WEBVTT - Doubleword CEO Meryem Arik Talks Scaling and AI Investment 

0:00:02.480 --> 0:00:07.480
<v Speaker 1>Bloomberg Audio Studios, podcasts, Radio News.

0:00:07.600 --> 0:00:10.039
<v Speaker 2>Now want to get into a big conversation this morning.

0:00:10.320 --> 0:00:12.799
<v Speaker 2>We are in the midst of London Tech Week. Lizzie,

0:00:12.840 --> 0:00:17.840
<v Speaker 2>you'lso out and about founder's forum British AI business. Double

0:00:17.880 --> 0:00:22.119
<v Speaker 2>Word the UK's first dedicated inference lab. It's one of

0:00:22.120 --> 0:00:24.040
<v Speaker 2>the first companies that has been backed by the UK

0:00:24.160 --> 0:00:28.440
<v Speaker 2>Government's Venture Fun for domestic AI startups. We've spoken to

0:00:28.480 --> 0:00:31.960
<v Speaker 2>the Air Minister about this and I'm really delighted that

0:00:32.000 --> 0:00:34.519
<v Speaker 2>we have Meriam Eric, who is the co founder and

0:00:34.600 --> 0:00:36.919
<v Speaker 2>CEO of double Word, in the radio studio with us.

0:00:36.920 --> 0:00:41.400
<v Speaker 2>Welcome to Bloomberg Radio. Yeah, Meriam, I know that you've

0:00:41.440 --> 0:00:44.040
<v Speaker 2>been right at the heart of this government activity with

0:00:44.120 --> 0:00:47.600
<v Speaker 2>London Tech Week. You've been to Downing Street already. How

0:00:47.640 --> 0:00:51.120
<v Speaker 2>do you think about the government's announcements in terms of

0:00:51.400 --> 0:00:53.080
<v Speaker 2>helping businesses like yours?

0:00:53.760 --> 0:00:56.240
<v Speaker 1>So the government is taking a very different approach to

0:00:56.320 --> 0:01:00.480
<v Speaker 1>how UK governments have typically treated the tech sector. So

0:01:00.560 --> 0:01:03.600
<v Speaker 1>typically UK governments have had a very hands off approach,

0:01:05.040 --> 0:01:08.399
<v Speaker 1>and unfortunately most of our competitive countries don't have a

0:01:08.400 --> 0:01:12.039
<v Speaker 1>hands off approach. And almost every single generational business that

0:01:12.080 --> 0:01:14.520
<v Speaker 1>has ever been built has had a very hands on

0:01:14.720 --> 0:01:18.119
<v Speaker 1>interventionist approach from their governments and so for the first time,

0:01:18.200 --> 0:01:21.320
<v Speaker 1>what we're seeing from this government is wanting to back

0:01:21.440 --> 0:01:26.320
<v Speaker 1>the strategically important AI companies of the future and really

0:01:26.400 --> 0:01:28.720
<v Speaker 1>back them so that they not just start here, which

0:01:29.120 --> 0:01:31.760
<v Speaker 1>plenty of companies do, but also stay here and scale

0:01:31.760 --> 0:01:34.280
<v Speaker 1>here so we can really build generational companies in the

0:01:34.360 --> 0:01:36.319
<v Speaker 1>UK as well. But are you taking what the Starma

0:01:36.400 --> 0:01:40.080
<v Speaker 1>administration is saying seriously when there's so much speculation about

0:01:40.080 --> 0:01:42.600
<v Speaker 1>who's going to be the Prime minister and for example

0:01:42.920 --> 0:01:45.480
<v Speaker 1>Andy Burnham, the Great Manchester may is talking about more

0:01:45.520 --> 0:01:49.480
<v Speaker 1>regulation on tech. Yes, I think the tech sector gets

0:01:49.480 --> 0:01:52.880
<v Speaker 1>painted with a very broad brush. I think it's very

0:01:53.040 --> 0:01:56.440
<v Speaker 1>like the Keir Starmer administration as well also wants a

0:01:56.480 --> 0:01:59.600
<v Speaker 1>lot of regulation on big tech and social media. But

0:01:59.640 --> 0:02:02.400
<v Speaker 1>I think that it's very different to the very innovative

0:02:02.560 --> 0:02:05.440
<v Speaker 1>startup companies and scale up companies that are kind of

0:02:05.440 --> 0:02:07.480
<v Speaker 1>British at its core, and they want to keep British

0:02:07.680 --> 0:02:09.799
<v Speaker 1>and they're really putting kind of their money where their

0:02:09.840 --> 0:02:13.519
<v Speaker 1>mouth is, and they've we are part of a five

0:02:13.639 --> 0:02:17.320
<v Speaker 1>hundred million dollars sovereign AI fund. We have received a

0:02:17.320 --> 0:02:20.760
<v Speaker 1>compute allocation from them. They've also recently announced two days

0:02:20.760 --> 0:02:25.200
<v Speaker 1>ago a one point one million pound fund to procure

0:02:25.280 --> 0:02:28.600
<v Speaker 1>computees so we can build more supercomputers in this country.

0:02:28.880 --> 0:02:30.840
<v Speaker 1>And so it does seem like they're taking this very

0:02:30.880 --> 0:02:34.200
<v Speaker 1>very seriously. Although something I'm worried about is a change

0:02:34.200 --> 0:02:36.519
<v Speaker 1>in government will lead to a change in the ministers

0:02:36.520 --> 0:02:39.040
<v Speaker 1>in the Department of Sites and Technology that we think

0:02:39.080 --> 0:02:40.160
<v Speaker 1>have been doing a very good job.

0:02:40.440 --> 0:02:44.760
<v Speaker 2>Okay, okay, some praise then for those ministers. I mean

0:02:44.840 --> 0:02:48.040
<v Speaker 2>that the issue though, with picking the winners and losers

0:02:48.080 --> 0:02:50.639
<v Speaker 2>in the in the UK simply that Britain isn't spending

0:02:50.760 --> 0:02:52.400
<v Speaker 2>very much money. I mean, if you look at the

0:02:52.480 --> 0:02:57.320
<v Speaker 2>trillions of dollars being used for the kind of infrastructure

0:02:57.480 --> 0:03:00.000
<v Speaker 2>build out in the US, it pales into infra significant.

0:03:00.160 --> 0:03:02.040
<v Speaker 2>But to understand that, it's sort of trying to pick

0:03:02.040 --> 0:03:05.440
<v Speaker 2>the winners in terms of double word though, and you've

0:03:05.480 --> 0:03:08.640
<v Speaker 2>got the backing of government, so that's one thing. Also

0:03:08.680 --> 0:03:10.520
<v Speaker 2>that they are going to help you with that compute

0:03:10.560 --> 0:03:13.800
<v Speaker 2>which is very important and very costly for startup businesses.

0:03:14.040 --> 0:03:16.320
<v Speaker 2>How quickly are you managing to grow though? I mean,

0:03:16.639 --> 0:03:21.320
<v Speaker 2>inference basically allows companies to use AI to make AI

0:03:21.520 --> 0:03:24.400
<v Speaker 2>work for them, and that is massively important for growth

0:03:24.600 --> 0:03:26.720
<v Speaker 2>in brit How fast you're scaling up, What are your

0:03:26.760 --> 0:03:30.800
<v Speaker 2>customers actually saying about whether they are doing good things

0:03:30.800 --> 0:03:32.679
<v Speaker 2>with AI exactly?

0:03:32.800 --> 0:03:36.560
<v Speaker 1>I think, as you say, the government doesn't have a

0:03:36.600 --> 0:03:38.400
<v Speaker 1>huge amount of money to throw around, and so they're

0:03:38.400 --> 0:03:40.640
<v Speaker 1>being very strategic in how they give this money out.

0:03:41.040 --> 0:03:44.960
<v Speaker 1>Something that I've been particularly impressed with is the choices

0:03:46.120 --> 0:03:49.800
<v Speaker 1>when they're deciding which startups to back. It's VC professionals

0:03:49.800 --> 0:03:52.640
<v Speaker 1>and investment professionals that are choosing those companies, and not

0:03:52.720 --> 0:03:55.880
<v Speaker 1>necessarily like government civil servants, and so they are very

0:03:55.880 --> 0:04:01.480
<v Speaker 1>selectively picking a couple of strategically important winners. US. For example,

0:04:01.520 --> 0:04:04.480
<v Speaker 1>we are a dedicated inference lab. The problem of tokens

0:04:04.480 --> 0:04:06.320
<v Speaker 1>and token costs is going to be a huge issue

0:04:06.320 --> 0:04:10.480
<v Speaker 1>in the UK, especially given how expensive are energy bills are,

0:04:11.600 --> 0:04:14.000
<v Speaker 1>and so they've picked just a couple of companies that

0:04:14.320 --> 0:04:17.000
<v Speaker 1>they think can really play in this space. It's not

0:04:17.040 --> 0:04:20.880
<v Speaker 1>like they're picking, you know, every AI application level company,

0:04:20.920 --> 0:04:23.279
<v Speaker 1>because actually there's also plenty of VC funding to go

0:04:23.320 --> 0:04:26.440
<v Speaker 1>around as well. We are predominantly VC backed and we

0:04:26.480 --> 0:04:29.280
<v Speaker 1>aren't taking money from the government. Just kind of compute

0:04:29.160 --> 0:04:30.960
<v Speaker 1>where they can be strategically important.

0:04:31.040 --> 0:04:33.279
<v Speaker 2>I mean, you've raised as far as I can make

0:04:33.320 --> 0:04:36.040
<v Speaker 2>out a little over twenty six million dollars when you're

0:04:36.080 --> 0:04:39.120
<v Speaker 2>backed by Dawn Capital, by Oxford's venture Atomic and so

0:04:39.600 --> 0:04:42.479
<v Speaker 2>that idea though, staying in Britain, building in Britain, staying

0:04:42.480 --> 0:04:47.239
<v Speaker 2>in Britain. Yes, Now what happens later as the business grows?

0:04:47.320 --> 0:04:49.719
<v Speaker 2>Would you be committed to staying in the UK.

0:04:50.320 --> 0:04:53.239
<v Speaker 1>Yeah. The common pattern that we see is that Britain

0:04:53.680 --> 0:04:57.320
<v Speaker 1>is great at starting businesses. We have great pedigree, so

0:04:57.520 --> 0:05:01.479
<v Speaker 1>we are kind of Oxford, UCL cac L scientists and

0:05:01.520 --> 0:05:04.320
<v Speaker 1>we start in Britain. The issue is is that Britain

0:05:04.360 --> 0:05:07.080
<v Speaker 1>doesn't necessarily have our customers. A lot of our customers

0:05:07.279 --> 0:05:09.599
<v Speaker 1>are abroad and are in the US, and so what

0:05:09.640 --> 0:05:12.599
<v Speaker 1>we often see is company starting here but then moving

0:05:12.600 --> 0:05:15.839
<v Speaker 1>abroad and then scaling abroad. And that's specifically the challenge

0:05:15.880 --> 0:05:19.440
<v Speaker 1>that this government is trying to stop. We think, and

0:05:19.480 --> 0:05:22.039
<v Speaker 1>we've committed to building our R and D tech cubs

0:05:22.160 --> 0:05:24.360
<v Speaker 1>in the UK and in London because we think there's

0:05:24.440 --> 0:05:27.320
<v Speaker 1>fantastic talent to benefit here and I would like to

0:05:27.360 --> 0:05:29.520
<v Speaker 1>think of the UK as more than just an R

0:05:29.560 --> 0:05:31.400
<v Speaker 1>and D hub, but also as a place where we

0:05:31.440 --> 0:05:35.479
<v Speaker 1>can have great customer relationships, firstly with the government but

0:05:35.560 --> 0:05:37.600
<v Speaker 1>also with businesses as well.

0:05:37.880 --> 0:05:40.839
<v Speaker 3>For our markets audience Burrium. I'd love to just point

0:05:40.880 --> 0:05:43.760
<v Speaker 3>out a figure that was in the TechNation report this week.

0:05:44.160 --> 0:05:47.120
<v Speaker 3>Thirty percent of UK tech leaders say that AI is

0:05:47.160 --> 0:05:50.440
<v Speaker 3>a bubble. If you're in this bubble, if it is one,

0:05:50.880 --> 0:05:53.159
<v Speaker 3>what's your view on that, is it going to burst?

0:05:53.720 --> 0:05:55.920
<v Speaker 1>Well, if it is a bubble, then it's an incredibly

0:05:55.920 --> 0:06:00.120
<v Speaker 1>productive one. It doesn't look like the bubbles that we

0:06:00.120 --> 0:06:05.039
<v Speaker 1>we saw in the two thousands. This bubble kind of

0:06:05.560 --> 0:06:08.279
<v Speaker 1>is a bubble of valuations potentially, but actually it seems

0:06:08.320 --> 0:06:11.719
<v Speaker 1>like this is largely being backed up by revenues by

0:06:11.760 --> 0:06:15.880
<v Speaker 1>all of these big players. We're seeing record breaking and

0:06:15.960 --> 0:06:20.400
<v Speaker 1>revenue growth from the likes of Anthropic Open AI in

0:06:20.560 --> 0:06:24.599
<v Speaker 1>video and others as well, and so obviously there will

0:06:24.600 --> 0:06:26.800
<v Speaker 1>always be some winners and some losers, as there are

0:06:26.880 --> 0:06:30.400
<v Speaker 1>adventure but actually this is being backed up by real

0:06:30.440 --> 0:06:31.719
<v Speaker 1>revenue for these businesses.

0:06:32.320 --> 0:06:35.599
<v Speaker 2>It's not really delivering productivity growth in the UK. But

0:06:35.680 --> 0:06:38.279
<v Speaker 2>then maybe that's also because we're right at the beginning

0:06:38.320 --> 0:06:42.320
<v Speaker 2>of this. I suppose again, what are you offering your

0:06:42.320 --> 0:06:45.520
<v Speaker 2>customers because part of the package is you know, you're

0:06:45.560 --> 0:06:48.200
<v Speaker 2>talking about global customers and customers in the US, but

0:06:48.320 --> 0:06:52.960
<v Speaker 2>part of this idea is allowing sovereignty in AI, allowing

0:06:53.440 --> 0:06:57.280
<v Speaker 2>a kind of ring fenced AI for businesses that they're

0:06:57.320 --> 0:07:02.200
<v Speaker 2>not so reliant on global US best based businesses. How

0:07:02.200 --> 0:07:05.080
<v Speaker 2>do you balance that the idea of delivering sovereign AI.

0:07:05.560 --> 0:07:09.720
<v Speaker 1>So the biggest problem in AI at the moment is

0:07:09.760 --> 0:07:12.600
<v Speaker 1>the problem of inference and the problem of tokenomics. And

0:07:12.680 --> 0:07:14.600
<v Speaker 1>so that is the problem of like, how do you

0:07:14.640 --> 0:07:16.760
<v Speaker 1>actually run these models and how do you actually run

0:07:16.800 --> 0:07:20.680
<v Speaker 1>them in a way that is semi affordable. So at

0:07:20.760 --> 0:07:22.760
<v Speaker 1>double World, we've been thinking about token costs for a

0:07:22.880 --> 0:07:25.440
<v Speaker 1>very very long time. But only a couple of weeks ago,

0:07:25.840 --> 0:07:29.640
<v Speaker 1>Uber announced that it had blown through its entire AI

0:07:29.720 --> 0:07:33.480
<v Speaker 1>budget or by April, so in four months, it had

0:07:33.480 --> 0:07:35.840
<v Speaker 1>blown through twelve months of budget. Because that is just

0:07:35.920 --> 0:07:38.680
<v Speaker 1>how expensive AI has turned out to be to adopt

0:07:38.680 --> 0:07:41.200
<v Speaker 1>for some businesses. And that's the problem we're trying to solve,

0:07:41.280 --> 0:07:44.400
<v Speaker 1>is how can we allow businesses to adopt AI at scale,

0:07:44.720 --> 0:07:46.840
<v Speaker 1>but also do so in a way that's affordable and

0:07:46.880 --> 0:07:48.040
<v Speaker 1>in a way that they can control.

0:07:48.480 --> 0:07:51.880
<v Speaker 3>Well, it's interesting that Anthropic's got the token model. I

0:07:51.920 --> 0:07:56.080
<v Speaker 3>wonder if you have thought strong opinions on Anthropic versus

0:07:56.120 --> 0:07:58.440
<v Speaker 3>Open AI, because of course the IPOs are coming at

0:07:58.480 --> 0:07:59.239
<v Speaker 3>the same time.

0:07:59.160 --> 0:08:02.800
<v Speaker 1>Roughly yeah, and they're both incredibly impressive businesses. And only

0:08:02.880 --> 0:08:06.240
<v Speaker 1>yesterday we saw Anthropic release a new model which was

0:08:06.320 --> 0:08:10.320
<v Speaker 1>a version of that much reported Mythos model that everyone's

0:08:10.320 --> 0:08:13.960
<v Speaker 1>been talking about. It seems in terms of capability there's

0:08:14.000 --> 0:08:17.760
<v Speaker 1>really not much between them. I would say at the moment,

0:08:17.800 --> 0:08:23.240
<v Speaker 1>there's probably slightly wider adoption amongst businesses for Anthropic, but

0:08:23.320 --> 0:08:26.720
<v Speaker 1>open AI seems to have much more consumerttraction, and so

0:08:26.760 --> 0:08:29.520
<v Speaker 1>they're really really neck and neck. But also I do

0:08:29.560 --> 0:08:31.400
<v Speaker 1>believe that there's space for both, and I think it's

0:08:31.480 --> 0:08:33.800
<v Speaker 1>very healthy that we have both in healthy competition there.

0:08:34.280 --> 0:08:37.199
<v Speaker 1>But also we are very strong believers in open source

0:08:37.200 --> 0:08:40.720
<v Speaker 1>models as well, and so we actually think open source

0:08:40.720 --> 0:08:43.040
<v Speaker 1>models are going to play a bigger role in the

0:08:43.080 --> 0:08:46.000
<v Speaker 1>AI landscape than some people currently believe.

0:08:47.760 --> 0:08:53.920
<v Speaker 2>How in terms of the UK, how close the EU

0:08:53.960 --> 0:08:56.520
<v Speaker 2>has a different kind of regulatory structure. How close is

0:08:56.600 --> 0:08:58.920
<v Speaker 2>the UK to the EU? I mean you say that

0:08:58.960 --> 0:09:03.600
<v Speaker 2>there's plenty of VC money in the UK, How are

0:09:03.640 --> 0:09:08.000
<v Speaker 2>you thinking about Europe in the contest of the US

0:09:08.559 --> 0:09:10.840
<v Speaker 2>compared to the US. I suppose now about that again,

0:09:10.880 --> 0:09:13.839
<v Speaker 2>that kind of sovereignty idea of the UK and Europe.

0:09:14.280 --> 0:09:17.800
<v Speaker 1>So I think London and the UK really stands apart

0:09:17.840 --> 0:09:20.000
<v Speaker 1>in terms of the amount of VC investment we have.

0:09:20.200 --> 0:09:23.040
<v Speaker 1>So in Q one we had record amounts of VC

0:09:23.160 --> 0:09:25.640
<v Speaker 1>investment in AI startups. I think it was five point

0:09:25.679 --> 0:09:30.120
<v Speaker 1>eight billion dollars, and so we are kind of out

0:09:30.120 --> 0:09:33.240
<v Speaker 1>competing our European neighbors, although there are very strong AI

0:09:33.280 --> 0:09:36.480
<v Speaker 1>companies in Europe as well, namely like Mistral and Lovable

0:09:37.320 --> 0:09:41.400
<v Speaker 1>for example. But when we think about sovereignty in this country,

0:09:41.480 --> 0:09:44.680
<v Speaker 1>I don't necessarily think about it in terms of as

0:09:44.679 --> 0:09:48.120
<v Speaker 1>a competition between any other country. I think about it

0:09:48.200 --> 0:09:50.679
<v Speaker 1>as what do we need to make sure that we

0:09:50.720 --> 0:09:53.600
<v Speaker 1>can keep the lights on when things if things go

0:09:53.679 --> 0:09:56.760
<v Speaker 1>pair shaped? And what that means is we need, you know,

0:09:57.200 --> 0:09:59.960
<v Speaker 1>lower energy costs in this country, We need more data

0:10:00.080 --> 0:10:02.440
<v Speaker 1>centers in this country, and we need you know, inferant

0:10:02.520 --> 0:10:05.200
<v Speaker 1>software and you know, great model companies as well, so

0:10:05.240 --> 0:10:08.800
<v Speaker 1>that we can run that AI you know in this country.

0:10:08.800 --> 0:10:10.040
<v Speaker 1>If the worst were to.

0:10:10.000 --> 0:10:11.920
<v Speaker 3>Happen, well, yeah, the Bank of thing the Governor Andrew

0:10:11.960 --> 0:10:14.840
<v Speaker 3>Baileys want the AI main I need to be rationed

0:10:14.920 --> 0:10:18.199
<v Speaker 3>because of the energy capacity constraints. And you know governments

0:10:18.559 --> 0:10:22.440
<v Speaker 3>facing these choices between deploying AI and healthcare versus defense

0:10:22.760 --> 0:10:26.840
<v Speaker 3>versus other sectors. How far off are we from those

0:10:26.880 --> 0:10:27.880
<v Speaker 3>sorts of decisions.

0:10:28.480 --> 0:10:31.720
<v Speaker 1>It's an incredibly good question. And at the moment some

0:10:31.760 --> 0:10:34.199
<v Speaker 1>companies are having to ration their usage because of how

0:10:34.200 --> 0:10:37.240
<v Speaker 1>expensive they are finding it, and it is particularly a

0:10:37.240 --> 0:10:39.839
<v Speaker 1>problem in the UK because of our energy costs. One

0:10:39.840 --> 0:10:41.839
<v Speaker 1>of the projects that we've been doing on is and

0:10:41.920 --> 0:10:45.960
<v Speaker 1>barred the UK supercomputer is at dou World. We've been

0:10:46.000 --> 0:10:48.160
<v Speaker 1>doing a lot of research in how can we make

0:10:48.160 --> 0:10:52.440
<v Speaker 1>this supercomputer more efficient for producing tokens such that we

0:10:52.480 --> 0:10:55.880
<v Speaker 1>can make the most of the limited kind of energy

0:10:55.920 --> 0:10:57.800
<v Speaker 1>and limited money that we have, so we can get

0:10:57.800 --> 0:11:00.480
<v Speaker 1>the most out of it so we don't have to ration. Otherwise,

0:11:00.520 --> 0:11:03.840
<v Speaker 1>if we don't look at efficient inference, then we will

0:11:03.880 --> 0:11:06.079
<v Speaker 1>have to ration this AI and that is not going

0:11:06.120 --> 0:11:07.680
<v Speaker 1>to be good for anyone.