Welcome to another episode of Category Visionaries — the show that explores GTM stories from tech's most innovative B2B founders. In today's episode, we're speaking with Nikola Borisov, CEO & Co-Founder of Deep Infra, an AI company that has raised $9 Million in funding.
Here are the most interesting points from our conversation:
- Inception and Focus: Deep Infra was founded in September 2022, focusing on hosting top open source AI models and providing them through a simple API.
- Market Demand Shift: The company pivoted focus to large language models after the release of ChatGPT, driven by increased demand in the AI space.
- Customer Base: Currently, most customers are startups integrating AI into their products, with medium and large enterprises slower to adopt.
- Competitive Edge: Deep Infra differentiates by offering cost-effective, easy-to-use access to open source AI models, allowing customers to fine-tune models as needed.
- AI Market Dynamics: Nikola sees the AI field as healthy and innovative, with continuous releases of new models from major players like Microsoft and Meta.
- Fundraising Insights: Fundraising is a necessary part of the entrepreneurial journey, requiring the ability to attract smart investors aligned with the company's vision.
//
Sponsors:
Front Lines — We help B2B tech companies launch, manage, and grow podcasts that drive demand, awareness, and thought leadership.
www.FrontLines.io
The Global Talent Co. — We help tech startups find, vet, hire, pay, and retain amazing marketing talent that costs 50-70% less than the US & Europe.
www.GlobalTalent.co
[MUSIC] >> Welcome to Category Visionaries, the show dedicated to exploring exciting visions for the future from the founders or in the front lines building it. In each episode, we'll speak with a visionary founder who's building a new category or reimagining an existing one. We'll learn about the problem they solve, how their technology works, and unpack their vision for the future. I'm your host, Brett Stapper, CEO of Frontlines Media. Now, let's dive right into today's episode. [MUSIC] >> Hey, everyone, and welcome back to Category Visionaries. Today, we're speaking with Nicole Boresoff, CEO and co-founder of Deep Infra, an AI company that frays 9 million in funding. Nicole, how's it going? >> Very good. Thanks for caring me. >> Yeah, no problem. Super excited for our conversation, and I'd love to just begin with a quick summary of who you are and a bit more about your background. >> Sounds great. So, I'm the CEO and co-founder of Deep Infra. Deep Infra hosts the top open source models and provides them using a simple API. One, we think of it as we do like OpenAI API, but instead of OpenAI as models, we use one or three and extra and the best open source models for each category. >> Take us back to the early days, the founding of the company ICU started in September 2022. What were those early conversations like with your co-founder? >> First, pick people I want to work with and then we brainstormed about a few possible ideas that we could go after. These are hard days because you have to pick one of the things you will focus on, and a number of things look like good ideas, but you have to focus, so you have to pick one, and you can have someone stuck with it for a while. So, we thought inference is going to be quite a big area. A lot of people were spending a ton of effort and money and training models at the time, but not as much was being done to actually run the models. At the time, newer and newer models came every day, and so like the old ones that forgotten quickly. So, we looked at what we have a pretty good background, and we thought we could do something in the inference space. So, that's kind of how it got started. We decided from beginning the general area that we want to tackle and organize around that. >> It sounds like the company was started shortly before Chachi BT really was unleashed into the world. How did you see the business change and demand for the product change after that announcement? >> I think basically the large language models came a lot more into focus at that moment. Even before I was hosting a number of language models, but I think right before Chachi BT came out, the coolest thing was stable diffusion in the image models. It's interesting, I think the language models have more, I think, a wider business application than I guess the image generation model. So, it was a cool moment. We had to switch gears and focus more on that and a lot more demand came from that area. >> When it comes to the types of customers that you're selling to, what is that ICP? >> Right now, most of our customers are startups trying to build some products with AI. I think the medium and large enterprises are still figuring out how to use AI into their products and they're slower to move into this new area. So, they do customers probably start up building something and then trying to integrate AI into their product. >> This show is brought to you by Frontlines Media, a podcast production studio that helps B2B founders launch, manage, and grow their own podcast. Now, if you're a founder, you may be thinking, "I don't have time to host a podcast. I've got a company to build." Well, that's exactly what we built our service to do. You show up and host and we handle literally everything else. To set up a call to discuss launching your own podcast, visit frontlines.io/podcast. Now, back to today's episode. >> I have to imagine that there is a lot of different companies that are battling for their attention. Obviously, a lot of startups are building around AI right now. What are you doing to separate yourself from all the other startups that are out there? >> I don't think we have anything super special in that category. I just want to provide a good service that number of other companies basically need. So, the open source AI models are fairly good, but trying to run them by yourself is quite painful, getting access to compute, and the compute is pretty expensive. So, naturally, we host these models for people and then they get an API access to them, they pay only for what they use. It's super simple to get started. The difference with opening is that you're still using open source models underneath, so you have control or the meaning you can find you in them and do other stuff with them. So, that's what we're focused on. >> Where are our customers coming from? Is it just word of mouth right now or where are you finding customers? >> Yeah, I would say most of our customers find us for word of mouth, as a costing provider for LMs, we get mentioned in a number of benchmarks. So, yeah, most of our customers are coming from word of mouth right now. >> I was reading in the Venture Beat article and was talking just about how you are significantly cheaper than other options that are on the market. How are you able to achieve that? How are you able to deliver value for more so much lower than others are? >> It's a mix of things. We just have a very well-integrated solution and a very strong technical team and software stack for serving these models, and that's what we're focused on. We believe that the inference market is not going to be a fine margin business, but it's going to be high volume, and that's the strategy we're executing at. >> When you say high volume, how high are we talking? >> What I mean is there's going to be billions of tokens that will be processed through large-language models every day, and so that's what I mean. >> Yeah, I just mean how big are we talking three years from now? What do you think it's going to look like? What's the scale that you're going to be at? >> Three years from now, I think the inference market, and that's in particular we should be doing quite a lot of revenue in tokens. As long as new and very competitive open source models continue to be built by the community, and I think there's going to be a lot of market for services like ours. >> Is there going to eventually be an enterprise sales team and an enterprise sales motion, or do you think it's always going to be adopted by the developers and the engineers themselves and then go up from there? >> I think down the line we'll probably have a strategy around that, but at the start I think the right setup for us is to just build a self-serve product. It's very simple to use that we don't really need long-term enterprise contract or anything like that to kind of get started with. But down the line I think to really catch a meaningful amount from the enterprise market, I think different strategy will be needed. >> This show is brought to you by the Global Talent Co, a marketing leader's best friend in these times of budget cuts and efficient growth. We help marketing leaders find, hire, vet, and manage amazing marketing talent for 50 to 70 percent less than their US and European counterparts. To book a free consultation, visit globaltalent.co. >> How would you describe the state of AI today? Obviously, there's a lot going on in the space. It seems like every day there's 10 big news announcements and things that happen. So from your perspective, how would you summarize the state of AI? >> Overall, I think it's very healthy in terms of the amount of open-source models that get released by various organizations, just this week Microsoft released some new models and obviously, one of three came out like less than a month ago and before that, Mistrao released 8X20 to be. So there's a healthy amount of innovation happening both open-source and cool-source. I think there will be less ground-breaking research being shared compared to what we had when the transformer framework came out. Overall, I think I'm pretty excited. I think it's a new paradigm and running the models or doing inference is a new paradigm of computing. That's very computationally intensive. It requires a lot of power. Still very early on, there's a lot of things that we could do to make this process better. >> What about from a regulatory perspective? Do you ever worry that regulation is going to come out that really slows down the pace of innovation here? Do you think regulation is good? What are your views when it comes to AI regulation? >> It's a hard topic. We need to strike the right balance of allowing the field to develop without over-performing with regulation, but also we want to make sure that they're not used in a bad way down the line. >> I'm not very sure what the right thing here is, but I think we're still early on and small models like seven or 70 billion. I'm not really in my mind potentially quite. They're not going to be harmful, I think. They're going to be way more positives out of them than negatives. But some very large models, then I think maybe there will be some concerns about them. At least I have some concerns if the model parameters get too high. >> What are some of those concerns that you would have? >> Well, let me put it this way. I think if the models are smaller in size, then I don't think they'll be potentially dangerous to humanity, but they will be quite helpful in useful. If the parameter counts of the models and the computation, they require increases, then that's when we might get into a yes problem. My main point is that I think the models we have right now open source are definitely not, I think, causing considerable danger. I don't think they need to be regulated right now. >> Let's talk a bit about fundraising. So as I mentioned there in the intro, you've raised nine million to date. What have you learned about fundraising throughout this journey? >> It's part of the journey of an entrepreneur, like something you have to do, you have to bring other smart people along with you on the journey. It's not like my favorite part. I'm an engineer, I prefer working on the product and working with my team to improve the service. But it's something that I think everybody building a venture back startup has to learn how to do. >> When it comes to things you're working on, things that you're building, what are you most excited about right now? >> I'm quite excited that the field is moving so fast, so I think there's basically a lot of opportunities in various areas. >> I'm excited about getting some other alternative hardware providers for running this AI models. I just thought that there's more competition between NVIDIA and other folks that tried to build essentially hardware for this. I'm excited about these new types of models that they're going to be able to maybe directly ingest of you and images in addition to text. This is obviously something that go up on AI demo recently, but I think more will probably come. >> Overall, I'm excited about the adoption. The way I looked about the AI models is they have this kind of capabilities to reason and understand information and summarize it and analyze it. There's a ton of things that as a whole, we couldn't do before because it was too expensive to hire someone to, let's say, mainly read through this documentation, train these documents and analyze them. But that's now possible with various AI models. >> It kind of opens a lot of new ways of doing things. >> Amazing. I love it. All right. Well, we are up on time. We're going to wrap here. Before we do, if there's any founders that are listening in and they want to follow with your journey as you continue to build or maybe they want to be a customer or a user, where should they go? >> Yeah, just come to deepinfo.com and it's very easy to get started with all the top AI models that we currently holds. We're going to continue to expand and improve the service and provide APIs, simple-to-use APIs, the top open source models. >> Amazing. Well, thank you so much for taking the time. I really appreciate it. >> Thanks, Brad. It was my pleasure. [MUSIC] >> This episode of Category Visionaries is brought to you by Frontline's media, Silicon Valley's leading podcast production studio. If you're a B2B founder looking for help launching and growing your own podcast, visit frontlines.io/podcast. And for the latest episode, search for Category Visionaries on your podcast platform of choice. Thanks for listening and we'll catch you on the next episode. [MUSIC]
