#286 Inside the USD 30 Billion Language Solutions and AI Market

0:00Florian:Perception used to be AI is disrupting this from the outside. The reality really is that the LSI and LTP sit inside and enable organizations to extract value from AI.

0:16Florian:Hey everyone, glad to be back. Hello, Esther. Hi, Florian. It's been a month. No, actually more than a month. Five weeks. Like two months. No, not two months, maybe five, six weeks. Anyway, so we are back from various activities, back from London, had the whole team there, or not the whole team, but many of the team and ran a quick conference, right? Quick conference. Yeah, that's a bit of an understatement, but yes, we had SlateCon London. Yeah, very, very nice conference. But just before we go and do a quick wrap up of SlateCon London, we haven't been inactive on podcasting. We recorded a podcast with DeepL called The New Fluency Podcast.

1:01Florian:So go and put that into Spotify or YouTube. That was Alex and myself. We recorded that. Alex Edwards, head of consulting at Slater, recorded a podcast with DeepL. So head over there and subscribe to The New Fluency Podcast. Now, yeah, no. Okay, let's go back to SlaterCon London. Massive turnout. We filled the room. 250 people. That's a record. Yeah, great turnout. Great turnout. Great turnout. I mean, we keep saying records, but we are actually setting new records with each conference. So this 250 people, that is a new record. It's also kind of capacity of the room. I don't think you, I mean, you could have maybe added a couple, maybe a couple dozen more, but then it would have gotten really cozy out there in the networking area.

1:46Florian:We actually also had a breakout room. What was this? What was it called? The Sayuri room, the Sakura room. You know, it's all Japanese themed. I can't remember which one it was because there are two that sound very similar. Right. And this was really cool. So we had these startup tracks and the research tracks, right? So we had researchers presenting in front of these research posters as, you know, you call them when you go to these big research conferences. And then we had a number of startups that were, you know, pitching their products, you know, five, ten minutes. So that was really cool. I hung out there for a couple of sessions.

2:23Florian:And then we also had amazing traffic in the exhibitor space. Lots of people going into the exhibitor space. The main conference room opened into the exhibitor space. And so they got a lot of traffic there. That was great. And also no last minute stress on our site. Remember Esther in 2025 when the screen broke? The screen, yes. So we have this giant screen that when you take the photos and stuff, it looks really great in the background. And in 2025, that giant screen broke because somebody, I spoke to the producer this time and he remembered. And he told me that last year, somebody unplugged, like some hotel staff unplugged the screen just accidentally, like in a different room.

3:06Florian:They just unplugged it. And so the AV people that we hired didn't know that that person unplugged it. And so they had to like manually recable it right before we started. I think it was about 20 minutes before we started. right well they fixed it that was long story short they fixed it and we uh ran a successful conference in 25 as well nobody realized it but you and i were standing there like all right well we just lost like uh one of the major pieces of infrastructure here 20 minutes before uh the show was set to start so this time we uh we were you know entering or you know climbed up to the stage in deep chill mode because everything worked.

3:46Florian:And so, yeah, that was good for NERVs. Yeah, cool. And I mean, agenda, look, I mean, we were doing a bunch of follow-up coverage. Check out the YouTube video and all the cool YouTube interviews and summers. We're going to be publishing a nice mix of kind of LSI giants and like fast rising LTPs. You know, we had the C of Gradium on a panel, Voice AI. So, yeah, pretty, pretty cool summary, a pretty, pretty cool lineup there. And, yeah, we also have an amazing lineup now starting to shape up for San Francisco for those of you who are in the U.S., Canada, or, you know, want to take a trip there. Happening right after the World Cup.

4:32Florian:So, you know, bringing back some much-needed tourism to San Francisco. No, just kidding. But let me go through to the agenda. So already have a few confirmed speakers, super exciting. So we have Marlon Duarte, who is the Director of Hospitality and Language Services at UCLA Health. And for those of us in Europe who struggle to understand the size of U.S. healthcare, UCLA Health is a massive academical medical system comprising of like over 126 locations in the greater L.A. area. they have a total workforce of 34 ,000 employees. 34 ,000 employees. That's a lot of people. I'm just surprised that Marlon has this hospitality and language services job title.

5:21It's part of the service. I'd love to understand. Yeah, yeah. I'd love to understand how those fit together or not.

5:27Florian:It's because it's part of a bigger experience in the US, right? It's a patient experience. And if you're there and you don't speak English, you know, it's part of the experience that you get that language service. So, yeah, but we'll ask Marlon. And then also super excited to have Emily Xue, head of enterprise AI at ScaleAI as a speaker. Really, really cool. Teresa Marshall, VP Globalization at Salesforce. Eliza Dublava from Coca-Cola, Europe Pacific Partners also joining. And then on the LTP side, we got Georg from Frey, CEO, Marco Trombetti, CEO of Translated and Gabriel Fairman from Works. And we're working on a bunch more speakers and panelists.

6:11Florian:So, you know, happening September 2nd and 3rd in San Francisco. So join us there. And let's move on to our market report. Major new piece of research out. Have you read all of it, Esther? To be honest, no. I mean I will I certainly will but obviously it's uh it's quite a lot of pages and quite a lot of content and it was released what the day before SlaterCon so I was a little bit busy that is true that is true no I mean it's also uh you know we're drawing from a bunch of research that we published already so we're I think especially Slater team we're going to be familiar with with quite a lot of it but yeah amazing piece of research 125 pages uh you know let me just go through some of the chapters.

6:59Florian:We got the language AI, language solutions and AI market overview. We got the scope, market size and scope. We expanded the scope and the TAM last year. So we're looking into this by vertical, by buyer profile, by geography. We again looked at buyer intentions, then basically language AI in the AI economy. So looked at the different capabilities and how these companies and players are positioned in the bigger AI ecosystems. Then we look at the LSIs and LTPs over 30 pages and how they fit into the picture. We ran a bunch of surveys, one of them on the buyer side, then on the linguist side. Then we have a couple of special chapters on US language access, AI dubbing, language AI in legal, and of course, your corner, M &A and funding.

7:50Florian:And then we do a market outlook. So amazing piece of research here that's informing our views on the market and it should inform yours as well. So go and get a copy or subscribe. I think it's coming with the Growth Pro and Enterprise subscription that we have. So go and get that. Now I want to just share some thoughts that I also discussed in London. And this draws heavily, of course, on that piece of research. So, I mean, first the mood was somewhat different this time in London. I feel that we definitely, you know, moved from a kind of like looking at AI as something that's kind of disrupting the, whatever, the established space that we've all lived in from the outside.

8:43Florian:to what really is much more the reality that LSIs and LTPs sit inside the AI economy. And they really enable organizations to extract value from AI. I mean, just having a Claude or a ChatGPT subscription isn't going to deliver your enterprise outcome across the entire org. And that was also a big topic, for example, that we had on the panel with TripAdvisor and Frays. where like, okay, you cannot just build all of this yourself. You cannot DIY the entire localization workflow with all its different assets and users and scalability and nuance just from scratch, right? So perception used to be AI is disrupting this from the outside.

9:29Florian:The reality really is that the LSI and LTP sit inside and enable organizations to extract value from AI. I think it bears repeating. I think it's a really crucial shift as well. Mentally, even if we know it to be true, I think articulating it is important and knowing how to speak about that when you're in conversations with clients or investors or what have you. It also switches the framing around. It gets you out of this defensive posture that, oh, no, there's this bad thing that's coming from the outside and it's kind of nibbling away to like, no, no, no, hang on. We're very much part of this.

10:07Florian:And when you speak to like AI native companies like Sonos, who we had on the podcast last time, they don't even think in these established categories, right? So it becomes kind of a how long have you been part of this like language AI space? And if it's, you know, two or three years, you think about it very differently than, you know, if you've ran a, I don't know, an LSP, formerly known LSP or like a machine translation company in 2016. Then we also broke, we tried to really frame the different parts of the stack here, the market that we're tracking. So, I mean, I'm not going to share the slide here because most of the people listen to this probably on audio, but like there's four layers, right?

10:52Florian:You got the foundational AI, you know, that's the, you know, all the hyperscalers, the Gemini's, the open AI's, the Anthropics. Then on top of that, you have more bespoke language AI models, you know, the gradiums of this world that we had at our conference, the deep Ls. Then you have the application and workflow platform layer, the phrases, the XTMs, what have you, many, many more. And then the managed solutions that sit on top, that's the LSI layer, right? And we're not measuring the foundational AI component. We're looking at the language AI, application workflow, managed solutions, and that's the 30 billion market that we're the total addressable market that we're tracking in the report.

11:33Florian:One thing I also want to share here is the adoption. Let me just pull this up from the presentation here. So first of all, yeah, let's talk about the market size, right? So total addressable market, according to our research in 2025, was$30.85 billion, right? And now that's actually a decline. That's a decline compared to 2024 of 2.7%, a total decline. Now for the LSI component, that's actually an even stronger decline. That's a 5.1 % decline. And I guess that tracks with some of the companies that you're talking to or seeing out in the market, right? Yeah, I'd say so. I mean, I think overall 2.7 is obviously not a huge decline.

12:33from a market level for the LSIs. I mean, 5 % decline roughly sounds more or less right. I mean, I think you have some that are more pronounced and suffering worse and some that are actually doing pretty well.

12:49Florian:It is a tale of two worlds, really. I mean, the 5 % figure is obviously the average. I mean, there's companies that are doing really well that we're seeing and kudos to them and then some that are scaling back quite rapidly. So 5.1 % decline for the LSIs. So this gives you a total TAM for LSIs at about$25.8 billion. And the LTP component here is nearly$5 billion. And that's actually up a lot. That's up nearly 20 % compared to 2024. So that's up 19.7%. And so LTPs now make up 16%, 16.1 % of the TAM that we're tracking and that's up from 13.2%. So the TLDR here really is there's a massive shift going from this lower margin type of high revenue, strong top-line world of the LSIs and the historic, like the LSPs towards much more tech where of course the top line is arguably higher margin but lower volume in terms of just the sheer dollar amount.

14:02Florian:So moving from services to a tech and Tale of Two Worlds, some LSIs are doing phenomenal, others aren't doing so well. Overall, the TAM is going back a little bit. And let me look at the outlook also. Maybe I'll share it as we do. So our base case for, you know, we mapped this out until like 2020. No, not 2020. 2030. 2030, right. map this out for 2030, we have in 2031, the base case would be the LSI component is down to 21 billion, the LTP component up to 14 billion. So, you know, very much. It's like two thirds, right? Yes, two thirds. So, it's going to be a very active space for the LTPs that we're going to be tracking.

14:56Florian:All right. Now, very different numbers are being thrown around. And this is, you know, let's talk about it. It's relevant to this podcast. It's these mega rounds and IPO announcements by, you know, the two biggest labs, Anthropic and OpenAI. So there is Anthropic. Let's start with Anthropic. So they raised$65 billion at a$965 billion valuation. I mean, that's – talk about a bubble. I don't know. If this stuff isn't going to work out, yeah, these are amounts that are unheard of, right? Yeah. And now they're also like literally a few days after they're saying they filed for an IPO. They call it – they filed confidentially for an IPO.

15:49Florian:Okay, they're telling the world that they confidentially filed. All right. At the time, maybe it was confidential. I don't know. I guess the content is confidential or something, but the filing itself is not confidential. Right. But yeah, it's interesting. You'd raise like what? $65 billion and then you file for an IPO and obviously the purpose of an IPO is to get capital. So you raise$65 privately and then what? What are you going to raise at like a$2 trillion valuation? So, yeah, and I mean, the numbers I'm hearing is like 45 or I'm hearing that that are out there is like they're at a$50 billion run rate-ish.

16:30Florian:This is highly relevant because of course, a lot of LTPs are building on top of or using some of this. And then if Anthropic decides to, I don't know, launch and like claw translation or something, this is going to be highly relevant as well. Same for OpenAI. They closed, I mean, that was a while ago. They closed like a$100-plus billion round at a$850 billion valuation. It's kind of weird to even mention these numbers. And they're also looking at a mega IPO in like late 2026. So, yeah. And hey, I found this amazing website. So, when you're looking at all these... Yeah, I need to share this. This I do want to share.

17:11Florian:Let me try to spin up the screen for those of you who are watching this on YouTube. So it's a website that's called, so if you're wondering if this is all profitable because they're spending, you know, literally we're getting into like the hundreds of billions of dollars of investment and trillions of dollars of valuation. So there's a website called is AI profitable? And then on the right side, it is a big... No. Everyone's broke. And so you can like sort by hyperscaler, by lab, by... So they're listing like Amazon, Alphabet, Microsoft, all the investments that are made and how much revenue that's been generated.

17:52Florian:And you have then the labs like OpenAI, Anthropic, X, Mistral, Cohere, DeepSeek. And like according to this page, the only company that's made money on this is NVIDIA, right? That's why NVIDIA is a$5 trillion company. But anyway, it's a fun website. head over to is AI profitable or not for, you know, a, I don't know, a potential, it's kind of a bubble tracker. Who knows how these figures that are ingested into this website, where they're drawn from and how accurate they are, but it's a fun website. I love the layout. Now let's talk about the challenge. So let me try to segue this. Okay. So you got these, you know, hundreds of billions of dollars that are invested.

18:33Florian:it, you got these frontier labs as they're now called. And so if you're an LTP and you're looking for a moat, you're looking for something that, you know, like the Googles and Amazons and Anthropics and opening the eyes of this world can't just kind of build or it's gotten challenging. For example, DeepL, you know, they announced that they were doubling down on voice translation, that they're going to open a San Francisco office, that they're going to be really doubling down on the US. And then we prepared the newsletter, our weekly newsletter that goes out to, you know, it's a free newsletter, goes out to subscribers.

19:13Florian:And I think we sent it and literally I think about 20 minutes and the angle was, hey, DeepL is, you know, doubling down on voice, going to the US, et cetera. And then about 20 minutes later, you open up X and you see, oh, OpenAI just launched GPT real-time translate. You know? So they released new audio models, 70 plus languages, output into 13. It's a speech-to-speech translation model. The model returns both translated audio and text transcripts while the speaker is still talking. OpenAI says it acts as an interpreter, not a conversational assistant. So like full on, like live speech translation, right?

19:52Florian:And so my point here is, you're the bell. you're like okay you know text got some of the control we have enterprise ai translation got you know thousands of clients let's go after voice yeah boom boom boom uh you know five minutes later after you announce that you're going after it open ai launches gpt real trend translate that's super difficult to really find a um a niche that has a kind of a frontier lab safe moat around it I guess that's my point. Yeah. I think we were talking a little bit about that when we had the investor panel at SlateCon as well, weren't we? Sort of finding a niche, finding a moat.

20:34How does one go about that? Very tricky. So yeah, super relevant. Yeah.

20:39Florian:Super relevant, super tricky. And then, yeah, how do you do that? And then it's not even these, not only these frontier labs, I think Zoom also, they moved into AI services with like a new translator and summarizer thingy. Let me pull this up. They also brought their AI Live speech translation in-house. You know, they had all these kind of apps that you could use on Zoom, but now I think they actually, they kind of launched it in-house. Yeah. So does that mean they're not going to have that kind of marketplace anymore? I'm sure they do, but they want to have it as a native feature. And hang on, so there's some limitations to it.

21:18Florian:So they say that Voice Translator currently supports five languages and is limited to paid US-based Zoom accounts. why US space? Yeah so we can't even test it maybe we can let's subscribe over in San Francisco get a new Zoom account up yeah we'll get one of our US colleagues to help out. I mean with Google it kind of works reasonably okay actually still like this type of interaction now will probably be too much but if you have like a basic interaction it works quite well cool alright so that's let's go from the world of you know trillion dollar rounds and modes and all of that to a couple of sad stories or sad.

21:58Florian:The first one's not sad. The first one is interesting. The second is a bit sadder. The first one is Lovo. Did you, have you, had you heard about this company before they blew up? Lovo? I don't think so. No, it wasn't really on my radar, to be honest. So this is an AI voice startup. And I think the website was still live a couple of days ago. So they say they're a hyper-realistic AI voice generator. and they have 500 voices and 100 languages. I mean, this whole languages seems to be... Everybody's saying they have 100 plus languages. Yeah. Except Zoom, they have five because they actually need to make it work.

22:37Florian:But yeah, on the website, Lova said they had 500 voices, 100 languages. But now they had to file for Chapter 7 bankruptcy in New York. And I Googled this and I'm still Googling. So chapter seven means it's liquidation. It's not like you're protected from creditors. I think chapter 11, you can still operate, but chapter seven is like, okay, you're done. You liquidate. Now, why? Why does Lovo, that voice AI startup, why did they have to liquidate the company? It's because they went, I'm trying to get this right. Otherwise go, I'm sure we're going to link the article in the show notes. They went on Fiverr, one of those crowd platforms, right?

23:21Florian:They bought some voice samples from voice actors and then they just started to commercialize it and use it without permission or fair compensation. That's what plaintiffs called Paul Lerman and Linnea Sage allege. And so they said that they're seeking to represent other actors whose voices or names were also used by Lobo and yeah, so it looks like the judge initially dismissed one copyright claim tied to voice imitation but then they allowed something, let me read this off here, New York civil rights law claims to continue and saying that voice clones can count as digital replicas. Point being that I think Lobo people, they realized that this isn't going to end well for them and they liquidated or they went into chapter seven.

24:18Florian:And yeah, then the plaintiff said that the bankruptcy may prevent them from seeking full justice in court because that's probably going to end the litigation here. Let me bring this back to our Data4AI report where actually rights management was a big part of it, like companies like Oxford Languages, making sure that everything's properly licensed. So, yeah, you want to make sure that if you use these types of data and these types of, yeah, for AI, you want to make sure that you actually have the rights to this data, which in this case was existential. It's like there's a bit of a crossover here because the rights management stuff has been a topic of conversation for a very, very long time in the entertainment world, hasn't it?

25:05So, you know, the right to use voices, actors' rights, et cetera. I suppose that still holds true, but it's just coming sort of front and center within data for AI now.

25:15Florian:And here, I mean, I understand. Like if you're a voice actor on Fiverr, you get paid, what,$50 to record something? And then it's just like commercialized your voice? It's like, oh. I vaguely remember. Hang on. Wasn't there the person, not the Siri person, but somebody who like was in like litigation with Apple or something? Anyway, we'll look this up. All right, staying with voice, text, and a sad story is a company called VIQ Solutions. We had their CEO on the podcast. I had to look this up. This was actually on 2nd October, 2023, Sebastian Perret, CEO of Vix Solutions. He joined SlaterPod. But now, that company is actually fighting for Survival, Canadian transcription company.

26:03Florian:They're publicly listed, so they have to disclose all this. So, they have a tiny market cap now of about 5 million Canadian dollars and their revenue fell to about 40 million US dollars, but the net loss nearly doubled to$13 million. So, they ended their 2025 with negative shareholder equity. That's a pretty major red flag for solvency. And then in Australia, I know when it rains, it pours, they had a data privacy incident and had to put their local division there into voluntary administration. So like some type of, I don't know, I guess insolvency in March, 2026. So yeah, court transcription, all kinds of other services they're providing.

26:50Florian:Looks like business isn't going well and maybe they're going to have to close up shop. So, VIQ solutions, yeah, we'll watch this very closely. Do you want to walk us through a few M &A transactions that happened since we last spoke? Yeah, there have been a few. Obviously, I think, as we said, it's been sort of five, six weeks since we last caught up on the pod. So, I mean, one observation that I made, and we won't go into too much detail about this, because I think Voice AI has already occupied quite a lot of the podcast today. but there are now an increasing number of acquisitions that we're tracking we're coming across in the voice ai space so that's acquisitions um i mean funding i guess comes as no surprise that there's a lot of voice ai funding stories out there um but yeah we've seen an increasing number of acquisitions um in the space as well um but more so in the world of lsi acquisitions We had news from a couple of super agencies and then three, four other stories as well.