Prime Intellect raises $130M at $1B valuation to build open sovereign AI stack

Speaker 2: I'm extremely excited to welcome our next guest to the show. We have Vincent Weiser from Crime Intellect. He's the cofounder and CEO, and he has some amazing news for us. Tell us what happened, then we'll dig into the news, the market, the story, the company. But first, give us the news because Jordy's already warming up over here.

Speaker 4: Amazing. Yeah. So we've announced that we've raised a 130,000,000 and 1,000,000,000 with Prime Analytics Wow. To build the open super intelligence stack.

Speaker 2: It's fantastic. Congratulations.

Speaker 1: Yeah. It it What I what we were talking about before the show Yeah. Is is the the the the ratio between dollars raised to run rate. Yeah. Absolutely

Speaker 2: incredible. Yeah. Right? Much have you raised before this round?

Speaker 4: Just 20,000,000 total. So I think we've actually got to this run rate on, like, less than 20,000,000 in spend. That's crazy. Which

Speaker 1: is north of which is north of a 100.

Speaker 4: Yes. That's amazing. Yeah. We we doubled the pricing since we actually closed the fundraise Yeah. To a 100. So basically, yeah, on track to grow much more from here on. Okay. So

Speaker 2: reset me on the latest and greatest in the actual product offering. What customers are coming to you? Is it I have used a frontier model to, you know, figure out that AI can in fact do some agentic workload within my organization? I wanna cut cost and you're gonna RL an open source model for me, fine tune it, and then am I paying you for tokens? Am I are you getting me GPUs? What is the full suite of products that you offer customers these days?

Speaker 4: Yes. So basically, it is kind of like the full stack to do training, deployment, and continue improvement of models. I think customers basically come to us for anything ranging from compute to inference to the full RL and post training stack to to train models. And I I would say it's like, with like customers coming from different buckets and categories, a lot of them are like the AI natives who are basically starting out, and and oftentimes also like being multi model. Right? Like, they they use like the frontier models. They use open models Mhmm. For different use cases. Like, good examples like RAM, for example, as one customer more in this like AI native bucket, a lot of the big AI native startups. But then we've also increasingly been able to track a lot of the new AI labs, like a lot of the Neo labs. So a lot of those are using our full stack from like large scale clusters to our pre training and post training and RL stack. But then also increasingly, like traditional enterprises and even sovereigns coming to us who want to have their own end to end stack that they can run like, on prem. So how I see this is, to some extent, it's like, there's just increasingly, I think, like a demand for having your own, like, sovereign AI stack. Right? It's like your own open end to end stack, but I think this is almost complementing, in many cases,

Speaker 2: the other, like, frontier, like, models and and stacks. And you're using Sovereign in not just the geopolitical sense, but also if you're a Fortune 500 company, you want to own your stack from start to finish, potentially. Exactly.

Speaker 4: Yeah. I think that there was a lot of recent talk about this. Obviously, like folks like Satya and Alex Carr were like, also making making those those points and arguments over the last few weeks. And I think it's increasingly something I think where, like, a lot of, like, enterprises want to basically build this, like, compounding data mode and flywheel, where ultimately, I think the future that we are starting to see is a lot of these companies want to build basically self improving agents. Mhmm. And the way to do this, I think, is generally you you have an RL environment for use case, you scale on it, and then you ultimately deploy that into production, and then continuously improve with the user interacting with that agent. It's almost like the the Tesla autonomy, like levels, but like for knowledge worker agents. So I think like the the ramp example I think was one that people really resonated with for basically them being able to within a week and less than 50 k of of training spend Wow. Outcompete a frontier model Yeah. At automating spreadsheets and and finance at a fraction of the cost of of running the cheapest models. Sure. This is like they were able to outperform Opus at a fraction of the cost of Haiku and and like 30% faster, like much more tailored to the use case. So I think in many ways, like, these, like, specialized agentic use cases are a big unlock. And I think in many ways, you still need the god like a big mega model to to do the orchestration or planning. But then I think oftentimes execution, I think, happens can happen with these specialized sub agents. So I think, yeah, those have been extremely, I think, like, fruitful and and and useful for a lot of the companies adopting and and training their own agents. Then I think the other big trend is just like an explosion in in neuro entrants that are, like, taking model training seriously. Obviously, hundreds of Neolabs really got stood up in the last probably twelve to twenty four months. So we started powering a lot of those, like, with our full stack, including with also large compute clusters. So, like, NVIDIA also joined us as part of the round, and we've been like, we've since day one basically operated large clusters. And and this is in in large part why we need so much money. Right? It's like to basically be standing up more and more GPUs. Like right now, we run like 15,000 GPUs and scaling to like over 30,000. Okay. And so so this is, like, a lot a lot of the capital needs even. And

Speaker 2: walk me through when you're pressing the huge cluster button. Is that because there's broad demand for specific models that you need to inference at scale, and so there's just a lot of inference demand? Or is it that for the level of training that you're doing, you need to be cluster scale to even kick off the training run?

Speaker 4: Yeah. So basically, it's like for for all the needs, like, from training to inference, you need obviously compute. In a lot of ways, like, our customers, like, are doing a lot of these, like, large scale training runs with us, like, leveraging our tools, like, including our compute. Right? So it's like pre training, fundamental and parameter models, like, for what we did with the RC, like Q4 last year, or really large also post training runs and RL runs. Right? So it's like increasingly basically, we abstract all of that info away. So basically, people can just, like, hit a button Mhmm. To do a post training run without having to, like, log into GPU cluster and and managing the bare metal GPUs. So we basically abstract a lot of this, like, GPU complexity away for but but yeah. Like, for that, we basically also ramping up our own clusters to basically just, like, serve hundreds and and thousands of customers, like, in parallel on those.

Speaker 2: Yeah. So what you're saying is, like, there might be a situation where I have a I have a 50 k budget like you gave with the ramp example. I have a 50 k budget to do some post training, and and I need to deploy that 50 k into, like, thousands of GPUs, but only for a short amount of time. And so Exactly. The budget's actually going to be relatively small even though if I were to buy all of that, it would be it's way more than 50 k, millions, millions of dollars. So you sort of absorb some of that cost, orchestrate it, abstract it away, and then you can deliver for the customer.

Speaker 4: Just a slice of it. So so basically, I think there's this, like, also big unlock, like, also on the inference side, right, where customers don't need to, like, front load hundreds of millions and and CapEx spend for large data center commitments. Yep. So we basically also increasingly saw a lot of inference. So it it really is the full stack, and I think a lot of the compute like, we we have, like, quite a a clear, like, visibility into the demand just because, like, people come with us with large requests or or large contracts to basically do larger training runs and deployment. So we we basically just, like, need to secure the compute to, yeah, to be able to fulfill all those. And I think, like, right now, we would probably be at, like, many orders of magnitude more compute like compute, but also revenue if compute wouldn't be as constrained as it is right now. So it's like right now, I think, like, in the in the current environment, it's like, basically GPUs are pretty sold out Mhmm. For the next few months. And for almost like the rest of the year, especially if you want to like stand up data centers in in The US. So I think this is actually like I think the the the main bottleneck to to scale right now for anyone in AI. And this is I think what we've been able to really scale through extremely well because we've basically partnered with almost every data center out there since, like, over two years. So basically, you're able to send up, like, data centers in in different geographies and orchestrate them globally.

Speaker 2: Yeah.

Speaker 1: Any predictions, forecasts for American open source over the next six months?

Speaker 4: Yes. No. Really great question. I we joined also, like, NVIDIA to help them train Nematron. So they're doing the Nematron Alliance and Coalition. So we are one of, like, think 10 or so partners. So it's like so like like, I can't I can't speak about some of that like all the things there, but I think, like, in general, both on that front and other things we're saying We won't tell any we won't tell any.

Speaker 1: Wait. Wait. Anything Just between us. Between us. Just the three of So

Speaker 4: basically, I think like well, like, in some ways, like, leading American efforts actually are Nemotron and actually like Aussie with the Trinity models. So both of those we're like very deeply involved in. And so I think those will continue to to like lead like with like American open source. Like we have like some of our own models cooking. And and then, like, we help a a ton of customers over there. So I think there will actually be, like, a huge resurgence of, like, Frontier open models coming out of The US, also led by players like NBA itself and a lot of our customers and partners. So, yeah, I think, like, in in general, I think the interesting geopolitical question is actually what happens with Chinese open malls. Right? It's like like, obviously, we've seen news that, like, China might consider, like, export restricting them, like and and and, like, there might be policy dimensions on on both the Chinese and The US side. Right? So it's like, I think this is, I think, actually, in some ways, the motivation to build a stack and and to also push American open source forward is to have, like, less reliance on malls that might get banned any day. Be it all more close. Right? It's like I I think this is something increasingly, like, a lot of enterprises are starting to to realize that, like, there's actually huge dependence, right, in the sense on like, mainly from the policy side. So this is something I think where we want to just, enable everyone to have, like, their own end to end FrontiOpenStack to be able to create their own models. And ultimately, think like there there will be like there's obviously a lot of like now, like also Neolabs and and LoftMRI actually like thinking about things quite in a quite open way. So I think it will be surprised that this may be one of the hot takes is to see more Neolabs release open models. Like a good example that we also partnered with recently was a poolside, for example. People were expecting them to be closed, but they actually opened up their models. So I I think there will be more of those, especially given there are, like, hundreds of well funded Neolabs now. Yeah. I think there will be huge resurgence just like of, like, the age of research in terms of, like, these nail labs, like, pursuing truly novel directions. Right? It's, like, going where, like, the big labs almost can't go necessarily, taking big, like, bold bets that also might not pay pay off all of them. Right? Or, like, if you have 100 shots on goal, like, some of them will probably pay off and result in interesting outcomes. So I think this is, like, really also, I think, what we are empowering, right, is is kind of, like, helping these new nail labs and and big AI natives and enterprises to move much faster and not have to basically all build the same stack each by themselves in house. Because obviously, a of them are like tiny teams of like tens of people instead of like thousands. So we're able to like help them move much faster basically, But I'm adopting our sec. When

Speaker 1: you hear of a, you know, major American tech company, let's say, someone in the mag seven that has extra capacity, do you reach out to them? Do you get do you do you do you ping them? Are you at the scale yet that that a deal would be interesting to them? Or are they really searching for these, you know, $10,000,000,000 plus opportunities?

Speaker 4: For sure, like, we already are, like, sourcing, like like, pretty large clusters, so that is, like, sometimes, like, up to, like, five to 10,000 GPUs. So it's, like, we we basically, like, looking at every source and pocket of supply that we can find. Like, we've literally turned every stone to find, like, the last tribute available on earth. And, like like, we we've we've been quite good at at that. So it's like in in some ways, like, I think we're, like that that has been, like, one of the core tenants almost, like, since the beginning is that, like, we we really didn't want it to, like, orchestrate, like, the global compute supply efficiently. So we did a lot of, like, optimization fault tolerance and this retraining even in the early days to, like, train across, like, distributed compute. So it's, like, definitely I I think to this question, it's, like, like, I think it's something that we're definitely, like, considering. Like, if there's any anyone with some spare GBUs here, you can always set us up.

Speaker 2: That's great. Alright. Back on on China, do you agree with my perception that if there is some sort of lockdown on Chinese open source export controls, that would not affect

Speaker 1: Yeah. Are they mad that American companies are distilling on or or post training

Speaker 2: their open source models? It was by design. That was goal of the project. But I what what I'm what I'm interested about is is would they try and claw back access to Kimi or GLM, any of those models that are already been deployed? Because there's an element of, like, once the the weights have been downloaded, it's sort of just out there. Of course, you could go and try and, like, sue every company, but that's very hard if you're in an American jurisdiction and you're a Chinese company. But how would you see that playing out? Would it just be going forward, no more open source, and that's the CCP telling Chinese companies or some other shape of of of, like, restriction?

Speaker 4: Yeah. I think it's really hard to know. I think, like, a good example was, like, in some extent, like, the most popular Chinese malts already have been closed for Nommo, is coin. Right? Like, in some ways, like, they've actually, like changed their licensings Mhmm. And stopped almost closing. So I think there will be both, like, companies changing their policies, right, but then also I think like, probably like state intervention, right, it's like in different shapes or form. Mhmm. So I think basically, one can't rely on the steady stream of frontier open, like, releases out of China in many ways, but it's, like, brought from different perspectives. And I think, similarly, a lot of other sovereigns, like, obviously, including The US also doesn't want to build on on necessarily Chinese op mods. It's like Mhmm. I think a lot of American enterprises, obviously, also, like, sovereign use cases, like, need American op mod, but I think similar for for other continents. Right? It's like, I think there there's been a lot of work, obviously, and and we're, like, collaborating with people also across Europe, across India, across other geos. So it's like, I think we'll we'll see a much more, like, multipolar, I think, open, like, ecosystem of, like, open frontend models. So I think that's, like, a healthy, like, balancing away from just relying on Chinese open models, even though they've obviously, like, been a great accelerant of, like, open source progress. I think it's also unhealthy to rely just on that ecosystem. Yeah. And, yeah, like, I think we'll be like, there will be, I think, a bunch of, like, quite unpredictable changes similar to all. Like, the last few months have been quite unpredictable Mhmm. I think, to everyone involved. So I think this will continue to be the case in the sense that it will be, like, quite a dicey policy like, AI policy landscape for the next few years. Mhmm. Last question.

Speaker 2: Data centers are data center construction, new data center construction, deeply unpopular in America domestically. Simultaneously, the workloads that are being done in AI data centers are basically latency is basically irrelevant. For training, certainly, for most inference tasks, if you're waiting a couple minutes for a result, an extra five hundred milliseconds to get across the world isn't gonna This feels like a recipe for global competition, for aggressive data center expansion in a country that does see it favorably, that does see an economic boom from it, or does have a lot of energy. What is the state of the non domestic, non US data center build out?

Speaker 4: Yes. So, like, we're certainly, like, building a lot of data centers and, like, partnering with a lot of them, like, outside of The US as well. But, like, I think there's still a lot of demand, actually, for data centers in The US. Think, like, even from a data perspective, like, there's a lot of now, like, the financial industries and and highly regulated industries, like, also adopting, obviously, more and more agents and and compute. I do think inference still matters. I like speed in a sense, like, maybe not for, like, long running agents, but, like, for a lot of use cases, people, I think, still really care about, like, having extremely snappy, like, responses. But I think, like, there there's definitely, I think, like, a general boom in data center build outs outside of The US as well. I think there's just, a limited set of countries who are, like, strong, like, allies with, like, ample amounts of, like, like, geopolitical stability, but then also, like, like, energy and and and and green like like, green fields and and enough space.

Speaker 1: Yeah. US has unique access to natural gas Mhmm. Which is if you're trying to bring energy online super quickly,

Speaker 4: it's hard to beat. Yes. And I I think, like, we we're seeing this across Europe even where, like, there's specific regions, right, like, are, like, much more active in data center build outs than others, and and those are generally the ones that, like, have their own, like, money internalized on them and, like, cheap energy and and a lot of, like, also local support for it. So I think, like, in general, from what we are seeing, I think there's, like, there there is new entrance almost, like, to to the data center build outs, where it's, like, in places like Australia and New Zealand, like and and, like, a small much small place, right, like, Armenia, Kazakhstan, and others, who, like, have a huge amount of energy and and huge like, and and making huge bets also on data centers by a while also being, like, allies of The US. So I think those are the countries, I think, generally, which will see the biggest boom in data centers. I think at the same time, like, I think we'll also continue to see see a boom in US data centers. But, yeah, there there's definitely it's definitely harder to get data centers in The US than outside of The US, and and so it's where I think we'll we'll see both continuously expand. And then I think there there might also be, obviously, like, your point, like, policy, like like, updating for like, what was the upcoming elections and and in midterms, I think, I think, like, this this whole almost, like, populism in in The US around, like, data centers and, like, a lot of the the kind of, I think, public sentiment as as potentially also, like, sovereign influences and other things, like, from other nation states who who are trying to harm almost, like, The US data center build out. Yeah. So I think there's, like, a lot of things at stake for The US, right, it's, like, to to not pull back on on their data center build out. But I think there will also be specific states who are, like, leaning in much more heavily than others. Right? Like, I I think this is already what we're seeing where, like Texas. Exactly. So it's, like, there's specific states in The US that are just, like, much more well suited to really ramp up data centers from here. Mhmm. And yeah. And I think we'll we'll see them ramp up globally and and hopefully also outside of, like, beyond Earth. And I think Yeah. Like I think this is the only way to really probably scale.

Speaker 2: I love it. Well, thank you so much for coming on the show. Congratulations.

Speaker 1: Incredible progress. We love seeing you and the team win. Yeah. It's fantastic.