 20:30
  going to be modernizing, let's say I'm modernizing the NHS here
• 20:33
  in the UK all. Right.
• 20:34
  God can only hope, huh? Yeah, but exactly right.
• 20:37
  But that or anywhere else, right, Those systems are old,
• 20:40
  right? They're 1020 years old in all of
• 20:42
  these markets, right, at least. And they radically need updating
• 20:47
  and that's going to be a radically different way of doing
• 20:50
  things. And they're going to be doing
• 20:52
  modelling on that huge amount of data that they have that is not
• 20:55
  anonymized for that purpose because they need to, they need
• 20:58
  to also use that internally. So it's a very different
• 21:02
  problem, you know, so that that's the one I'm, I'm going to
• 21:06
  be referring to more here. From a sovereign AI perspective,
• 21:08
  anonymizing data from a research perspective, that's a different
• 21:12
  problem. And and yeah, I mean, give the
• 21:15
  research as much data as they can for sure.
• 21:18
  But what do I do? So for instance, can the AI,
• 21:23
  let's say I go in and get a body scan here at the NHS, right?
• 21:28
  And they use that data and they model that data against others
• 21:31
  and they come back and they tell me X about me, right?
• 21:36
  That is very sensitive data that needs to be protected.
• 21:39
  So, but the people that are going to be developing that
• 21:43
  application need to access my data.
• 21:47
  So you know, that's that's a different problem in a way.
• 21:52
  Yeah, also you have the epic company in the US, but every
• 21:59
  country has their own epic or a few different epics that want to
• 22:04
  be creating these ecosystems so that folks can be pushing
• 22:09
  forward in the progress area. And I think I fall victim of
• 22:15
  this a little bit because I am American of thinking that there
• 22:19
  is the US and then everybody else.
• 22:22
  And that's kind of we don't think about that.
• 22:24
  Obviously living outside of the US for so long.
• 22:28
  I recognize there's this great big world with a whole lot of
• 22:32
  people out there and it needs these type of things to these
• 22:37
  initiatives and all this stuff. And when I was in the when I was
• 22:41
  in the Zurich Jenny I conference a few weeks ago, one thing that
• 22:45
  was top of mind for them in a few different talks that I saw
• 22:50
  was, hey, all this AI stuff is happening.
• 22:53
  How is Switzerland being part of it?
• 22:57
  And of course, Switzerland is traditionally very neutral.
• 23:01
  So for them, sovereign AI felt like a natural next step or a
• 23:06
  natural path to take. And it it also helps that they
• 23:11
  can say, yeah, we're going to go invest in this project.
• 23:16
  I wonder if you see different come countries or scenarios
• 23:22
  where maybe they want to invest in it but like they don't
• 23:27
  understand or is there just budget right now everyone
• 23:30
  understands that this is as important as possible.
• 23:33
  I think it's a great question. I think taking a step back
• 23:37
  again, like what's what Trump has done has is, is end
• 23:43
  globalization, right? And not just in trade, but in
• 23:47
  how things are built, right. Traditionally a company would
• 23:50
  have been a country doing completely fine using AUS
• 23:53
  company to do whatever they did, right?
• 23:56
  No more, right? He has turned everything to
• 23:59
  localization instead and and in some ways incredibly cut the
• 24:06
  foreign business development opportunities for a lot of U.S.
• 24:09
  companies, which is incredible, right?
• 24:12
  And so, but on the other side, that means that a huge
• 24:16
  opportunities opened up for local regional companies around
• 24:20
  the world. It's, it's, it's just
• 24:22
  transformed the landscape incredibly, right?
• 24:25
  So now Europe is going to go, I need as few U.S. companies in my
• 24:32
  sovereign AI or banking facility as possible, right?
• 24:37
  And so I need local and, and the local suddenly these local like
• 24:41
  it's, it's so interesting. I mean, you know, suddenly
• 24:44
  they're going to be I need local local companies for this and I
• 24:48
  had local companies for that and, and all elements of the
• 24:51
  stack. There are some banks in Germany
• 24:53
  have said like I cannot have AUS company in the stack, right?
• 24:57
  I can't even have GitHub. So like that's, so that's huge,
• 25:03
  tremendous opportunity for companies around the world to
• 25:06
  suddenly become taking advantage of that.
• 25:10
  And that will start primarily in the finance and government
• 25:13
  sectors because both the finance was like finance defence and
• 25:16
  government suddenly go, we cannot, we can't use U.S.
• 25:20
  companies. We may not even be able to use
• 25:23
  the UK company. We have to use a European
• 25:25
  company. So, and, and UK is like, we need
• 25:30
  to do this and Saudi is, well, we're going to need to do Saudi.
• 25:34
  I mean, it's just tremendously opening up the opportunity
• 25:37
  globally for local talent and local development of core AI
• 25:43
  functions just from the geopolitics of everything that's
• 25:46
  happened. I mean, Trump has literally
• 25:49
  suddenly woken up the world who was like pretty much reliant on
• 25:53
  U.S. companies and US software to go, wow, well, wow, what do
• 26:00
  we do? So I mean, it's kind of at that
• 26:03
  level now, right? Everyone is a bit just at that
• 26:05
  conference that you were at, Aaron's like, well, what do we
• 26:08
  do? And so, yeah, I mean, that's
• 26:11
  where we come in and go, OK, let's help you through this from
• 26:14
  a local level. It's going to be tremendous
• 26:17
  opportunities. I think AI is truly going
• 26:19
  global, thanks in large part to Trump.
• 26:23
  So. Yeah, he's great for your
• 26:24
  business. You've got to send them a thank
• 26:27
  you card or something. 100% I mean, there are a lot of people
• 26:30
  that are in the, you know, technology space who are going
• 26:33
  to be doing that now because yeah, it's the same thing.
• 26:37
  Like if I'm, you know, let's say I'm like, it just replies to so
• 26:41
  many things. If I'm a again, if I'm a country
• 26:44
  and I'm, let's say there's this kind of rumour, you know, that
• 26:48
  the F30 fives that if I buy an F35, the US can switch them off,
• 26:52
  right? I mean, would I, I mean, that's
• 26:55
  super interesting right now, you know?
• 26:58
  Now you got to look for a new provider.
• 27:00
  He's got to Remus, you know, whatever.
• 27:02
  But it's entirely possible that the US could switch an F35 off,
• 27:08
  right? So I've just spent a billion
• 27:10
  dollars buying an F35 and you can switch it off with a kill
• 27:12
  switch. I mean, you know, like, that's
• 27:15
  really interesting, right? And he has, as we just saw from
• 27:20
  the Canadian election, like completely transformed
• 27:23
  countries. Canada, Canada is now more
• 27:25
  nationalistic and more local than ever.
• 27:28
  And that is happening in many markets and it's going to open
• 27:33
  up a lot of opportunities for AI developers and stuff all around
• 27:36
  the world. Yeah, the First off, let's just
• 27:39
  have a moment of silence for those engineers that have to
• 27:42
  work at the banks that don't allow GitHub because poor folks
• 27:47
  out there, you know they are going to be going through hell
• 27:50
  and back. But the other thing.
• 27:52
  Just think about what they've been having to use instead,
• 27:54
  yeah? That's what I'm thinking about.
• 27:57
  It's like, oh man, that's going to be rough.
• 27:59
  But the the other piece was around this idea of we've seen
• 28:05
  the US weaponize the dollar and like you're saying, there's this
• 28:13
  jump or there's a step where they can weaponize the weapons
• 28:18
  that they've sold to other countries.
• 28:19
  So if you get on their bad side and they say potentially they
• 28:23
  just turn off their jets that they sold to you.
• 28:25
  Whether or not that's true or not, whatever, it doesn't matter
• 28:28
  because the thought has been planted into different
• 28:31
  governments and politicians minds that they are making an
• 28:36
  effort to move away from the US as the supplier.
• 28:41
  But then if you extrapolate that forward, you can think about it
• 28:45
  in this frame of, well, what's going to stop them from
• 28:49
  weaponizing AI? And maybe today with the use
• 28:54
  cases that we have, it isn't as strong of an argument.
• 29:00
  But potentially in five years or sooner or later, maybe it's 10
• 29:04
  years, that could be a whole different environment that we
• 29:09
  are living in. If we are as reliant on EU s s
• 29:14
  AI ecosystem as we are on the dollar.
• 29:19
  Then if for some reason you get on Trump 3 or whoever the next
• 29:25
  president is on their bad side and they say, well, we're just
• 29:28
  going to turn off your AI, that is another way that it can be
• 29:32
  weaponized. So I fully see the other side of
• 29:35
  this story where it's like, yeah, we got to make sure that
• 29:38
  if AI is as transformational of a technology as most folks are
• 29:45
  saying it's going to be, whether or not it is I it's debatable
• 29:49
  100% when I think there's a lot of early signals that it's going
• 29:53
  to be. But even if it isn't, it's
• 29:58
  better to have it and not need it than to need it and not have
• 30:01
  it per SE. So you don't want to the US to
• 30:04
  have that capability of just like being able to shut off your
• 30:07
  AI. Yeah, Look, I mean, I think the
• 30:09
  US at the end of the day is, is just hyper focused on itself
• 30:12
  right now. And and that's fine, right?
• 30:16
  I mean, that's that's really fine.
• 30:18
  I do think that the Trump administration will, will
• 30:20
  gradually, you know, start to and is starting to realize that,
• 30:25
  you know, maybe this is, you know, things will change, right?
• 30:29
  So I do think it's just hyper focused on itself and just
• 30:32
  making sure that the US has the right resources and the right
• 30:35
  chips and everything that are available to it.
• 30:37
  And, and like, it's going to go like, we need the best chips.
• 30:40
  We need a lot of them and we need to, we need to conserve
• 30:43
  them, right? And, and that's the US is right
• 30:45
  completely. I think that and that's like a
• 30:50
  lot of markets are doing that, right?
• 30:51
  China does exactly the same thing.
• 30:53
  So the US is allowed to do that as well.
• 30:56
  I think at the end of the day, you, you, you break the US out
• 31:01
  from the rest of the world. And that as we said, like
• 31:04
  globalization for whatever reason is, is coming a bit to an
• 31:08
  end in a lot of ways. And localization is coming back
• 31:11
  again. And that means that by having a
• 31:15
  local, especially for something as important as AI, having a
• 31:18
  local deployment that I can rely on is important, right?
• 31:23
  And that I, you're absolutely right, you know, can it, you
• 31:26
  know, something maybe they do switch it off, right?
• 31:28
  And that that could have a huge impact because I have AI
• 31:30
  throughout my infrastructure, right?
• 31:32
  I mean, we just saw the, the power outage in Spain, right?
• 31:36
  I mean, what happens if AI is in the banking sector and it just
• 31:39
  suddenly, you know, you know, collapses or takes over or, or
• 31:44
  you know, so. That's why I saw stuff that
• 31:47
  folks were saying they couldn't get cash out because there was
• 31:51
  no ATMs there. They couldn't pay for anything
• 31:54
  because all of the credit card machines weren't working.
• 31:57
  And so effectively it was like, well, I guess I'm not going to
• 32:00
  do anything for a few hours. Yeah, that's wild, right?
• 32:03
  I mean, now that's an energy problem, which, you know, comes
• 32:08
  back online, but again, what happens if I've got AI, you
• 32:10
  know, heavily into my energy systems and everything?
• 32:12
  So it really has to be protected.
• 32:14
  And I, I, I do think that yes, you can make your own mistake
• 32:18
  nationally and you can fix it. But if you've got another actor
• 32:25
  considerably embedded into your AI local infrastructure, you
• 32:30
  need to be careful of that, right?
• 32:32
  And that, you know, the US isn't going to allow China to run
• 32:35
  their AI infrastructure and neither should anyone allow
• 32:38
  anyone else to write. I mean, like that's the
• 32:40
  localization of it when it comes to the, the government side,
• 32:43
  which is so important. I think especially if you, you
• 32:45
  look at, you know, energy and finance and where we could be in
• 32:47
  five to 10 years where AI systems are AI agents.
• 32:52
  And, you know, potentially, you know, general intelligence is,
• 32:57
  is, is running these things and we don't even have much control
• 33:00
  over them. And we're all gone to Skynet,
• 33:03
  which is not doesn't seem that far away now.
• 33:07
  You know, there is that potential of, you know, that
• 33:09
  that famous scene in T2, right when you know, the guys are
• 33:13
  switching on Skynet without having any idea what it can
• 33:16
  possibly do. That I think is going to be a
• 33:18
  lot of governments. And so like you've got to be
• 33:23
  able to how you're going to implement that and control that
• 33:27
  at least in a national level is important.
• 33:29
  Yeah. What are some of the other key
• 33:32
  players in this? I mean, you mentioned how you
• 33:35
  have to have the government's body and you have to have
• 33:39
  financers. You almost want to be looking at
• 33:41
  the different tools that you can offer in these data centers too,
• 33:46
  because it's not just the straight hardware, you've got
• 33:48
  the software. Is there anybody else that I I
• 33:51
  am missing? So I think, you know, you look
• 33:53
  at the stack. So I mean like, firstly, I've
• 33:55
  got to go. Let's say I'm implementing
• 33:58
  whether it's like I'm repurposing an existing data
• 34:00
  center, but let's say I'm building 1 from scratch.
• 34:02
  I've got to find the right power, I've got to find the
• 34:04
  right location. And, and, and also very
• 34:08
  importantly, before I build, I got to have the right customers.
• 34:11
  So I need to know before I'm committing a billion dollars or
• 34:14
  even $100 million of GPU's, who is the customer for this, right?
• 34:19
  So the government or someone has to say, I'm going to take the
• 34:22
  off take for that. Now in France, Macron twisted
• 34:25
  everybody's arms and said, OK, you all these corporates are
• 34:29
  going to come together and we're going to all contribute and and
• 34:32
  say basically put a billion dollars worth of compute off
• 34:36
  take spend on the table. I think that's very smart and
• 34:39
  that's a really good way of doing it.
• 34:40
  So I do think the other players are going to be big corporates
• 34:44
  in local markets. He will work in tandem with the
• 34:47
  government to do that. For instance, in in Argentina,
• 34:51
  you know, the government there has been really, really strong
• 34:55
  that they want to make Argentina a local AI centre and there and
• 34:58
  the corporates are being brought in to figure out the usage of
• 35:04
  it. Because at the end of the day,
• 35:05
  who's going to use it? How's it going to go right.
• 35:08
  So do you think that's the big, another big player is the the
• 35:11
  corporates themselves and figuring out how they will
• 35:15
  interact with systems locally. And the other players are going
• 35:20
  to be local, local players on the ground who can build local
• 35:23
  applications. You know, great, a lot of
• 35:27
  people, I'm sure in the ML OPS community, you know it, it's
• 35:31
  really interesting. I mean, like, you know, you
• 35:33
  think where I think is it's so interesting for a lot of
• 35:36
  markets, instead of just being a little outsourced team of some
• 35:40
  big U.S. company, the the opportunity is to is to build
• 35:44
  local and build something and then potentially regionally and,
• 35:47
  and, and keep going. We've seen in the Middle East,
• 35:49
  there's a lots of great regional players who do things in the
• 35:54
  Middle East, you know, because of different requirements, not
• 35:59
  least language, but you know, so how does that, how does that
• 36:03
  come together? It's really interesting.
• 36:05
  Do you feel like that leads to a world of more quote, UN quote,
• 36:10
  lifestyle businesses where you get smaller teams, they have
• 36:15
  smaller markets, but it's less venture backed or less fundable
• 36:21
  because they're not going to go and conquer the world and
• 36:24
  there's not that infinite Tam. It's a great question.
• 36:27
  I mean, we've obviously been talking about that opportunity
• 36:30
  and I think it, it's certainly enable say the tools that are
• 36:35
  available for everybody. Now I, you know, we can, we can
• 36:38
  spin things up faster with smaller teams and we can use
• 36:41
  tools to, to run things right. We can have agents running
• 36:44
  customer service and we don't need to pay $10,000 to some
• 36:48
  team, some group of people who are going to do customer
• 36:51
  outreach for us. We could probably use agents
• 36:53
  much more effectively now. So I think people can spin
• 36:59
  things up and prototype things much, much faster.
• 37:02
  They can come together in little teams to do things locally much
• 37:06
  faster and solve local problems. And those local problems can go
• 37:10
  regional, right? I mean, I'm a partner in a, in
• 37:13
  a, in a great fund that is out of Oman and the UAE.
• 37:17
  And a couple of those companies, even on somewhere as as small
• 37:23
  and as tight as Oman have gone regional just because they
• 37:26
  solved a particular problem that just works regionally, right?
• 37:31
  And gradually scale it out and now they can do it faster.
• 37:34
  And, and one of those is in in the health sector.
• 37:36
  And so, yeah, I really think that's going to be the case.
• 37:42
  I think it's a change, right. You know, instead of, as we
• 37:45
  talked about beforehand, like, you know, the, the model, the
• 37:50
  old venture model of giving $5,000,000 or $10 million to a
• 37:55
  company and saying, hey, let, you know, you know, nine months,
• 37:59
  12 months, produce something. And that's an MVPS are great.
• 38:03
  You know, just limited functionality, just get it out
• 38:05
  there. Like that doesn't work anymore.
• 38:07
  Like, you know, someone else can prototype something so fast with
• 38:11
  a lot of function and a lot of features and test it and roll it
• 38:14
  out faster. And I think that's really
• 38:18
  exciting. And I think that's where the
• 38:21
  power of like what the ML OPS community has is, is really
• 38:25
  powerful. Like that's a group of very
• 38:28
  great people who are in the AI and ML space and who are
• 38:32
  thinking about these things and prototyping these things.
• 38:34
  And they're not on LinkedIn. And if they are, they're not
• 38:37
  going to be really sharing their life experiences on LinkedIn,
• 38:40
  right? So they're doing that in a
• 38:42
  community like ML OPS. And so how does that come
• 38:45
  together and build cool new things and prototypes, I think
• 38:50
  is the the next kind of logical step for communities like like
• 38:54
  you guys. And I know that we had talked
• 38:57
  about too these traditional ways going back to that idea of, hey,
• 39:02
  I've got to go now and raise a bunch of money, go all in on my
• 39:07
  idea and spend the next nine months launching it, bringing it
• 39:11
  to market, etcetera, etcetera. That has been a bit turned on
• 39:15
  its head where you can in a way you can have your job at your
• 39:22
  big fancy company making all that cash, but still you can be
• 39:28
  moonlighting on this project, which it has the potential of
• 39:36
  like the iteration speed. I think is what is different
• 39:39
  because you've always been able to moonlight.
• 39:40
  You've always been able to have your you working at your big
• 39:43
  company. You are now moonlighting for
• 39:46
  something and then you bring it to market or you'd put it out
• 39:50
  there. But the ability to have that
• 39:52
  feedback loop be tightened is the coolest thing ever.
• 39:56
  And we're seeing it so much with a lot of these bottoms up
• 40:00
  companies like the Lovables and the Cursors where they have such
• 40:05
  a product LED growth motion that they can get to a gigantic scale
• 40:12
  with the small teams because they don't have any sales and
• 40:15
  marketing teams. Spot on I mean look at lovable,
• 40:20
  right, another Swedish, great Swedish, you know like Spotify,
• 40:24
  right. You know when Spotify started,
• 40:26
  you know, like in Sweden, right, you're going to take on the
• 40:28
  music market from Sweden and yes, they did completely right
• 40:33
  and lovable's done the same thing like you know, because of
• 40:36
  the ability to scale things faster and smarter and the hyper
• 40:41
  efficiency focus of Swedes is great.
• 40:44
  You know, I used to live in Sweden and it's amazing.
• 40:47
  I don't know how they do it. You know, six weeks off in the
• 40:50
  year went from sort of eight in the morning till four in the
• 40:53
  evening, four in the afternoon. But still like IKEA, yeah,
• 40:59
  mobile, Spotify, everything right.
• 41:00
  But I think the point is, is that you're absolutely right,
• 41:03
  right is that companies can can prototype and spin things
• 41:07
  faster. And I can do like, and you know,
• 41:10
  I, I would be, you know, if someone's at a big company and
• 41:14
  they were like a little cog in the dev team and it's really
• 41:18
  kind of bit boring. And they do the same thing over
• 41:21
  and over again with a little bit more next, you know, in a couple
• 41:24
  of months you're going to be looking at other things to do.
• 41:27
  But those other things to do now are infinitely more interesting
• 41:30
  and infinitely more scalable and faster to prototype and get out
• 41:33
  there. But also, I don't need to leave
• 41:35
  the day job until I know it's working and then suddenly it's
• 41:37
  working. And then, oh wow, OK, maybe I'm
• 41:39
  going to make a jump now. And and same as a VC, a VC is
• 41:44
  going to be like, I think VCs need to change their idea of
• 41:47
  what it is, right? I don't need to have 20-30
• 41:50
  people all be all doing something before I invest in
• 41:54
  them. I need to have like 5 very smart
• 41:57
  people who've launched something and I don't care how they got it
• 42:00
  out the door. I don't really care, right?
• 42:03
  I just know they got it out the door and it's really cool and
• 42:06
  it's scaly. And I'm going to put those.
• 42:07
  I'm going to invest into those five people.
• 42:09
  I don't care where they are or where they are, but I'm going to
• 42:11
  invest into them. That's a different story.
• 42:15
  And I think VCs have to change the way they're approaching the
• 42:18
  market. No, going back to the sovereign
• 42:21
  AI for a second, because one lingering question that I have
• 42:26
  is how you see regions cooperating.
• 42:32
  Specifically, we have the EU. It is very friendly in a way.
• 42:40
  But you have been mentioning, hey, France is doing this on
• 42:43
  their own. They're building their sovereign
• 42:46
  AI system very much for them. I know that Germany has talked
• 42:50
  about, actually, I think Germany's doing something cool
• 42:52
  where they're taking old factories and they're trying to
• 42:56
  spin up data centers with them in the old factories because of
• 43:00
  the old factories are now defunct.
• 43:02
  But they're like, well, the structure is still there.
• 43:04
  It's actually a big space. You know, it'd be perfect here.
• 43:06
  Whole lot of GPS. How do you see the
• 43:10
  interoperability of the countries?
• 43:12
  Maybe we'll start with Europe, but then go towards, like you
• 43:16
  were saying, in the Middle East and there's these companies that
• 43:19
  will spin up on top of the structure and it starts locally
• 43:23
  then goes regionally. Do you feel like these joint
• 43:26
  partnerships and the data centers are going to be only a
• 43:30
  specific country thing or is it going to be like that more
• 43:33
  regional thing? And how does that play?
• 43:36
  How does that work? Yeah, I think it's a great
• 43:37
  question. I mean, we don't know yet, but
• 43:40
  we can surmise that let's say France builds that $1 billion
• 43:45
  facility or even a $5 billion facility.
• 43:49
  I, I do not think that enough customers are in France for that
• 43:55
  type of facility. There's, I mean, what I mean is
• 43:57
  there's not enough usage in France for that facility.
• 44:01
  And I think France will then go, Hey, let's make this available
• 44:05
  to the rest of Europe and maybe just carve out, you know, some
• 44:10
  racks. And we're going to need to give
• 44:13
  the ability that like like some hyper localization on some of
• 44:17
  these racks. But yes, I do think so.
• 44:19
  And then is there AEU sovereign AI?
• 44:24
  Of course the EU is going to do it.
• 44:25
  I mean, that's that's, you know, the.
• 44:27
  Most regulated sovereign AI. 100% there's some guys in
• 44:30
  Brussels dreaming about this right now.
• 44:33
  And is that going to work? Well, yeah, let's.
• 44:38
  See another. Question.
• 44:39
  But they're certainly going to try and they're going to have
• 44:42
  some sort of European AI, but nobody really even knows what
• 44:47
  the definition of that is, you know, and, and the problem of
• 44:50
  course in the EU is it's so over regulated that anytime I try to
• 44:53
  launch something, I've got to check so many hoops that by the
• 44:56
  time I launch it, it's out of date.
• 44:58
  So I think there's that vast problem of regulation needs to
• 45:03
  get solved in the EU 1st as well.
• 45:08
  I think other markets have an opportunity to not be so hyper
• 45:11
  regulated and cooperate on a regional basis without the huge
• 45:18
  regulations that the EU throws in things that just stifle every
• 45:22
  innovation. You know, I think you've got a
• 45:25
  clear opinion about that. But yeah, so I think it's it's
• 45:29
  different, right? Markets are going to be
• 45:31
  different. Like how does Lat am, you know,
• 45:33
  collaborate Africa, right Asia. Like there will be collaboration
• 45:38
  for sure. Yeah, I like that idea of
• 45:43
  there's going to be access to man, so it's going to get sold,
• 45:47
  and it feels like potentially there's going to be that in
• 45:53
  these Middle Eastern countries too.
• 45:56
  And so it's interesting to think, wow, I imagine a world
• 46:01
  where even if you're not fully bought in, maybe you're not
• 46:06
  doing something so sensitive that you need the sovereign AI.
• 46:11
  It's just me, app developer in the EU, and I go on the streets
• 46:16
  looking for some GPU capacity. If there are a lot of different
• 46:22
  data centers that have been built up around me, that's just
• 46:25
  going to make my cost go way down.
• 46:27
  Yeah, that's right. Definitely.
• 46:31
  And I think that's, that's going to happen and everything, of
• 46:34
  course, at the end of the day comes down to power.
• 46:36
  So, you know, let's take something like Bhutan.
• 46:40
  Bhutan suddenly has become incredibly interesting because
• 46:43
  it's got 30 gigawatts of potential hydropower, of which
• 46:46
  it's only used about two, right? Is so much hydropower.
• 46:50
  So just Bhutan, if they put a little a major data centre in
• 46:54
  the South of Bhutan where they're building the new
• 46:56
  mindfulness city, we could have a mindfulness data centre down
• 47:00
  there. But I mean, the, the point is
• 47:03
  their power is so cheap. And so, I mean, they could be a
• 47:09
  regional player just from the, from the source of power they
• 47:13
  have, right? So, you know, El Salvador has
• 47:17
  is, is looking at building Bitcoin city in a, in a volcano,
• 47:21
  right, using geothermal right, hopefully extinct volcano.
• 47:24
  And now maybe Bitcoin City is, you know, I don't know.
• 47:29
  Well, today it's probably pretty good idea.
• 47:32
  Last week it probably wasn't. So who knows, right?
• 47:35
  But there's a little bit more, you know, surety around AI,
• 47:39
  right? So they're looking at thinking,
• 47:41
  OK, should we build a data center there, right?
• 47:44
  And could that be regional? Like, you know, there's a lot of
• 47:48
  thinking that goes into it, but I do think yes, there will be a
• 47:53
  degree of localization and regional support a lot of driven
• 47:57
  by which country has the lowest cost power.
• 48:01
  And it feels like from what I've understood it, you have to have
• 48:05
  the governments that sponsor this in a way, otherwise you're.
• 48:11
  To a certain degree, because no one's going to go like I'm not
• 48:14
  going to it. No, like I as a hyperscaler or
• 48:20
  whatever or a Black Rock or anything, I'm not going to put a
• 48:22
  billion dollars into building an AI data centre if I don't know
• 48:25
  who's going to be using it and who's going to pay for the
• 48:27
  usage. So yes, to a certain degree the
• 48:30
  government does need to put in like maybe they say I take 20 or
• 48:33
  30% of the off take and guarantee it, which means that
• 48:37
  yeah, OK, now I feel more comfortable about financing
• 48:40
  this. Is there anything else that you
• 48:44
  wanted to talk about that we haven't hit on?
• 48:46
  Man, I could talk about things for I, I could talk about
• 48:48
  anything really. I mean, right, you know,
• 48:50
  whatever really to be honest of issues.
• 48:52
  But I think today's topic has been exactly the key thing,
• 48:57
  which is really for me going back to it again, AI is truly
• 49:01
  going global. We saw that with DeepSeek how
• 49:05
  that came out of nowhere and you know, and in surprised the
• 49:10
  Chinese government as much as surprised anybody else, right.
• 49:12
  And where did that come from? It came from a hedge fund which
• 49:15
  was looking to do better modeling and decided to to build
• 49:18
  this thing and wow, right. So I think we're going to see a
• 49:22
  lot of things that are going to expand AI globally and the
• 49:29
  opportunities are going to be tremendous.
• 49:32
  And we're just seeing the start of that.
• 49:34
  You know, what I wanted to talk to you about for a second is
• 49:38
  take me back to the Siri days pre Apple acquisition.
• 49:44
  What and how did you even think like was give me the scenario
• 49:49
  because I think it's a fascinating story on how Siri
• 49:53
  got rocking and rolling. It was a bunch of researchers.
• 49:55
  Right. It's a bunch of researchers out
• 49:57
  of Sri International and it was really like, you know, I have to
• 50:03
  say, you know, my, my boss at Horizon Ventures, she's
• 50:07
  incredibly, incredibly smart. And she saw the opportunity.
• 50:13
  She was thinking about data and AI way before anybody else was.
• 50:16
  And it was together with, and this is out of Hong Kong, right?
• 50:19
  And, and she's so smart. And the others that were in that
• 50:24
  was Menlo and Morgenthaler and like kind of older VCs from the
• 50:29
  90s that knew deep tech, right. And we're like, deep tech is
• 50:34
  still always going to be important.
• 50:35
  And those days everything is just about social, social
• 50:37
  network scaling. That's where all the money was
• 50:39
  going. But you know, Menlo Morgenthaler
• 50:43
  and us were like thinking a lot longer term about what are the
• 50:47
  impacts of deep tech. And this is like deep tech
• 50:50
  wasn't even a thing. Like, you know, people aren't
• 50:52
  really talking about it, right? And these guys led by Adam
• 50:58
  Chair, who's super, super smart, just great guy.
• 51:02
  I'd been working for 10 years in Sri on AI, right?
• 51:06
  And the the opportunity came to do something and spin it out and
• 51:11
  look, it was so under the radar. We launched in February 2010 on
• 51:15
  the App Store and the App Store was very small and it did sort
• 51:20
  of help you with reservations and cinema bookings and things.
• 51:26
  And there was one also the other guys that were very forward
• 51:29
  sighting was Verizon. There was a guy called Ed Ruth
• 51:31
  at Verizon who was very forward thinking and had done a deal
• 51:35
  between Verizon and the Siri team.
• 51:39
  Said that Siri would launch their first.
• 51:42
  And. It was so interesting.
• 51:44
  And so basically what happened then was it launched in the
• 51:47
  store and then Steve Jobs saw it and I about two days later, the
• 51:51
  founders got a call, which they didn't believe, but obviously,
• 51:54
  you know, it turned out to be Steve Jobs and they went into
• 52:00
  the Jobs reality distortion field completely.
• 52:04
  And it was great. Like 4 weeks later, it was sold
• 52:06
  200 million and it was fast and it was crazy.
• 52:10
  And, and the, the, the amazing team.
• 52:12
  And fortunately Steve got sick and, and Siri got lost
• 52:15
  internally for a while, but the team then went on and built
• 52:19
  amazing things and for Samsung and then Airbnb.
• 52:23
  And it's just, I mean, they've just an incredible team that
• 52:26
  have continually innovative in AI constantly with new
• 52:29
  companies. But yeah, back in those days, I,
• 52:33
  I mean, Steve was, you know, his jobs was so far ahead of the
• 52:36
  thing, he saw it. But I have to say, you know, my
• 52:39
  boss at Horizons and, and those guys at Menlo and Morgenthatter
• 52:43
  also saw it, right? You know, they saw it way above
• 52:46
  the people were just, they were just had that long term deep
• 52:50
  tech focus and and none of them were bankers, right?
• 52:53
  They weren't finance guys, so they saw it from a technology
• 52:57
  perspective, not from a finance perspective.
• 52:59
  And no. One ROI series, not a question.
• 53:02
  Or deep Mind when they pitched we're like, we have no idea
• 53:05
  where the revenue's coming from. None.
• 53:07
  And in those days that was novel, right?
• 53:10
  So no VCs doing it like hardly any VC would do it.
• 53:13
  Like no VC was investing in deep mind in Europe in the early
• 53:16
  days, right? Because it was like, well,
• 53:18
  where's the, where's the revenues?
• 53:19
  Where's the, we don't know, Like we're just going to build the
• 53:21
  best AI engine on the planet. Well, why would you do that?
• 53:25
  I mean, like, you know, so that kind of question like comes down
• 53:29
  to it. And I think, you know, again,
• 53:33
  coming back to it, I think there is again, a huge shakeout going
• 53:38
  on where a lot of investors do not like work very
• 53:45
  traditionally, right? And there's a a new way of
• 53:49
  things going to happen where people are going to spin things
• 53:52
  up really quick and jumping on that really quick is going to be
• 53:57
  the key. So yeah, watch this space.