00:23:50 · Carles Reina, Guillaume Moubeche
So as a business, if I had to start doing it again today, I would segment it in in building centers based on, like, in the market, on the industry tackling, what is a higher priority versus a lower priority, and essentially, like, de incentivize some sort of products as well. It is much more complex selling an agent's product versus, like, selling a single API endpoint. So there is some of these things that actually need to like craft it into the entire compensation plans. [00:24:13] S1: Before like talking about your fund strategy and the the investments you've made, I'm curious to like touch base on everything that's happening right now in AI. Like, we're seeing, you know, the rise, obviously, I mean, we've seen the rise of Anthropic, OpenAI as, like, closed source models. And we're seeing also that open source, all the Chinese models are getting, like, better and better. We're seeing, you know, like, Hugging Face, like, being acquired by NVIDIA. I mean, there are a lot of things, like, happening. What's your view, like, for ElevenLabs versus, you know, like, all those open source and kind of competitors in some way that can pop up like over the years? Like what's the technical, like strategy do you think you guys have been able to develop that is very hard for other people to catch up on?
00:25:01 · Carles Reina
[00:25:01] S0: I think like the market is saturated with like models, right, across the board. Like, whether that is like text LLM, that is video, that is image, that is like audio, voice music, whatever you wanted. I think like what we ended up doing at ElevenLabs was like, let's try to play the card of like the best research for audio technology. Right? Like, let's always have the best models out there in the market and so on. Now, of course, like you end up like the moment you do it and you go this quickly and you go this quickly, it pays like a lot of people pay different attention points and end up having like competitors pop up left and right. Some of them using... Were using our own API endpoint and claiming it was their own technology. Some other ones building like on top of open source. Some other ones like building proprietary research, all of that stuff. Right? And I think the core element was like always the same, which is like, let's build foundational models, the best in the market, and then let's build products on top of the foundational models that we're able to actually keep, like, keep just going to the market and providing customers like the best research and the best product together. So we pay a lot of attention on the Chinese model because they are absolutely amazing. Everyone like from a tech generation to video generation and so on. And you can see like video generation, the Chinese models are the best ones in the world right now. Right? No. Like, I mean, see that it's like absolutely amazing. What is the downside? The downside is if you're selling to large companies in The US and those companies might have a contract with the US government, the reality is like you cannot implement open source models. And a lot of like what the media publicly like forgets about it, and everyone... I know like the buzz is like, oh, those Chinese models are amazing. And they are. But if you truly have big American customers, you cannot use open source models from China. And, like, people don't fully understand that because, like, it's something, like, not that many people talk about, but that's the reality. So even today when, like, do you think about building agents? You're like, oh, let's use a quen model. Amazing product. Like, an absolutely amazing product. One of the best one for, like, like, low latency, like, real time conversation and so on.
00:26:59 · Carles Reina
But the reality is, like, you're not gonna be able to use it. You do it for demos. So, like, you feel that... Like, everyone feels like this technology is amazing. It works really fast. Like, there is no latency. All... Per a practical level, when you take it to a company, you no longer can use a coin. Right? So, like, there is a lot of these things that, like, when you scratch a little bit the surface, you realize, like, yes, amazing foundational models that, like, Chinese have, but at a practical level, they're just not, like, to the point that you can embed in your product and get specifically US customers start using it, or European companies that have a lot of, like, US operations to start using it. Right? So there's some of these things on the politics side that actually, like, are not ideal, that essentially are being much more favorable to, like, US companies right now just because of the structure of everything it is. Whereas in Europe, like, can still use Mistral or you can use, like, any variation of it. And depends on the company. Like, they will tell you that, yeah, open source Chinese models. What is it hosted? Are you connecting to the API endpoint or are you hosting it yourself? What is that hosting? Is it like in Europe or is it somewhere else? So you end up answering a bunch of questions on maybe some of the big companies are also using open source Chinese models if they don't have US operations. Right? So there is some of that stuff, like there's variations on which market you operate in.
00:28:15 · Guillaume Moubeche, Carles Reina
[00:28:15] S1: I think for Europe, it's gonna be a good test for Chinese and Chinese like open source model because in The US, I mean, it's obvious that there is the open competition kind of like the global business war, etcetera, while Europe, you know, is a bit in the middle. And I don't think, like, Europeans have the same view on Chinese model as The US have, like, for many, and and I think it's gonna be interesting to see, like, how great, you know, like, some models are getting, to what point, to what extent can they be used and and see how how it's evolved. For you, like, whenever you see people [00:28:51] S0: talking about... Correct. To me, like, it reminds me a lot to, like, the Chinese vehicles, the Chinese cars. Like, there's a very interesting comparison. Like, in The US, you can still not sell Chinese vehicles Yeah. For a number of restrictions in there. But in Europe, like, we kinda embraced it. And then, like, in many European countries, like, you see mostly Chinese vehicles out there, and then look at the, like, European manufacturers that are all struggling. I'm not saying, like, it is right or wrong. I'm saying, like, it's a completely different approach, and usually Europe is like a testing ground for, like, what the Chinese companies are able to do later on in The US once the market clears up. Right?
00:29:24 · Guillaume Moubeche
[00:29:24] S1: Yeah. And the funny part is, like, if you look at at The US economy, and I think people forget it a lot, but if you look at the iPhones, Nike products, like most of the American brands, whether it's electronic or, you know, like clothes, fashion, etcetera, at some point, it's been produced in China, you know, like... So I think for a long time, underestimated Chinese, and they were saying, like, okay, they can only copy, they can't innovate, etcetera. But for me, it's like... It seems that everything has been shifting, and for Europe, it's a bit sad also, like when you look at the manufacturing capabilities and when it comes to cars, and this was a big industry back in the days, but after moving like all the facilities, like in Asia, etcetera, etcetera, like, now we we have, I don't know, maybe like twenty years delay on on everything. 100%.
00:30:16 · Carles Reina
[00:30:16] S0: Yeah. Like, and look at the, like, the foundational models. Right? Like, literally, like, what are we producing? We have Mistral. We have, like, Black Forest Labs. We have ElevenLabs. We have some of the newer labs. Some of them are French. Some of them are, like, UK based and so on. But we don't produce as much core foundational model like technology. And when you even look like there might be a bubble or not in the, like, data center space and we could spend, like, hours talking about this. Right? But, like, I was looking at some of the stats, like, this weekend, for instance, and I was talking to my brother and we were saying like, oh, it's great, like in Spain, like we are starting to push like towards like data center, although it might be a bubble. Like, well, are we? Right? Like, the Spanish government has committed to spending like €720,000,000 on a crazy, like, one gigabyte of, like, data center power. I was like, €700,000,000? That is laughable. Like, it's not gonna give you anything. Right? Like, at how much is it actually gonna cost to build a single gigabyte or, of a data center. That's gonna be, like, in the tens of billions of dollars, like around like $20,000,000,000. Right? Like, literally, like The US is spending it. A lot of investors are actually spending it. Like here, we're not that serious. Right? Like European Commission should be saying like, hey, like, there you go, like $400,000,000,000 to actually create all of this technology from scratch. And by the way, we're gonna be building, like, models, and we'll give, like, another $2,000,000,000 to Mistral, and another billion dollars to ElevenLabs, and we're gonna do this, and we're gonna do that, all of that stuff. So you build the ground up from the four core infrastructure. We're just not doing it. We're not serious enough. China has been very serious. US has been very serious. Of course, they're actually a front... Like, at forefront of everything right now.