Dylan Patel, SemiAnalysis, Nebius, Glean, Legora.. 12 Hot Takes From The Biggest Names in AI

[00:00] Raised. Raised. Raised. Raised. Paris. Paris. Paris. Paris. Paris is beautiful, but it's hot as... Paris in the middle of the summer. Middle of the heat. People are ignoring it. They're showing up. [00:16] We have Kasser of Applied Intuition. Kasser, how are you doing? I'm doing fantastic. Thanks for having me again. Good to see you. So you're coming off of what some people are saying, [00:30] get canceled for talking about AI. So what happened there? Yeah, it was at my undergrad. I think... [00:35] I mean, to give credit where credit is due, I think it's, you know, I went to this place called the General Motors Institute, Kettering University now. I talk about it in the commencement speech. [00:45] And let's say the... [00:47] audience, it's not like East Coast, West Coast elite schools. I think the GMI, mostly guys, the GMI guys are more... [00:59] Midwest, pragmatic. So I didn't expect booze, thankfully. And I'm from there. And hometown... [01:05] Hometown person. But, yeah, I mean, I gave, I think, what is my authentic opinion, which is, [01:11] I think you can have a lot of anxiety about this. You're a new grad. I mean, if you talk to new grads, [01:16] We hire, obviously, lots of them. That's the prime fear. Like, is my career done? Is everything... And I just don't believe that. Like, fundamentally, I don't believe, like... [01:26] jobs are all going away and [01:28] I just... If there's something in my mind... I can be...

[01:31] This might age really poorly. By the way, it's also not true already. It has been reported that jobs have only increased. Yeah, and if you look at, like, when I was graduating, there was this whole big, looming technological reckoning, and that was the Internet. [01:47] and you know all these things were being changed at the time and uh... [01:50] there was a whole view of like [01:52] what's going to happen to the businesses I'm still in 25 years later. Obviously, we still drive, and we still need construction equipment. [02:03] I think there's something similar in the future. Like these things are not going away. They're going to change because technology changes things. So I talked about that, and I think it landed well. [02:12] the risky bet in the commencement speech was, [02:15] It wasn't like [02:17] everything will be [02:18] lollipops and rainbows. It's like, hey, that's life. [02:22] Like your 20s are actually a pretty... [02:25] tough time in your life. You've got to kind of... [02:27] figure out how to get in the right [02:29] you know, groove. [02:30] And you have to do this in this very volatile environment that we have. [02:33] Yeah. That's a soft way to put it. I think Jensen Wong once said at Stanford, I wish you all pain and suffering. Yeah, exactly. It's part of that. I think like [02:45] Like there's like the other extreme, which is like... [02:47] uh... imagine you don't have to work or you know you [02:51] That's very unfulfilling too. [02:53] And so if you are going to work, then you should work on things that are really important. [02:59] Because you only get to live once. [03:01] And so I think like--

[03:03] I'm a believer, like, you know, doing the laundromat, like, you know, the classic immigrant, like laundromat or doing those jobs. My parents, those jobs are just as hard. [03:13] And every time people ask me, like as a founder, like, oh, it must be tough. [03:17] applied or whatever. [03:18] It's like... [03:19] Yeah, but... [03:20] You know what's harder? Like working at like Taco Bell. [03:23] Like that's a hard job. [03:25] And so I think like... [03:26] for new grads you have to have that perspective too which is like [03:29] there's going to be difficulty [03:31] even if there wasn't an AI boom. [03:33] There's always difficulty. [03:35] That's the part of growing up and finding your way and stuff like that. [03:39] So I managed to wrap that in like a more cohesive way in 10 minutes. An inspiring bow. Exactly. Yeah, yeah. Okay, so to bring it to current day, we're at the Rays Summit in Paris. [03:52] AI is buzzing all around us. It's like a big expo. [03:55] What are the trends that you're seeing in the room and around what's going on right now in AI? [04:01] Obviously, the broad stuff that we don't need to talk about are things... [04:04] the IPOs that are happening and kind of just how some companies are sucking up lots of revenue. Super positive. That proves that there's something there. It's not hollow. I think the... [04:17] below the surface which is like the rumblings is what's that reconciliation and [04:21] And then reconciliation in enterprise, the world that we are, we're an enterprise company, is, [04:27] like [04:28] and cost consciousness is becoming like a more core theme.

[04:33] um... whereas let's say if we're talking twelve months ago there was way more of [04:37] give me everything and I'll take [04:40] five of it. And now it's like [04:42] What are you giving me and how much I'm going to pay? [04:44] And I think that reconciliation will have downstream impacts. [04:47] on everybody. Not only the big companies going public, but also the small startups that are trying to find their little niche groups. I think that's... [04:54] a big talking, uh, talk, but in the physical AI world where we're at, uh, the, uh, [04:59] The way to think about physical AI versus, let's say, the large language model universe is... [05:05] the diffusion is slower because you're dealing with safety, you're dealing with physical machines in the real world. [05:12] And so it's way more linear. [05:14] so that you don't have these big spikes and crashes, there aren't new entrants that come in. It's just because it's safety oriented. [05:21] So that's a bit more of what, you know, when we last talked or in the last year, it's kind of continuing the same trajectory. So when we last talked, you had a lot of big announcements that came out. Since then, you've had even more. So what's going on at Applied Intuition? Yeah, I mean, one part of it is we are like a sizable company. So we have a thousand plus engineers doing a lot. So every month we're doing. [05:45] hopefully producing a lot of good new products that our customers are consuming. [05:49] But like at a high level in all of our major areas, our mission is to get intelligence onto a billion machines. [05:57] So you kind of deconstruct that, like how do you do that? [06:00] the obvious ones that people always think about us are cars and trucks because that's

[06:05] They understand that. [06:06] But similarly, we've had announcements in defense and where we are [06:10] We're working with shipmakers to put intelligence on the actual ships. [06:14] Huntington Wells in that specific example or we're working with Heidelberg materials to put [06:20] intelligence and queries. [06:22] and in ports and in mines. So [06:25] we continue on we had a bunch of announcements I won't bore the audience with them but you can go through all those verticals and we're just [06:33] continuing to push intelligence into each of those verticals. [06:36] and our... [06:37] are like macro hypothesis which is [06:41] you can take a model and put it on many machines. [06:44] and it can perform really well, I think continues to hold. Why we believe that's really important is it's very expensive to train and deploy these models. [06:55] And these are our models, and we're not like repurposing other folks. [06:58] collecting data, training, pre-training, we're talking big numbers [07:01] in the way everybody talks big numbers. [07:05] What's different about us is we want to also... [07:08] reconcile that with customer revenue. So we're always trying to balance that, which is different about us than other companies. There's a version where you're just like, "Hey, we're going to [07:17] spend a lot of money on this and we'll figure out how to make money in the future we try to [07:20] creates some balance there. [07:22] Isn't it right you barely touched any of your funding? Yeah, we're still in that phase, which is great. Which we're super happy about. It's less about... [07:32] like the ego of saying that. I think it's more about...

[07:35] We have a lot of resources to put towards [07:38] any opportunity that we think is worth deploying a billion dollars for or deploying multiple billion dollars for. [07:45] And so, yeah, and I think... [07:46] We have some big announcements coming in the next, like, [07:50] like the biggest announcements in the company's histories. Yeah, yeah. So I can't say it. [07:55] In my notes today, they said, whatever you do, don't talk about X, Y, and Z. Because internally, it's something we've been working on for a long time. A couple of things we're working on. [08:03] product side and the customer side. [08:06] you know I that's the way I was think about our company were a thousand engineers with some resources [08:11] Our mission is to put, you know, make-- [08:14] put intelligence on a billion machines. So how do we do that in the most effective way possible within the context of everything happening in the summit and everything happening in the competitive world, everything happening with our customers? Yeah, but I think one thing that is worth saying is like, [08:27] I think everybody who makes physical machines, whether it's defense companies or construction companies or automotive companies, they're not like... [08:38] Their head isn't in the sand about AI. [08:41] They recognize they need to put intelligence on these machines. And so that really works well for us. I don't know how you're going to go bigger than the last Physical Intelligence Day. You had Marc Andres in there. You had so many big announcements. So we're going to have to sneak into this one somehow. Yeah, exactly. Yeah, I think, I mean, I'm a believer that I'm pitching my own book here, which is like, I think, [09:03] the physical world and AI going into that, I think when we look back, I think that's going to be the bigger story. As I look back 25 years from now, I think we'll look back. It's kind of like if you look at the early days of the Internet.

[09:16] There were companies that were really focused on getting websites up, static websites up. So when you look back over the last 25, 30 years of the Internet, [09:22] you really think about Amazon you really think about Apple [09:26] you know so I think like those broader themes of like [09:30] phones getting in everybody's pockets and delivery becoming ubiquitous [09:34] those macro themes I think will be the big ones rather than like a specific model that's [09:39] launching and then the government saying no and then like I had all is gonna be forgotten so I think I [09:44] And that's where my mind is always at. On that note, I have to ask you, what is your hottest take right now? I think... [09:54] Oh man, the hottest take. [09:55] I think... [09:56] Thank you. [09:57] the cost structure that we're currently seeing from [10:01] whether it's from how much we spend on model training and development to employees to [10:07] We're at that phase of the bubble where it's like, you know, [10:10] All the opulence in the companies is what people talk about rather than the products. [10:18] I think there's going to be a little bit of a pullback. Maybe that's not a hot take, but I think it's coming. And a lot of times when people talk about this concept of, [10:26] of boom and bust. [10:27] The view is [10:28] The buses always takes longer. [10:30] than you think it is. [10:32] But I think the cycles move a lot faster. [10:34] So if there was really a bust in 21... [10:38] It's like, oh, well, you typically have a longer time horizon where you have a buildup, and maybe that pullback happens. [10:42] It's not necessarily a hot take. [10:44] But it's like... [10:45] I think about it.

[10:46] I think about like [10:48] So I'm just, you know, I would talk about books. I read, you know, a lot. And I just finished this book called... [10:53] the rise and fall of LTCM. The book title is When Genius Failed, but that's really, LTCM was this hedge fund [11:00] in the late 90s. [11:02] that was a bunch of these savants, geniuses, [11:06] And they... [11:08] Amen. [11:09] They kind of do what's happening now with Compute, which is like they just use leverage. [11:14] and they found themselves in a really bad position. And suddenly this... [11:18] untouchable firm. [11:20] so lauded in New York, in Wall Street, [11:25] And it collapses very, very, like, you know... [11:29] dramatically. [11:30] And like maybe because I read right before I go to sleep, like it's in the back of my mind. It's like, could there be an LTCM situation in our business? [11:39] where you have something that's so elite and hopeful and then suddenly like something so where it stumbles and it has [11:46] it has real repercussions. In the LTCM case that was because the broader global markets in Russia and in Asia [11:55] started [11:56] behaving in ways that nobody thought would happen. [11:59] and it compounded. So like, is there something that happens more peripherally in the AI business, and then it has a domino effect into our business? It's like, that's what I kind of think about. [12:07] a little bit [12:08] So, but I'm like, you know, I'm in the, you know, only the paranoid survive. So I'm like, I think you'd be asking me this question at any point in the last 10 years, I'd always be like...

[12:18] where is the hidden risk? [12:21] Like, that's like... [12:23] one thing I'd really implore founders to do. [12:26] is that you always want to be thinking about hidden risk. [12:30] in your company, in your leadership team, in your technical strategy, [12:35] in the market at large and how you play it. [12:39] I was going to ask you about the books. Yes. So what else are you reading right now? Like literally at this moment, and I always have to keep reading new stuff because if I get an interview, I can't be recycling books. [12:50] the non-technical books I'm reading, which are [12:52] interesting are [12:54] Society of Captives, which is about the American prison system, and then it's written in the 1950s. [12:59] Really good book. [13:00] Also, I recommend to founders to read stuff. [13:03] which is outside of our industry. [13:05] to give you like some [13:07] you know, [13:09] reflection on like... [13:10] How does this kind of apply here? [13:13] I'm reading The Courage to be Disliked. [13:17] Which I think... [13:18] people think [13:19] is hilarious because nobody thinks that I... [13:22] don't mind being disliked. So it's like, that's a great book. I'm reading, I'm also reading something else, and I just, it skips me. Genghis Khan? I just finished Jack Waterford's Genghis Khan book as well. Also fantastic, Making the Modern World, also fantastic. What were your biggest lessons from Genghis Khan? That's just a nice, like, popcorn book. [13:43] A better book that I just finished, I would really recommend, especially if you are...

[13:50] an enterprise watching this, like you're on the consuming side of AI and [13:55] It's a book written by John DeLorean, who is supposed to be the next CEO of General Motors in the 70s, or next president of GM. And he leaves, he created the DeLorean, the car that's in Back to the Future. [14:07] So he wrote this bombastic book called On a Clear Day That You Can See General Motors. [14:14] But then after he wrote it, he was like, this can't be published. Like, it's too crazy. [14:19] and I'm starting a car company I can't alienated but his co-author basically publishes it anyways [14:24] through lawsuits and everything. But it's a phenomenal book because... [14:27] At the time, General Motors was at the top of the business world. [14:38] It's like right now, like saying like if somebody wrote a bombastic book about... [14:42] SpaceX or Anthropic or something and saying like it's all screwed up [14:46] 300 pages of [14:48] why it's not good. It's written in the late 70s. [14:52] When GM is doing so well. [14:53] But the punchline is like, [14:55] those things actually came to be true. The reason GM actually struggled, it would be another 20 years. We're very clear in that book. [15:03] So if you're an enterprise, the reason I was just with the leadership at one of the big OEMs in Germany, the board and the CEO, all the most senior people. And I told them, you should read this even though it's from the 70s because I think large organizations have the same almost endemic problems, AI or not AI. It's like how leads work together. But anyways, long way of saying on a clear day you can see General Motors is a good book.

[15:33] some personal audits going on there. So as we close out, what are you most looking forward to? I, you know, it could be in any timeline. Some people say 12 months is too long. Maybe it's the next weekend. I don't know. What are you most looking forward to? [15:49] I mean, it's like to a hammer, everything's a nail. [15:53] self-driving cars is really everything. [15:56] Uh, and you know, you, you're, when I go around Paris, you, you see data collection vehicles as you're like, it's even coming here. Like, you know, it's just like not Sunnyvale. And so now like every city I go to typically with a trained eye, you can see like, oh, that's, that's a data collection vehicle like that. [16:15] Those are sensors that are not... [16:16] for self, to collect. [16:18] data to train models. And so I think that's all I see all the time. Yeah. That's fantastic. Well, Kaser, thank you so much. Yeah, thanks for having me. [16:30] This episode is brought to you by Brex, my favorite. You become what you spend on, and I refuse to spend my time on work that shouldn't exist. Expense reports, receipt chasing, and manual closes. The companies building what's next from Vercel, OpenAI, Anthropic, Granola, and DeepGram [16:50] all made the same call. They all run on Brex. Brex is the intelligent finance platform that combines cards, expenses, and banking into a single stack with agentic finance built in. AI agents that handle expenses automatically, enforce policy before spend happens, and close your books in minutes. That's why Sorcery runs on Brex, so I can spend time on

[17:20] That's B-R-E-X dot com slash S-O-U-R-C-E-R-Y. [17:28] Bye. [17:28] We're here with Dylan Patel of Semi Analysis out here in Paris. Dylan, how are you? I'm doing well, yeah. I mean, Paris is beautiful, but it's hot as... [17:39] Yeah. I don't know AC. Paris is meant to be enjoyed in spring and fall, not in summer. It's not an optimal place for data centers, right? I think it's mostly a regulatory thing, but yeah. Paris is, you know, they have all this spare power. One of the funny things is like, they're like, oh, okay, let's build a bunch of data centers. But then Spain, Italy, Belgium, the Netherlands, and Germany all freaked out because it's like, oh, wait, all this spare power. [18:09] It's actually like keeping our grid alive. If you stop selling us the spare power and you build data centers, our grids are screwed. So there's been a lot of pushback from the other EU countries like don't build data centers in France because we need the nuclear power. It's very interesting. So you are a conference surfer. You love conferences. I just listened to your interview with Sequoia on this. [18:31] he built semi-analysis off of riding the wave of conferences. [18:36] I'm very curious, what are the biggest trends that are going on at Reyes this year? [18:41] I would say a lot of people here are just trying to figure out what the hell is going on. And I think there's a wide array of people. There's a lot of people who are super deep in infra and they're making deals here. But then there's a lot of people who are just like, "I spent too much money on my token. What do I do?" Right? I spent too many tokens. So there's a wide range of people here.

[19:07] I think it's fun. It's fun that way because then you get, like, such a wide perspective of people. And you can walk up to someone and they're, like, a G and they know a lot of stuff and, like, a great connection. Then you walk up to someone and you're like, wow, I need to get out of this conversation as soon as possible. Like Molly, you know. Stop. [19:26] Oh, God. Okay. Well, anyways, let's cut to the chase. What is your hottest take right now? [19:37] for data centers, and they're just looking at like a backward looking view, right? Like what happened at the lab six months ago and what does it look like there? Or what does it look like now? Okay, I'm gonna build my three-year infrastructure with that in mind. [19:50] instead of like flexibility and general purpose, you know, capabilities. And then when they actually end up building at infra, it's going to be, a lot of it may be useless. Or not useless, but like less optimal than something that is more general purpose or more flexible. It feels like a lot of people are trying to optimize on the current rather than like think about where the workload is heading and then optimizing. And that's going to lead to a lot of wasted infra spend. [20:20] I think most people don't know what they're doing. They're just buying the NVIDIA stuff. But a lot of people are also just like they think they know what they're doing, and then they're buying and building and optimizing and customizing, and then ultimately they're going to waste a bunch of money. So it'll be fun to see where things head out and how they plan out and then what people do with all this infra that maybe isn't super useful with where models are in two years. Okay, we're going to run through some quick topics, okay? I want your take on all of them. Let's start with co-design. How do you feel?

[20:50] is the important thing, right? But when you try to do software hardware co-design without knowing where the software is headed, where the models are headed, then you end up in a pit, right? That's sort of the issue. But if you don't do software hardware co-design, you're also going to end up with, like, infra that is not optimized. So it's tough. [21:09] Open source first closed. [21:12] The theme here is that all, like, there's multiple Chinese model labs who are telling all the inference guys, hey, yeah, yeah, our next model [21:20] We're going to license it to you. So open is dying quickly, unfortunately. How do you think the costs of memory should trickle down? Should they go to the customer, or should players be taking that price on themselves? I'm a very big fan of trickle-down economics. Out of pocket. So trickle-down, I think they're going to trickle down a lot. I think you've already seen server pricing go up. [21:50] V300s that you're buying now versus last year, the price is significantly higher. We've seen next generation hardware receive price increases before they've even started production. Like, here was the quoted price, and now the quoted price is higher. So it's definitely going to get passed down. And we see that with cost of tokens, you know, not falling as fast as they were previously, because the cost of memory is flowing through. How do you feel about token

[22:20] I think anyone who doesn't token max is gonna get left behind. I think all this token budgeting stuff is loser mentality. If you're token budgeting hardcore, then your people are not gonna learn the new workflows and then they're not gonna reshape your company and make it more efficient and drive a lot more work with fewer people. So ultimately token budgeting is a fallacy. Now ROI is important, but ultimately... [22:49] The only way to get people to change their behavior is you measure them and you force them to do things differently. And token budgeting is like pulling back on them. [22:58] There's got to be some rationality. Don't just waste money on tokens for the laws. Don't give bonuses to people based on how many tokens they use or comp. But at the same time, don't just clamp down on people wholesale. Is it true you fired someone for using Haiku? I did not fire. I got mad at someone because their model was set to Haiku, and they didn't even realize it. They were like, Cloud Code's mid, blah, blah, blah. They still were trying to use a lot of tokens. They spent like $1,000 on Haiku. [23:28] what the hell? What are you doing? And they're like, oh, my bad. I didn't even realize they changed to Opus or Fable now. And they're so much more productive because they're not just like asking the model to do the same thing over and over and over again. Okay. Lastly, this is a pressing question. What is your favorite chip right now? [23:47] I've been a real big fan of the Takis Blue Heat. It's quite spicy, limey. It's good. It's good,

[23:58] is true or not, but chips in Europe have less seasoning than chips in America and Mexico. [24:07] There is this entire thing, right, where developing countries or less cultured countries are less flavoring their food. And the European idea is like, oh, our ingredients are high quality. And so I think this extends to chips as well. It's just horrible. You need more flavoring. More MSG, please. Amazing. Thank you so much, Dylan. All right. See you. Thank you. We have Mark here from Nebius. Mark, how are you doing? Fantastic. [24:37] How are the vibes at Raze? You know, the energy is real high. You can see it around us right now. [24:42] ton of attention, ton of participation. [24:45] Really impressed at how they pulled a global audience to Paris in the middle of the summer, in the middle of the heat. People are ignoring it. They're showing up, and it's been a fantastic experience. What are the biggest trends that you're seeing here? It's incredible. The discussion shifting from what our AI native is building to when our enterprise is adopting it. [25:03] So it's really incredible to actually have the discussion [25:06] where we're seeing the [25:07] Bridges being built to see enterprise adoption take place. [25:11] and [25:12] lots of discussion about uh... you know obviously the cap ex power and [25:18] stock prices, which I think at the end of the day is a bit of a distraction because this is a marathon that we're running and actually being able to drive that enterprise adoption, that's the critical next stage for the entire industry. Well, I can understand you guys have a lot of fans over there at Nebius and a lot of fans of the stock, too.

[25:35] So I want to ask you, what is your hottest take today? [25:40] How does Tate today... [25:43] And boy, that's such a wonderfully... [25:45] generous offer. Hottest take today is... [25:51] incredible opportunities in front of us that stay focused on the long-term [25:56] to not only [25:58] serve today's immediate needs, but how are we going to help to transform industries and how are we going to help to make enterprises successful with AI? [26:07] What is the biggest problem you're seeing with the build-out of data centers? The biggest problem with the build-out of data centers is something that Nebius is not experiencing, which is that a lot of folks are... [26:18] A lot of people are focused on single individual projects. Our organization has built a portfolio of opportunities. Today we have 20 data centers. We're multiplying that into the future. So [26:29] the biggest challenge many of our competitors are facing is [26:32] being single threaded which is strange when you're in a multi-threaded parallel processing industry and getting stuck with a single project. The challenges are that a lot of things are happening locally with different community challenges, regulatory challenges, [26:46] power challenges. [26:47] which everybody faces. Fortunately at Nebius, we're actually pursuing a broad, diversified portfolio approach. The beautiful thing about the data centers being created today is they have some of the most advanced technology data centers have ever had. So whether it's from new memory to optics, how are you guys thinking about that? All the time looking at the interesting innovation that's taking place, the fact that we can actually build

[27:11] from scratch, from dirt up [27:13] and be able to look at each layer of the [27:16] entire data center stack. [27:18] and be able to consider new... [27:21] methods, new technology, new providers. [27:24] puts us in a unique position to be able to take advantage of all those advances. [27:28] So we're always looking at the new and innovative things that are taking place to be able to improve the quality, the capabilities, the reliability of the solutions that we're delivering. [27:37] Amazing. Okay, as we close out, what are you most looking forward to in the next 12 months? The next 12 months in AI is an eternity. I mean, I've been at Nebius for 12 months, actually 13 months. [27:49] And if you look back 13 months ago, it was a completely different company. So the next 13 months is... [27:56] talking about [27:59] the [28:01] tens of billions that are being generated in nebius revenue. [28:06] and the extraordinary traction they were gaining with [28:11] not only the phenomenal AI natives that we're working with, [28:13] but also making our way to scale enterprise and big brand adoption. [28:19] and [28:20] being able to expand our solution set [28:23] and from the [28:25] model training that people are doing today into the inferencing that we started to offer [28:29] and into agentic workloads. [28:32] Fantastic. Well, thank you so much, Mark. Pleasure. I really enjoyed the conversation. We're here with Arvind of Glean. Arvind, how are you doing? Doing very well. How's RAISE treating you? This is a fantastic conference. I didn't expect what I saw here. What is the major trend that you're seeing that's going on here today? Have you been able to actually see anything? Well, I've not been able to attend most of the talks because there are so many good people at this event.

[29:02] my time. But some trends are clear. Number one, all the new AI companies are here. And the big trend, in my opinion, is open source and how that is fundamentally changing the full AI stack and bringing rise to a whole bunch of new companies that have new opportunities with open source. So what's going on at Glean? What's hot right now? Well, Glean has actually, we're finding ourselves in the midst of very, very good timing. So when [29:32] making it work in the enterprise. The two big things are context, like how do you bring all these agents that you want to automate the work that humans do. They need that context, that data, that information that humans use to do the same work. And that's actually something that we are really, really good at. So being the leader in context graphs is actually helping create a massive demand for Glean. [30:02] the business value coming from. And so Glean comes in handy on that front, at least from a bottom line perspective, because we do really, really good in terms of helping a customer reduce their token usage in two different ways. One, we can pick the right model. Since we work with all the closed domain and open source models, we can pick the right model for the right task, which is cheaper for them, but still gets the work done. And second, with our context graph, we can actually, when a model is trying to do some complex work,

[30:32] going to have to spend all this time just trying to assemble the raw materials to do that work. With Glean, they get that context in one shot. So we actually make most of your AI workloads much faster and very cheaper. What is the biggest challenge that you have been hitting recently? [30:48] A challenge, I think from a business perspective, it is finally businesses are running into this, you know, that, well, we're going to actually measure our spend on AI. And so that is creating [31:02] having so many different tools, so many different choices, you know, to bet on. It's getting confusing for them to pick, you know, who's the right partner to, you know, to actually start their AI transformation journey on. So that's been always, like, you know, the challenge in AI, like, you know, because, you know, like suddenly when AI became hot, every software company in the world became an AI company and it's confusing the buyers. So, you know, clarifying that doubt in them, like helping them understand the landscape, [31:32] and why is Glean relevant in this crowded mix of technology providers, that's always been the difficult task for us. All right, now I have to ask you a difficult question. What is your hottest take right now? Well, I think open source models are going to dominate AI inferencing. You'll see in the next two years, we'll shift to where there's almost no open source to where there's going to be almost all open source. Amazing. Okay. Arvind, thank you so much. Thank you.

[32:02] We're here with New York Stock Exchange, star of the show. How are you doing? Well, I'm better now that I'm with you. I mean, but this is kind of incredible, actually. It's my first time in Paris. I'm Laura, by the way. Hi. I run social for the New York Stock Exchange, so I like to think I have the best job in the world. You might have the best job in the world, but luckily, being at the New York Stock Exchange means I get to work with you a lot. So it's kind of a win-win. But, yeah, this is kind of incredible. First time in Paris. [32:32] I'm in Paris. Yeah, I get the whole hullabaloo. I get it. It's kind of cool. Paris is amazing. Yeah, it's pretty amazing. Also, it's hard to beat New York in my eyes. Tried and true. So I understand the lore and the love of the Louvre. How about that? Whoa. When I was on my way over here, I was like, you know what? I could commute. Straight up. The plane wasn't that bad. I came from L.A. I just slept the whole time. I could probably just commute here. Hop, skip, and a jump. Yeah. And we have the match tonight. I'm very excited. [33:02] Morocco so there might be riots we're gonna see and you will see Molly and I on the pitch no kidding we are not going no that's in the United States right all right Laura so you have been around raise you've been seeing all the companies come in and through and around the New York Stock Exchange booth you've done media yourself what is something we should all be paying attention to what is the trend of the day hottest thing happening here at raise I would say folks

[33:32] day, eugenic, cloud, quantum, all of our favorite words. I believe the thing that's making these companies stand out is their access and visibility. It's their partnerships. A lot of those partnerships are happening here. We're seeing it in real time, watching real C-suite founders, entrepreneurs, actually meeting in person to close deals. It's actually kind of crazy to see it with my own eyes. You only hear about it, I think, in closed doors and golf courses and karaoke bars. But [34:02] I firmly believe that it is about the story you're telling, how well you're telling it, and where you're going to go from there, how it's going to set you apart. And that has a lot to do with what we're doing, right? Like, there's a reason why NYSE and NYSE Word are the main media sponsors of Raise. Hello. There's a reason NYSE is one of your sponsors. And also, Raise rang the bell. Come on. Come on. [34:32] that like sets people apart. And if you can stand out and if you can, you know, cut through the noise, I think there's a lot of noise, honestly. And I think if you have a clear message and you have a clear goal with really good people, that's what helps with being in person and stuff is you meet the actual real good people. Talk about vibe coding. It's all about vibe, you know? Vibe meeting. Vibe media. Our new podcast, Molly and Laura. [34:58] Amazing place to end it, Laura. Thank you so much. Thank you so much.

[35:02] We're here with Apoorv from Altimeter. Apoorv, how are you doing? Good, glad to be here. How's race this year? Things have definitely upgraded since last year. [35:13] Yeah, really good. It's big time, big guests, big speakers, big companies. Great. Like really good setup. Lots of great founders. Really excited to meet a lot of the CIOs. I got to know the CIO of Goldman last night, CIO of Procter & Gamble. [35:27] and where they are on their AI journey has been the biggest learning for me. [35:31] I've been waiting to ask you this question for quite some time because you've gone viral a lot recently with your Stanford lectures. So, Apoor, what is your hottest take right now? Well, today of all days, I'm going to be a little bit of a challenge. [35:46] There's four seasons in AI. They're called OpenAI, Anthropic, [35:52] SpaceX and Google. [35:53] So if you're a CIO, if you're a CEO, if you're about to make a multi-million, billion-dollar decision of which lab to build your intelligence on, plan for the climate. [36:05] not for weather. What does that mean? That's multi-model routing, that's evals, that's thinking about post-trained custom models. Yeah, that's out of stake. It's four seasons of the year. [36:24] to plan for if you're about to spend a hundred million dollars you don't want to get stuck [36:29] in one of the seasons. Sometimes the summer is long, sometimes it's a heat wave. But I think you want an evergreen season. And I think that evergreen season, the climate looks like you'll have a portion of your workloads going through...

[36:42] Um... [36:43] open source, as Jesse from Decagon wrote very eloquently, a big portion of their volume goes through open source, about 90% now. And then the frontier stuff, you know, discovering new use cases will always go through the most intelligent models. Coding might stay there for a long time. [37:03] Planning for that multi-model world, planning to not get stuck in one regime is the climate. [37:09] For people who have not seen your lectures just yet, highly recommend it. [37:14] Who have you spoken with? I know you brought on different guests, and you bring them for the students. What was the plan there? Oh, you know, we had a great lineup. It's a labor of love, and I was really excited to bring back some of those conversations back to school. You know, if you're a student and you're making a big life decision, I wanted to make sure that you saw the whole AI stack from chips to data centers to models and ultimately the AI applications. And so we had a great set of speakers across all of the parts of the stack. [37:44] folks from NVIDIA and GROC. We had Ali Gozi from Databricks. We had Doohan from BaseStand. We had Suchin from OpenAI, the folks from Anthropic. So it's a great lineup and we're very lucky to have them. Amazing. Okay. To close out, what are you most looking forward to this year? Oh, wow. I'm looking to the seasonal changes in the climate this year. Well, you guys saw what happened today. Saul from OpenAI goes live. SpaceX's Michael Truel just launched a model at his opus level. [38:11] you know it's it's it's it's what a great time to be alive and and and watch the seasons change amazing anything else we miss anything anything you want to share to come later in the year more to come later in the year thank you so much support thank you we have nakil from turbo puffer nakil how are you doing so well love being here in paris

[38:30] A little hot, a little chaotic, but, you know, we're telling people to puff. [38:34] so we have to address the rumors you don't sell [38:37] Puffer jackets. No, those are internal only. We do have puffer jackets, but only for a select few. Now, if you go to tpup.supply, we do occasionally drop new merch there, so stay tuned. Oh my gosh. Okay, so let's get into it. For people who don't know Turbo Puffer, [38:57] What is it? [38:58] So we are a search engine optimized for AI workloads. So we power retrieval for Cursor, for Notion, for Lagora, for Anthropic, some of the biggest names in the business. And what that means is if you think of a product that has a search bar, [39:11] we want to be the thing powering that search bar. We are the thing powering that search bar in many cases. But for any agentic application, they're searching on your behalf behind the scenes. [39:20] So you might be asking for a report on sales. [39:23] the first thing that agent is going to do is search Turbo Popper for everything mentioning sales. [39:27] and get that all into the context window of your model. [39:30] Amazing. Okay. [39:31] Given that, what are you seeing as the biggest trends here at Raze? So much data. I've been talking to folks. We used to think in terms of gigabytes or terabytes. Folks are coming up to me with not just one, but ten or hundreds of petabytes that they want to search over. No way. Now, I'm not sure everyone is prepared for the cost of searching over 100 petabytes. There's a bit of a negotiation. What is the cost on that? [39:54] So our largest workloads cost tens of millions of dollars to search over right now. And those are web scale use cases with tens of terabytes.

[40:03] Searching petabytes is going to cost another order of magnitude or two on top of that. [40:08] And what is the architecture behind it? So the reason that this is viable at all, the reason folks are able to do vector search over the entire web on TurboPover, is because we're based on object storage. So that's S3 or Google Cloud Storage, which is rock bottom storage prices. The absolute cheapest storage you can get. The problem is it's really slow. So our innovation is we had all this caching on top of S3. [40:29] so that you see performance that's almost equivalent to what you get with a system that's not based on object storage. [40:34] but you get something much closer to the economics. [40:36] of object storage. [40:38] Damn. [40:39] Okay. [40:40] Are you ready for the hardest question of the day? I'm so ready. What is your hottest take right now? I think search is still too expensive. [40:47] And Turbo Puffer was founded because we thought Search was an order of magnitude too expensive at the time. We brought it down by an order of magnitude. [40:56] But it kills us that there are still products out there that are limited in their ambition because search is still too expensive. So we're always looking for ways to bring the cost down even more. [41:05] If you think about it, if you're spending [41:07] $5 on turbo puffer per user, but you're only charging your users [41:12] five dollars a month, the economics just don't work. But if we can bring that cost down by an order of magnitude, so you're only spending fifty cents a user on searches, [41:20] Suddenly you're able to build a product. [41:23] where you couldn't before your margins work, you get crazy growth, you explode. So that's what we're always looking for. What product isn't currently in the market because search is too expensive? [41:31] Amazing. So what's next? [41:33] more relevance

[41:34] We started with vector search, we added keyword search, we added regex, [41:38] But what people actually care about is, are my searches good? [41:41] They don't care about... [41:42] Did I get 95% of the results I should have gotten for this vector search? They care about, is my agent performing? Is it completing the task successfully? So our task is, how do we get the best possible search results fed to these agents? [41:56] So they actually complete the tasks. [41:58] So we're taking more and more of that in-house. There's a bunch of new techniques like [42:02] late interaction like search agents. [42:04] that we're going to build into the product. [42:05] so that you as the user just-- [42:07] You puff harder. You send more queries to Turbo Puffer, and you can trust that you're getting good results, and you don't have to worry as much about the mechanics of search. [42:14] Amazing. Thank you so much, Nikhil. Yeah, thank you for having me. [42:18] If you're building what's next in AI, you need to know MongoDB, the database platform developers love and built for the agents you're running. MongoDB stores searches and reasons over your data in real time with vector search and embeddings from Voyage AI all in the same system. No separate pipelines, no stitching together 10 different tools. It's why 75% of the Fortune 100 and leading [42:48] to billions of vectors. Go to mongodb.com slash AI to learn more. That's mongodb.com slash AI to learn more. [42:59] Bye. [42:59] So we are in the middle of chaos here at Raze. It is very popular this year. And we have Barack from Wonderful AI. So, Barack, welcome. Hey, Molly. Good to see you. It is absolutely chaotic here in Paris at the Raze Summit. I'm Barack. I'm from Wonderful. I'm the chief strategy officer at Wonderful. I'm actually the first institutional investor in the company that joined because I believe in it so much.

[43:22] We're an enterprise applied AI partner. [43:25] to the world's largest enterprises outside of the US? [43:29] We're both a platform and a partner to them. So what is your hottest take right now? [43:34] I believe geographies will be even bigger than verticals. [43:38] when it comes to AI, when you look at actual labor and labor displacement and what's actually happening in the world and where it's likely to go, [43:45] countries are way more interesting from an expansion perspective and a focus perspective. [43:50] than verticals. [43:51] and that you can go horizontal without a vertical specialization with AI, as long as you're able to solve the go-to-market problem. [43:58] strategy with it. [44:00] - Damn, we are back in Uber Land Grab competition stage. - That's exactly right. [44:05] like the enterprise applied AI playbook with the Uber go to market strategy. We're in over 30 countries around the world. [44:13] and pretty much the last six months that we've expanded to most of them. So it's quite a wild ride. So what is the secret to Wonderful? It's really the focus and the strategy. I think generally in the AI universe, the opportunity is massive, but one of the hardest things for companies is how to actually differentiate and to break out from the pack. What it requires is a non-consensus insight. [44:34] that you need to actually be right about. And I think for Wonderful, that was really the geographic focus, to go generalized, to go horizontal in terms of the use cases that you can actually offer for enterprises, but to actually focus on [44:46] rest of world versus US, that was the non-consensus part. The rest of it around the challenges with applied AI and what you actually need is very similar to the way most of us probably think about it, but it's the playbook that you bring to it.

[44:57] and local delivery that was... [44:59] most non-consensus about wonderful. What's your favorite country? [45:02] Wow, great question. I live most of the year in Tel Aviv, so I'm biased towards Israel. But to visit and that I've had the pleasure of visiting was Japan. Really? We opened up an office and I was there about a month ago. Tokyo is just, I mean, Japan in general is like a completely unique world, and especially as it relates to the enterprise playbook and how they do business and the respect culture and everything. It's fascinating and I love it. Amazing. Great place to end. Thank you, Brock. [45:32] Lovely to meet you, Molly. Bye, everyone. We have Max from Legora. Max, how's Raze for you? It's very warm. The electricity went out yesterday. I keep getting pestered by customers, but I think that's a great thing. [45:47] And... [45:48] It has been a fun time to come to Paris, I think. [45:53] I've done a few of these in the US, in London, in the Nordics, and it's always a bit of a different flavor, but it's been great. [46:00] So you're based in Europe. Yes. How is the Europe vibe right now? Well, I think the vibe [46:09] in Europe overall versus companies like Legora is quite different. [46:13] I think the general vibe is [46:16] There's a fear that some of the best frontier models get locked up in [46:20] the states [46:21] And there's a lot of discussion around how do you continue winning in manufacturing and these types of markets that are important.

[46:29] And then you have really fast growing software companies like Legora, where we are [46:33] adopted a global mindset from day one. And so we don't really view ourselves as only European. I think we view ourselves as very global. And we have been competing at the global stage from day one. And so we still wake up every day thinking about what can we do more, what can we do next? [46:48] which is really thrilling. [46:51] Did you expect this level of growth? [46:55] I think going into this, I didn't really know what to expect. I think I've had a very unique experience fundraising and working in our market that most startups doesn't have. [47:08] Typically you don't raise 600 million dollars in two weeks. [47:12] But we got some great advice from Y Combinator, which is if you build a great company, it's easy to fundraise. And if you build a great product, I think it's easy to work with customers and to deliver value. [47:24] No, I did not expect this, but I think that we are leveling up and we are... [47:31] you know, [47:32] meeting the opportunity when it's now being presented to us. [47:35] So legal AI is one of the fastest growing categories right now and it's super competitive [47:42] How are you dealing with the token situation? So we were actually the first legal AI company to move into consumption-based pricing. And I think that's really important because [47:53] 8.1 just continues to solidify that we're the product leader and innovator in our space. [47:57] But secondly, it really aligns the value that we bring with the way that the business model works. So if you have a flat rate,

[48:05] and you say, you know, $200, $300 per user per month, [48:10] and then they use $2,000 worth of tokens, you have a problem. [48:14] And so we've been running at a very positive gross margin for a long time, and this will help us continue to do so. [48:19] And I think it's... [48:21] It helps our customers already see where their tokens are being spent. [48:26] and how do you think about that long term? [48:28] because token consumption have been the business model in coding tools since inception. [48:33] It's the pricing model of the big labs. [48:35] and now it's going to be the pricing model in legal, and I expect many of the other companies to follow us into that. [48:40] Damn. [48:41] Okay, I have a really hard question for you right now. Are you ready? Soon. What is your hottest take right now? [48:49] My hardest take is... [48:52] that there is [48:53] A lot of complaining. [48:55] from European startups [48:57] rather than just [48:59] locking in and [49:02] building for the global stage. I think there's a bit of laziness. It's nice in the summers here. [49:08] to go to Italy or to go to France where we are today. [49:12] But if you want to build the biggest companies in the world, you need to look past that. And you need to understand that you're competing with the U.S., you're competing with China. And if you want to win globally, you need to work as hard as they do. [49:25] Sorry, you can't go to Saint-Tropez. No, sorry. [49:29] The great lock-in is year-round. Yes, that's exactly it. [49:32] Amazing. Thank you so much, Max. Thank you, Molly. We have Gil here, CTO of Merge, which we recently did an amazing episode with you guys at the New York Stock Exchange. It was great. How are you doing, Gil? I'm doing great. Happy to be here in Paris for the conference. Okay. We have one burning question that I know you've been waiting for. Yeah. What is your hottest take right now in AI? I think some people are starting to agree with this.

[49:57] Token maxing? Not working. [49:59] you're getting zero results from using more tokens [50:01] It was cool to encourage your employees to use more AI, but I think what we're seeing now is [50:06] We're not getting more output. The only thing that people are seeing a real connection between, I guess, sort of [50:11] productivity and usage of AI is how you use AI to bring your cycle times down, get feedback, and iterate really quickly. [50:17] I love asking you this question because you are so technical and it is a fun fear mongering topic. [50:23] So what are the biggest concerns people should be watching out for with integrating all these APIs with all of these new AI applications? [50:32] I think mainly it's security. I think down market, who cares? You're going to let your own personal data flow. I do it for my personal consumption. [50:39] But when it comes to businesses, [50:41] They're just afraid of data flowing out of the system, so integrations are the point. [50:46] where an LLM actually becomes dangerous. [50:48] something isolated, I mentioned this before, but something isolated, an LLM, [50:52] It might insult you, but it's not going to do much worse than that. But the second that thing can send your data elsewhere, that's where all the problems come in. [50:58] So I think we're just going to see a lot more governance and locking down before we see AI really opened up to the masses to talk to all your systems. [51:05] I think Alex Karp recently went viral for this rant of AI sovereignty within your organization. You agree? I absolutely agree. I, it's just, it's, there's never been a good... [51:16] position in business to hand over, you know, the fate of your, your, [51:19] your future, your development to another [51:22] company. So I think people are going to want sovereignty, they're going to want control over their own systems. Whether that means, you know,

[51:27] hosting in-house or using third parties, they want the flexibility to switch whenever they want, they do not want to be locked in. [51:34] Amazing. Well, Gil, we will let you go, but what are you most looking forward to at Rays this year? [51:39] Um... [51:40] Honestly, I think what's been so cool, I know this is probably an untraditional answer, [51:45] but the mix of sort of this [51:46] really classic French architecture, and these statues from the year 1500 mixed with all the AI, and it just... [51:52] It's such a cool juxtaposition. I'm loving being here. [51:55] Perfect place to end it. Thank you so much, Gil. Thank you. [51:57] Ariel, welcome to Sorcery. I'm happy to be here. We're here at the Illuminati. Yes, actually I've never been in such a bizarre place to do an interview, so that's cool. [52:09] So we're fresh off stage. We were just talking about Navon and how you guys are dominating travel. What is your hottest take right now with travel? I think the hottest take is what we talked about, how humans are still so important, both for meeting and being here, but also supporting all of this. It was actually a cool outcome of this interview. [52:28] I mean, it is true. We were talking about this, but hallucinations. You can't have an AI agent do everything, especially with travel, because it's [52:36] so, so complex. [52:39] So how are you at Navon helping solve this? It's so, so important. Like think about it. If I send you to the wrong flight, [52:45] I'll tell you that I upgraded your flight and we didn't do it. Like, I'll send you to the wrong room. [52:51] People are so, so emotional when it comes to travel. They care about it. So you cannot have any fuck-ups. Halluzination is a huge, huge, huge fuck-up.

[53:01] We've built our own platform, our own model to prevent that, and we are supporting most of our customers by using our own AI platform that actually does not hallucinate. Before you've had a customer come over to you, I would assume, what has been the worst travel story you've ever heard? It's actually my travel story. This is the travel story that started Navan. I was one of the first ones to start an offshore operation in Ukraine, you know, Dessa, Ukraine. [53:30] And I went there. [53:32] arrived there in the middle of the night [53:34] freezing cold by the way I hate the cold freezing cold came to the hotel because of some credit card issue [53:40] The hotel canceled my reservation. [53:43] and they told me to go down the street and find a new hotel. The travel agency didn't pick up, the software didn't work, [53:50] I walked with my luggage down the street, one hotel after the other, until I found something. But this was by far my worst travel experience and really what inspired me to start the company. [54:00] So you've been public now for almost a year? Almost a year, like nine months, yeah. That's pretty incredible. So you guys are also ripping. So what's going on there? What are the latest stats? [54:09] Yeah, we grew last quarter by 50% in terms of usage. [54:14] we grew our revenue by 40%, we became cash flow positive, [54:18] became profitable so our kind of a stats are amazing but the most important thing [54:23] We have more and more users using us. We have more than $10 billion of bookings. [54:27] and you're now going really really fast [54:30] So more people are joining this kind of Navan thing.

[54:33] Amazing. Well, thank you so much, Ariel. And I think we closed out on stage with Bullish on Humans. Yes, and I think that's the most important thing. It's the most important thing to figure out because nothing matters. [54:44] do all of this AI infrastructure, the AI fund stuff, [54:47] But we need to remember that we are all humans and this needs to stay. [54:52] Perfect. Thank you. [54:53] Okay. [54:54] CJ, welcome to Sorcery. - Great to be here. - Great to be in Paris. - Great to be in Paris, it's fun. - You're fresh off the stage. You're very popular today, by the way. I keep hearing your name from everyone. - Thank you, thank you. - So what excites you right now? [55:09] Number one, [55:10] It's about... [55:12] customer-focused innovation, [55:14] and making sure we are the foundational layer for all [55:19] AI applications and agentic applications. So that's number one. Number two, [55:24] making our customers really successful [55:27] as they roll out AI and there is clear ROI that they can see, [55:31] and they have a peace of mind [55:33] when they use us as the data layer so that's really exciting and third is just the pace of innovation [55:39] We want to move fast for our customers. We do these things called .local conferences that are very local, hence the name .local. Next one is in San Francisco. [55:50] We want to make many announcements there. After that one is in New York, we want to make announcements there. Then it's in Mumbai, we want to make announcements there, which are all related to innovation. So the pace of innovation... [56:01] in service of our customers really excites me.

[56:04] I have to ask you a very difficult question. Yes. What is your hottest take in AI right now? [56:10] My simple answer is [56:13] which I believe to be true is [56:17] Data is the unsung hero and data is back. [56:21] You cannot create an AI application without a great data layer, and your AI application is as good as your data. As the kids would say, facts. Yes, facts. And it tracks. Well, great. So you heard it here. Big data is back. Thank you, CJ. Thank you very much.