Baseten’s CEO on why every company will want to own its AI

>> To end for people who don't know what AI inference is beyond, it's how you serve the models. I'm really curious to hear you explain it because it feels like the most exciting frenetic part of the AI stack right now, and you're thick in the middle of it with base 10. I'd love to know what it means for people who, yeah, they know the term, but they don't know much else. >> Yeah, absolutely. >> So, what is inference? So, inference is the way you serve models, quite often, customers will come to us, they say, "Hey, we have this model, we want to use open weights model, or we have post-trained this model. We need to figure out how to actually use it in our applications." And so, when you work with a closed frontier model provider, you kind of get this to free. They have the model, but they vertically own everything from the model all the way to the chips it runs on, all the way up to the APIs that you are using to call that model. When it comes to post-trained models or your own models, or open weight models, you don't get that for free. And so, you'll come to an inference provider and you say, "Hey, I need to run this in production. Here is some scale, approximately how I need to run this at." And then you'll come to a company base 10. Now, there is so much that goes into actually running a model. And for us, that's the infrastructure-level problems. So that is like, "Hey, where is the actual compute going to come from to run this model?" So we take care of that. There is the orchestration piece, which is, "Hey, how do I get this user request to the right chip with the right model on it?" We solved that problem. And then there's the chip level details where it's like, "How can I make this model run very, very fast?" And so, really, when it comes to inference, you're really like, you're not just looking for an API, you're not just looking for GPUs. You're looking for like a performance reliable system that puts it all together and puts it behind an API and base 10 kind of takes care of all that. All right, that was kind of marketing. So let's double click on that. Let's actually double click on what does that actually mean. So, you know, base 10 is like three different layers of problems that we're solving for our customers. So the first one is this idea of infrastructure and compute. So you need chips to run these models on. So for our customers, you know, if you're a customer running this model at a decent scale, you'll probably need thousands of chips to run this. If you go to the market right now and say, "I need 1,000 P200 to GP300." - Good luck. - Good luck, it's not happening. Not happening as they're gone. Someone's already taken them in. So you'll come to like what we do is that we kind of take care of this compute procurement from the standpoint of what you need that is necessary to run inference on for our customers. We do this using like a pretty complex architecture where we're able to pull together compute from all sorts of different regions and clouds. So we sit on top of 20 different clouds and 90 different regions, which is completely abstracted away from our customers. So that's the bottom heterogeneous compute layout. - I don't even know the word of 20 different clouds. - Yeah, exactly. Well, you call them clouds, neo clouds, data centers, hyperscalers, you call them to us. They are, you know, sources of compute to some extent. - Yeah. - Okay, let's go up one layer. That becomes a core inference layer and that's kind of what I described, which is like, hey, I care about running these models really, really fast and really, really reliably. And so that's kind of like what base 10 is done for a long time now. That is what we are known for. And our customers come to us not only saying here we want the compute, but here we need the whole software stack on top of that, on top of this. So we can run this or it doesn't go down. We can run this or it's very fast. We can run this with the right numerics and so on and so forth. That's the second layer. We can talk as much as you want about that. The third layer is around all the primitives that come on top of that. And that's kind of like, you know, these are inference adjacent primitives a lot of the time. What I mean by that, these are things that either make your inference more powerful. They give you different parts and different types of inference. They make your models more powerful, but they're all related to inference in some way. So that might be something like post-training models and RL on top of those models, which not only helps you train more models, but has a bunch of inference baked into it. It might be something like sandboxes, which is how do you actually execute code? - We're gonna get into that. - Yeah. - How do you actually execute code generated by these models? It might be something e-values and routing. When people talk about inference, they're kind of talking about all these problems together, all these problems together. But there are lots of different parts to it. - It's everything except training, right? It's serving AI in production. It's when you get a chat GPT response back or you use MUSE or GROCKBOT or any of these things. They're all inference. And inference is growing like crazy. It seems like as a market. But it also feels like it's hard to make sense of it. There's a lot of players. They're all approaching it in slightly different ways. And I'd love to like better understand how base 10 differs in its approach from not only like your direct competitors and inference, but the traditional clouds. 'Cause I think someone listening or watching this who is maybe not as in the weeds as we are would go. Well, why would someone not just use AWS or Azure or Google Cloud to run this stuff? 'Cause they have AI. Why do we need these AI clouds doing inference? - It's a good question. Look, I think the most important thing with all this is that the market is just ginormous. - And this is important for a reason. But if you kind of project forward four or five years and some silly number associated with how much AI spenders, let's call it $5 trillion, what they say in five years from now. It doesn't matter what that number is. It's somewhere between $2 and $10 trillion. Let's say AI runs about like a 50% gross margin. So that means there's about one to $5 trillion of inference spent in the market. And if you think about how that's gonna be dived out between hyperscalers, between closed frontier model, between inference brightest like base 10, any which way you cut it, that's somewhere between 10 to $50 trillion on market cap that's gonna be divided between the people who are capturing inference. If you then start to break down how different folks are approaching inference, everyone has a slightly different take, I'd say. So there's the hyperscalers who have historically been really good at infrastructure like things, but they've also been really good partners. So I think the hyperscalers will have inference solutions and they do today, but they're also really good partners to the inference companies themselves. We have really strong relationships with all the folks you mentioned there. The reason why you'd use us over hyperscaler, really a lot of it just comes down to building specialized software. And it's like, hey, we do one thing. They do many things. They're like, when you're in a market that is moving as fast as this market, where you need to run software production and when your inference provider is down, your product is down. You would generally tend to who will go, who will move as fast as you in the most customer-aligned way and give you the most attention. That is base 10 today because just 'cause we give you, we do fewer things and allows us to be best in class at those few focused things that we really do. Yeah, and I think it goes beyond that because I think there's just so many layers of this problem that I described. And we're kind of solving it layer by layer there, whereas I think a lot of other folks and like, you know, whether it's hyperscalers or whoever else are like, kind of like, hey, I've got some basic inference solution that works off the shelf, but it isn't really dialed in to serving users in production at scale. This episode is brought to you by Mercury, AI native banking that's loved by more than 300,000 entrepreneurs, including me, visit mercury.com to learn more. Mercury is a fintech, not a bank, check the show notes for details. Thanks also to granola, the AI notepad for people in back-to-back meetings. It works everywhere you do and lets you focus on what matters. Try to granola.ai/sources and use the code sources for three months off. This episode is also brought to you by Jira Byad-Lasian, where teams and agents get the context, coordination, and control to move work forward. Try it free at Jira.com. That's J-I-R-A.com. The criticism, maybe, of the marketer end, is that these inference clouds, they're just reselling GPU capacity at a very thin margin, and they really have to prove the software layer still. And you just said the software is what matters, and you have an acquisition that you just announced that I want to talk about today on that. But can you explain that a little more like why the software matters so much? Because as you said, the chips are hard to get, and if you've got chips, you see the base 10 ads in San Francisco and it's on the billboards, and it's very clear, it's like you're saying I've got compute, then that seems valuable, but at the same time you're saying, well, the software is actually what matters. Yeah, the compute's necessary, but the software is what differentiates, right? So it's like, I don't want to downplay the role of computing or computer obviously, like the market-containing thing right now, at the same time, at the same time, there is so much value to be driven to be built on top of that. The way we think about this is like, if you go, if you wind back a few months from now, there's been this really big push towards owned intelligence. And what is the idea of owned intelligence? The idea of owned intelligence is like, hey, because of cost control, because of data sovereignty, because of, you know, credible permanence, which is the idea that you don't want models just to disappear on your one day, is very, very important for every company in the world to some extent to own their intelligence. Cool, was that mean? Well, what that means is that they need to have the capability to not only run open source models, but an open-way model. models in addition to close frontier models, but to figure out, like, when to use which, how to use their data and the user feedback from their customers to improve these models, what customers have realized is, like, inference is one part of that, and how to argue is probably, like, the backbone of that, to some extent, but there's all these things that you need to do to unlock that continual learning loop that makes customers kind of internalize their intelligence. So that might be inference. That might be how do you eval, eval each of these models that they come out. It might be how do you post-trained models with our own environments to make these models better. It might be like, hey, the execution environments for when these models are producing code and doing their own things for them to do this work. And then eventually, it's like, how do you get a better model out of it all that you run again and kick off this whole cycle again? That whole, like, the guy just described you is what we call like a continual learning loop to some extent. And that is what the software layer that we have, like, that's the software stack, like, the software stack we are building is all those things I just described to you and all of them in the entirety, which in a lot of ways become their own hype scale or new cloud when we build it. And compute is necessary underneath that until you have that thing, until you have that loop working, the compute is kind of useless. Or it is exactly what you said. It is reselling of, you know, taking access to it and putting it somewhere else. But the minute that we build the software layer for that continual learning loop, what we are doing is, like, providing every enterprise, every customer in the world, some credible advantages they are getting from running their intelligence. And what that does for them is it protects their margins. And as a result of them protecting their margins, they willing to pay a lot of that margin to us because we are kind of making them independent, again, to some extent. That's interesting. I haven't heard continual learning framed in that way. I mean, when you spend time in the labs, you know, the frontier labs, they're talking about continual learning in the sense of the models themselves learning and getting better constantly and the research loop closing and being aided by AI. But you're talking about it in like a sovereign company AI sense of controlling your own destiny, which I haven't heard before. Yeah, well, I mean, and it's exactly that right. This idea of owned intelligence, like Satya has talked a lot about the gents has talked a lot about this. Yeah. You know, which is like, hey, your user data, your signal from these models, the improvements to these models themselves will become the core IP of every company in the world. I think every enterprise, like, the extent to which you thrive in AI first world is the extent to which you own all those things. In order to own those things, you need to own that loop that I just described here. And like, that is what we're building for. And like that is like, you know, you don't want that all that data to be the input to someone else's continual line loop, which is the loop you, which is the loop you described about how, how, how the model provider companies get better at trading, trading their models. Yeah. And like, like that, that to me is the software stock. That is the differentiate value that you're providing. That is when this gets a lot more interesting than just bare metal compute. Does it feel to you like it does to me, where in the last like eight to 10 months, it just really feels like AI has started becoming useful in a way that it was not before. I mean, I think that, you know, post-catch EPT, obviously like that product took off and kind of kickstarted this whole boomer end. But it's was this kind of like better chap bought Google search things for a while. And I mean, even in my own work, both, you know, as a, as a podcaster, but, you know, investor now too, like it's utility is incredible. And I'm curious if you've been feeling that as well, because you started base 10 and what, like 2019, which was well ahead of all of this. The models were not capable at all. No company would be running LLMs in production, even in like 2021, 2022. So, yeah, I want to go to the founding story and how you saw this when you did. But also, do you feel that what I'm describing about the shift in AI right now? I know something in December that happened where these models describe how to capability shift. I think it's happening again right now, like in the last month where there's another capability shift. But the last like six, seven months have been insane, right? Because so much has happened, like obviously felt like coding models family got really good. The computer shortage happened. Openweight models started getting very, very good. And like, you know, it's like release up to release up to release. It's just like, oh, there's another one. There's another one. There's another one. I mean, and I think what has happened as a result of this is that one, people are paying attention to intelligence. So intelligence, people using intelligence and, and enterprising you using intelligence as skyrocketed as a result. Earnings calls have become about using AI. Companies are being judged by how much they're spending on AI. That in turn is like making them, you know, rethink, hey, what is our long-term cost right? You know, how do we, one, you know, get AI in more places and two, how do we do it in some way that doesn't fall up our companies? And I think that is just taken over the entire narrative. But also just pushed adoption even further. It's kind of, it's wild. It's wild. And, and I can't imagine being inside it the way you are. I mean, before we started this, you were saying, I may need to step away because I'll get paged because one of our customers needs help, because everyone's growing like crazy. Before we get into more stuff, go back with me to, to 2019, you're starting this company with your co-founders. You were in banking before. Is that right? How did you see this because this market didn't exist then? Yeah, so my background is that, I grew up in Australia. I moved it for uni, graduated uni in 2009. I did what everyone does in 2009. It's like, oh, you know, you go working banking. Yeah, that's right. So, so I moved to New York. I worked in banking. I worked in infrastructure finance. Basically doing a project, financing for infrastructure projects like toll roads and bridges and airports, which is, which is interesting. They're all data center people now, funnily enough. After that, I started working technology. I ended up working on a lab in Boston, working on using machine learning to predict and diagnose neuromuscular disease. That was in Japan 12. From 2012 to 2019, machine learning was very, very early. It's kind of more classical models than more classical models than large language models. If you go think about what machine learning was being used for 15 years ago, it was stuff that like PayPal was doing. It was stuff that Facebook was doing and then the law was like fraud classification. It was recommendations and like we were kind of working in that paradigm for a long time. I've started a bunch of companies. I didn't really go anywhere in 2019. I started another company and we were really thinking at the time, like two things. One is like, you know, my two co-founders of Phil and Amir and then soon after that punkage. But Phil and Amir, you know, they're my best mates and so really the idea was like, how do I how do I set a couple of my friends? I was number one, number two, number, number two, which was just as important was this idea that in 2019, we kind of knew that machine learning was going to be a big deal. Like opening it out and being around for a few years at that point, doing stuff like, you know, smart people were going to work there. But it was still very early. Like there's no commercial proof whatsoever. Like machine learning looks like it's going to be pretty, pretty big. We don't know what it's going to look like. We had no idea. You know, we're just like, you know, I like, you know, that's why I'm like so, so in all of all of our customers, because I think application layer work is so challenging. Like actually, you know, actually understanding that end user problem in a domain requires so much product industry than domain expertise that frankly, we just didn't have in that way. But what we knew was a mission, we knew by, by drum machine learning, we thought it was going to be a big deal. Like up, well, if we believe in this that we know a bit, let's go to build a picks and shovels business alongside, to support machine learning. And, you know, as long as the market, the market is big enough, we'll have a shot of building something big. I think that what happened in 2022 with the markets accelerated in a way that none of us expected. And then like quarter after quarter, it just feels like, you know, we're like, surely this is it, right? And then, you know, and then it takes another turn. I was chatting with some folks that, you know, you work with a guy in an FE and Jill about this Monday and basically this idea that even 2022 and, you know, they started thinking about machine learning and AI, like it felt really early. It didn't feel like it was obvious than either. And it just feels like quarter after quarter, it's like, it just come online in a way that we, like, it's kind of gone beyond all our wildest streams. And so in 2019, when we started the company, we were definitely building a infrastructure company. I think that infrastructure, that infrastructure company had many different facets, one facet of it, which was seem to us the time with the least important facet was being able to serve models. And then that just became the entire company. And you raised at five billion valuation in January of 2026. And then 13 billion valuation in June. So in six months, you almost tripled your valuation. What changed in those months? Is it what we've been talking about? Yeah, it's all those things, right? Which is like, I think in the last like six months, what has changed is that we've gone from thinking our open weight model is going to be a thing is like to, oh, wow, this might be a majority of token volume. You think open will be that. I open and custom and owned intelligence will become, I think, I think the future is mixed intelligence where you're using closed frontier, using open weights models using post train models. But I think custom frontier models will probably like be some version, some percentage of the models that are uniquely served to needing the most powerful models in the world at all times and you think about right now, think about what AI could solve three months ago versus what AI could solve now and AI could solve a lot three months ago or two months ago even. If you can run that two-month-old intelligence for 90% of your tasks at a fraction of the cost and own all that data sovereignty and ownership narrative that I talked about earlier, that's a no-brainer and it's very rational. How are you feeling about the state of open source, though, because the U.S. is so behind China is still dominating. I think a lot of people, a lot of tech leaders are worried about what's happening and a bunch of COs are recently advocating for the government here in the U.S. to protect open and open weights and open source. But yeah, I curious what you make of what's going on right now. Yeah, look, I think it's both a very exciting time and an interesting and challenging situation, right? Look, we want more intelligence everywhere and I think open weight models are fundamentally a very important part of the ecosystem. In absence of open weight models, all that ownership narrative that I just told you about is gone. You need it. There are necessary preconditions. We understand their importance to the independence and sovereignty of American enterprise. That's what I was saying, one, two. I feel really good that there are options now, so that's amazing. On where they come from, there seems temporary to me. I think there will be an American open weight model. I hope, from what we can tell, there will be really good American open weight models in weeks, not even months or quarters. Soon enough, there will be really good American open weight models. I think once that flywheel kicks off or we have stuff that is competitive with Chinese open weight models, I think a lot of the narrative that you're seeing right now, which I think is a little bit more fear-driven and I'll be falling behind driven than there's capability driven. That makes sense, I think that will kind of wash away as well. I'll give you a good example of this, where all the open weight stuff really kicked off 18 months ago, when DeepSeq, when DeepSeq V3 came out, which was a fantastic model came out over Christmas of '24. The first thing we were obviously serving that to a bunch of customers, but the first thing I did was I was like, "Hey, we've met so many enterprises over the last six or seven years." I was like, "Let me go and talk to 10 of them right now and see how they feel about them." The reality was that I'd say 90% of them, if not more, were like, "Well, we would never use this production." We don't know whether they come from, we don't really understand how this works, whether they've been trained on, yada, yada, yada. I'd say they were DeepSeq V3 was interesting because I think it was the first time you had something that would approach the frontier, but it was still behind. It was still like 386 behind and it wasn't clear how you'd run them and there was all this uncertainty around what it was most of like. Even though it was behind, it was such a big moment everyone was freaking out. The air race has been reset, yada, yada. It was like the paper that they trained it on like a fraction of a cost and blah, blah, blah. Fast forward a year. Now you have all sorts of models, you know, you would have some really good American models like Inkling, but then you have GLM, you have Kimmy, you have DeepSeq, and hopefully a Neymatron, the Neymatron family thing, and really good from Nvidia as well. All that narrative from 18 months ago is completely evaporated. Every enterprise we talked to now is very excited to run open source, open point models. I think even debating where they're coming from isn't really relevant. I think this is just the new world and turned out that fear and certainty and doubt with more just capability driven as opposed to actual, what does this mean driven? So the minute they got good enough, everyone's like, "Can we use them?" And the infrastructure needs to exist, open point models need to exist to run them, but we're in a really good spot right now in terms of owning that entire stack that's necessary to be able to run this model's reproduction. I think as that stack matures, and obviously we are to some extent building that stack. We are building that stack until that stack matures. And as that stack matures, I think the barrier to adoption of our boy model is going to go down. Is it going to go down as well? Does open have to keep growing for base 10 to keep growing? Are you really kind of tied at the hip with this? I've got the good question, the intellectually honest answer is yes. There's probably people who don't want me to say that, but the intellectually honest answer is yes, which is like we believe in the open weighty ecosystem and we want to get back to it and push it as well. We launched something called base labs a few weeks ago, which is our research lab, which is like, "Hey, how can we make it?" We acquired a research team, and then last year, they're doing really good work in which it's like, "How do we put all this stuff out as well?" But we are fundamentally behind the open weighty ecosystem and the extent to which I don't think open weight models need to exist, but I think customers need to want to own their own intelligence. And that needs to be a idea and concept that matures aggressively, and I think open weight models are just an accelerant to that. So does that make sense, which is like a core ingredient of that? And does base labs signify that you're going to start doing your own models, base 10? If we need to, we will do whatever is necessary for customers to own their own intelligence. We post trained models extensively with customers today. So what that means is that we take a base model and then we take their data and we set up their learning loops to be able to have those models around them in production. Today, we are not doing large-scale pre-training. It's folks like Nematron who are doing a lot of that stuff and we're building and we're part of the Nematron Coalition and we're doing work with them there. Look at some point, if it's like, "We feel like we have a differentiated advantage to training models ourselves," and now the compute situation starts that it allows it. 100%. I think that is just our way of giving back to that ecosystem that we believe in when we try to power. Mercury is a modern take on banking, built for startups like mine. When I decided to start my media business, Mercury was by far the most straightforward, full-featured banking solution for me to set up quickly. The interface is intuitive and simple, saving me valuable time every day. I use Mercury to track my spending, bills, and invoicing. I love that I can delegate permissions to my team so they can keep things running for me and exactly the way I want them to. My favorite part is how forward-looking Mercury is with AI. Legacy banks are stuck in the past, but Mercury is built for how modern software works today. I use its built-in command assistant to analyze cash flow and help me move money, and Mercury also connects to other AI tools like ChatGPT and Clawed. I use this feature all the time, and the folks at Mercury actually let me know that I'm one of the top users of it. So trust me. It's finally easy to get real-time financial data about your business wherever you need it. Visit Mercury.com to learn more and apply online in minutes. Mercury is a fentech company, not an FDIC-insured bank. Connecting services provided through Choice Financial Group and Call M&A members FDIC. I spend a lot of time context switching between meetings, often with no time to process one before the next starts. Thankfully, Granola runs in the background the whole time. It's an easy-to-use AI notepad for meetings that works everywhere, even on phone calls. I use Granola to recall what was set in meetings and create helpful summaries. I use it every day to stay on top of what I need to get done with my team. It connects to my email and suggests follow-ups for me to quickly review and send, saving me valuable time. Granola isn't just a core part of my workflow, it's basically my second brain. Try Granola at Granola.ai/sources and use the promo code "Sources" for three months off. AI is only as useful as the context it has. But when that context is scattered across tools, threads, and DMs, your team, and your AI agents are flying blind. That's the problem Jira by Atlassian solves. That's the goal tied to your project. What got decided last week in Slack DMs, Atlassian's teamwork graph pulls all of the valuable pieces together. From Jira, Confluence, GitHub, Slack, and more, so nothing falls through the cracks. You get 44% more accurate results with 48% less token usage. With Jira, you can easily share your work context with AI agents you already love, like Claude, Cursor, and GitHub Copilot. Assign them work directly or connect your tools through MCP. All of this lets you spend less time digging through endless links and messages, chasing down what got decided and by who, and spend more time actually shipping. Learn more at Jira.com. That's J-I-R-A dot com. Framer is the AI website builder that brings agents into the same canvas where your website is designed, managed, and published, so you can move faster without giving up your taste or control. Framer powers the sources podcast website at podcast.sources.news, where you can find new episodes, transcripts, and a lot more. Learn how you can get more out of your site from a framer specialist or get started building for free. Today, at framer.com/sources, for 30% off a framer pro annual plan. That's framer.com/sources, for 30% off, framer.com/sources, rules and restrictions may apply. It feels like everyone who has a stake in the compute build out is going vertical. You've talked about in video, which I know they're an investor as well, but they've got Neematron now. They're doing models. They just bought hugging face. They're going more vertical in the compute stack up and down. It feels like everyone is realizing this is such a big market. It's so important to control your own destiny that you've got base labs. You're thinking about the model layer and how base time can play there, not just at the compute and the software stack for inference. I'm wondering do you agree with that take? that kind of the incentives of everything right now are driving companies in your position to try to own as much of the stack as possible. Yeah. It's a good question. I can just speak for ourselves, like, look like the reason why you own more and more of the stack is for two reasons, right? It's either to deepen customer value and, you know, ideally, you know, add more value to them at the end and hopefully scale off some of the top for yourself as you add that value. Like, that's one place. The other one is like, to the extent that you feel blocked by the rest of the ecosystem. The reason why, like, you know, we, we think about going, like, building up and down the stack is like, you know, we build up the stack again, that's all the primitives on top of it. First, I talked about that third layer because we think, um, that is how we accrue more value and add more value for the customer and that we, we own more of that learning loop. That's like a very value driven thesis. I think we go down the stack because, you know, we need to be in charge of our destiny and we need to know that, um, you know, we, we, we think that in 12, like 12 to 24 months, we will be, well, you know, one of the largest, largest individual users of compute on the planet, like, you know, like that, that is, that is the way that, that is, you know, there will be the labs and then there'll be based in, you know, like, that is the way that this is trending in order to make sure that we can continue to grow the pace that we want to grow at. We just need to at least believe that we have control over thumbless things that sit down through them of us or upstream of us, how do you want to think about it? And like, that is why we voted, like, vertically integrated that way. Can you give me a sense of the scale of base 10 today? You mentioned you work across dozens of clouds. Can you talk about revenue, growth, custom number of customers? Yeah. Anything that kind of puts a, a weight, so what you do, what's the scale, what's the scale that we can do? Like, you know, 40 to 50 trillion tokens a day. That's a lot. That's, you know, bigger than a lot of them. How much were you doing six months ago? I think, I think, I think token volume has grown 40x year and year. Wow. Yeah. It's kind of insane. And then, so that's one, two, it's like, you know, we have thousands of customers, I'd say, you know, the way to think about base 10 is that, you know, if you have, if you have used any of the application layer companies, and we've probably touched part of your workflow today, probably the last year else. And it's like, tokens that you have consumed have probably run through base 10 through all the apps you use, through all the, through all the products you love. And then, in terms of revenue, like, I don't know what we've probably said, but it's like, I think we have 10x in the last 12 months, like, it's like, you know, it is everything about our revenue about our business is just kind of multiplying as a market is accelerating. And like, that's, it's a weird thing, it's a weird moment, the way we're very, very blessed to build in this market. And then, obviously, there's a ton of like macro tailwinds that are continuing to push. And speaking of that, a trend I'm obsessed with right now is the rise of these agents using virtual machines and browsers for people to get things done. And I think, you know, very, at the consumer level, there's mues which metageship that had Zach on the last episode talking about that instinct is taking off in Silicon Valley world. There's startups like town. And then there's Grock Bott and Cheshire PT work, obviously. But the idea is you take a model and you give it a harness that includes a computer and a browser and the ability to log in and do things. And most people in the world who use AI have not experienced this yet, right? They're still on that chatbot paradigm of before. But this is really starting to take off and it also implies that token volume continues to go through the roof, I think. And you just made an acquisition in this space. So I want to understand that and how you think that is going to start intersecting with what you do at base 10. Yeah. Look, I think this just becomes, you know, these run times in the execution environments for these, the place where agents do work. That's the way to think about it. And like for us, like this is like our first version of like scientific build stuff, specifically for this idea of like agents, you know, like like instinct, like town, like Grock Bott, like where they're going to need to do their work. And we've been thinking about this space for a long time now. We met Paul and Blackswell and the entire team and we were just blown away by the technology that they built. And really the idea was was like, hey, I don't think inference and sandboxes are different problems. I think they are extensions and related, like, you know, where these models run is where the execution environment should be. And that makes a lot, that makes a lot of sense to me. And then we start to think about, hey, like, it's so interesting when you start to think about all these problems together because, you know, a lot of these sandboxes, all these execution environments run on CPUs, not GPUs. Right. Then you start to think about that own infrastructure thing again. It's like, oh, should these GPUs and CPUs be pollucated? Should they, you know, how do you, and like, really, we start to think, you know, it isn't both, again, the extension of the value that I talked about earlier for our customers, where it's like, you know, where the age, if inference is primarily going to be driven by agents in the future, we're also going to need to understand where the agents are doing their work and communicating themselves. And that comes into the execution environment that comes to the browser use and that's where we start to think about flexible and how that fits into based sense, very, very natural. But then we start to go down the layer again and think about the layer of heterogeneous computing. It's like, how does that change when these two things are collocated? And like, that is why this is such an interesting company, I think, like, it's so complex. And I think, you know, I just want to bring this back to the first question you said about the competitive environment, like, inference is a very, very, very special workload and it needs, like, new hyperscalers and new clouds to be built around it. And it's like, that's really how we think about not only what we do, but how we are acquisitive and how we join forces with folks like Baxel. Do you think that billions of people are eventually using agents with virtual machines? Like, do you think this gets really big? Yeah. I mean, you kind of see, you know, with, like, instinct and, you know, like, and love to instinct. I love it. I use it. But it's, I don't think it's huge yet. I don't think it has millions of users. Oh, no, no, no, totally, but you could see it, right? Which is like, you know, we've been trying to solve this, do work for me problem for so many years. Like, think about how many chat bots there are where like, you know, can you make this reservation for me? Can you pay my taxes? Can you, can you, can you book a flight for me? I think this fundamentally changes how we do work. This is, this is what's so interesting, which is, you know, I think about our customers, our customers a lot. And I think about customers, like, open evidence, which is like, you know, basically bring intelligence to the fingerprints of, of, of frontline physicians or bridge, which is like making, you know, healthcare providers be more present by like, you know, being the ambience cry abroad than making them take notes on the end time. All I think about, I feel it's like clay, who are like kind of changing how good a market teams, good a market team. Well, go even what the texture of good a market looks like, or you obviously think of stuff like cursor, which is like coding or I don't know if you're familiar with, but flow, whisper flows. Awesome. Yes. But we'll look around our office in Jackson Square. I see a bunch of engineers whispering into their microphones to their agents. Yeah. And what I see with all these things is like, we, in a lot of ways, like tech and like the community that we are in are really, really early adopters, but you see the impact is having on healthcare. You see the impact in having on the legal field or with support with companies like Sierra and Decagon and Brandon, so on and so forth. There is just no way that it would make sense for all those productivity gains and the changes in how they work to not flow to the consumer experience. I think like instinct and town, and these are just like the first example, these are the first examples of that. And I think, I think like, you know, could I see my, my, my mom and dad being, you know, massive instinct users or town users or, you know, 100% and I think that that's where things get really exciting for like how much more we have to, like, again, this goes back to the first thing, you said, it's like, I'm just about to get like 100,000 X bigger, like that's a crazy thing. It's like the amount of penetration we actually have in the market right now. It feels like, you know, one percent or two percent, not, not, you know, not some saturation point. Is it scalable, though, if hundreds of millions, eventually billions of people have agents or multiple agents running computers in the cloud, don't we just need a lot more data centers? Like, is this actually scalable? Yeah. Yes. IVS is there. Yes. We need a lot more data sets. We need a lot more compute. But you also can remember that we're, we're like, we're aggressively like lowering the cost of inference as well. And so like, I think both things are true. One, like, we need a lot more inference optimization and two, we need a lot more power to show on land and, um, to do data centers and we'll get more efficient at building data centers as well. But the good news is that, you know, that a lot of land is like, you know, I, I, I, I, like the country that we're, you know, I grew up in is like, you know, like the majority of it's unoccupied. And, you know, the, the so much space in the US, we're obviously dealing with stuff. But, um, you know, we are, we are, we are thinking about and they're really interesting people thinking about how we create energy in a, in a sustainable way. How do we figure out land? How do we make these data centers more compact? Um, I think that'll just become a very big, big topic of conversation in the coming weeks and months and years. A big theme on my show so far, both with Sam Altman and Zach was the data center backlash. And I'm curious how that touches your world, is it something that impacts you base 10 and your customers? Is it something you're talking about a lot? People are really reacting vis-à-vis to, to what's happening and I'm curious how you think about that. Look, does it affect our business? Yeah. Like, yes, like, you know, like we, we need a lower compute and we need to scale and we need to figure out how to get intelligent as the hands of. people and as we go a lot of computer, I think all these things to some extent just slow us down. I think at the same time, there are both valid concerns about what this means for these talents and the countries where we're putting these things up, but also a lot of education to do. I think we've done a pretty terrible job at the industry so far, kind of bringing everyone else along and describing, "Hey, what has this actually been? How does this change things?" To me, it's more just it just shows to some extent the responsibility that we all have to be able to make sure that this thing that we have seen the value occur from grow to the potential that we see it has. I'm out of no if it helps for us to be going in a spreading flood around these things. To me, it's a lot of just education driven, I suppose, anything else. Well, I think it's very tough because yeah, I think a lot of researchers at Anthropic, especially, but also open AI, they really do believe that what their building has potentially negative effects. Everyone has seen recently the ex- Anthropic researcher who said 10% chance that make ill humanity most viral tweet of all time in terms of views within 24 hours. That's true. That's true. Yes, I think Elon confirmed that. You see that, and then that researcher is on Anderson Cooper. Later that evening, it's hard to balance all these things because you just listed a bunch of your customers who are doing genuinely impressive things with AI, helping doctors, helping lawyers be better at their jobs, etc. But at the same time, there's this existential fear and dread about what AI will do. I'm wondering if a company like Base 10 that sits in such a critical part of the stack, what role do you have to play in this conversation? Have you thought about that? Yeah, look, I think a lot of it for us is just, there's two things you can do, right? There's like one, make sure we are listening and we understand the actual environment. We can't live in a bubble. This is kind of the thing I said to you. Oh, there as well, which is like living in tempers, let's go. We just have a very specific view of the world or the way we see it and seeing how it's affecting doctors out, affecting lawyers. That helps ground you to like, hey, this is thing. And then too, it's like telling those stories is very important and highlighting the use cases. And three, I think the biggest thing is like, look, we have had lots of powerful technology for many years. And the way we work around them is to figure out what are the Godrails and what are the balances that we put to make sure that they don't cause unto you harm. But also, that's an amazing thing about academic research and policy is that there's a counterbalance of these things. When we encourage, we encourage security hackers to find data or vulnerabilities. And that is so we can build a ecosystem around that to patch them and fix them and roll them out. I think that's the same thing will happen to me. I think the biggest challenge right now is like things are moving very, very, very fast. We're not having that time to react. I think that is probably where the responsibility comes in, which is like, take stock of what's happening, and just be thoughtful about the future we don't live in and then make sure that we bring people around and along and maybe don't drink too much over on call it. Yeah. Speaking of things coming fast, I've heard you say that inference may be the last market post AGI, which is an interesting take. The idea that maybe the labs just reach their training goals and AI training stops and the billions of dollars spent on training shifts to inference. Do you really believe that and how do you see that practically playing out? Because I just spent a lot of time in OpenAI and they were talking about AGI with Astra and Sam and Greg told me they basically got it. They feel like they have AGI, but the world is continuing and new models are still shipping. I'm curious if even Astra and what A/OpenAI has said recently has changed your view on that or not. All that claim is is that we just have a lot of AI and whether it's the final market or not, it's definitely going to be a larger market, that's for sure. Running these models in production and serving and user value is going and agents doing work this way is going to be massive. The final market claim is really just that even in a world which you have AGI and again, I'm not educated enough to be able to speak to that. The researchers who are building it can't speak to it either, don't worry. They're all having different opinions. I can't speak to that, but when I get there, he's like, look, without AGI, it's going to be the biggest market ever. With AGI, well, the only thing left to do is for these models to run. It'll be the only market left because these models are going to run in loops and do everything for us and if that's the case, it's like, well, we're going to need a lot of compute and a lot of software to be able to run these models really, really fast and that's what they claim. I think I believe that. I want to end here because I think it's a really interesting look at where things are going. You all shipped, I believe it was an MCP server and a skill so that coding agents can use base 10 directly. Does this imply that agents are becoming your customers? Yeah, agents are deploying models in base 10 for sure and running them, which is kind of wild. Agents are deciding which model to use when. And how recent of a phenomenon is this? I think this has been happening in some form for a while or just getting the agents to see it and the models are getting good. Are you imagining a future where your customer support, instead of you getting paged mid-interview to go deal with a customer that needs something fixed, like base 10 agents are talking to customer agents? Well, if you go into our incident channels when things happen, the agent is debugging for like, while we are working on figuring stuff out, the agent is debugging by itself. It has access to all I could. It has access to everything about the model. It knows everything about the architecture. It knows about the customer issues come up and it's like, hey, there might be something unique about this model that's using this data in this way. You should go check this out. And then the extent to which we give it power to do that thing itself and patch it in like today, we're doing, we run stuff in production for, you know, and these are really important things, and we don't want to delegate all that such large impact actions. But I think for sure, I think all these problems just taking salt pages and what happens, the human skill just has to change to figure out how to interact with the data to give them the most leverage and vice versa. What is the last thing you and your team are going to be delegating to agents? The last thing, hanging out with each other? I don't know. The human contact. I went to dinner with a few of my colleagues to say that. That was a highlight. There's no time I'm delegating my social interactions. But is that maybe all we've got left? That's all we've got left. I mean, that's all we've ever had. I guess what I would argue with you. All we had is people. All we had was people. That's a great place to head, I think. This is getting way too philosophical. We got to end this. No, too. And I really appreciate it. I appreciate you chatting through all these things with me. Thanks for joining. Thanks, Alex. That found me. Banking should feel like modern software. Get everything you need in one place. Mercury's a fintech, not a bank. Check the show notes for details. Grinola is the best AI notepad I've tried. It works everywhere on a video or phone call in person or an Apple watch. Try it now. Grinola.ai/sources and use the promo code sources at checkout for three months off. Giro by ad lasing is where your team and your agents work from the same context. Try it free at gira.com. That's j-i-r-a.com. Framer is the AI native website builder that lets you build faster without giving up control. Visit framer.com /sources for 30% off. Rules and restrictions may apply.