My $250M AI Startup Almost Bankrupted Me

[00:00:00][music] Nobody complains about AI costs until the bill gets too high. [music] AI cost went up to 500,000 and the month after that it was a million [music] because if I continue like this and the next month I have like a 20,000 customers and 3,000 customers, 100,000 customers, I'm going to go bankrupt.

[00:00:26][music] >> [music] >> Can you tell us the story of how you started using AI and when you decided to start building with AI?

[00:00:46]>> Yeah, I mean it actually started in LA.

[00:00:50]I mean the inception of Pulsia. I lived in LA for the longest time. I was in LA for seven years. I was working for Travis Galanic. That was in October in LA. [music] I decided to start [music] going all in on this company OS Idea and working on Pulsia when I was in LA in like around October of 20 2025.

[00:01:18]Essentially, I was building all these AI native products, but honestly, none of them really took off. And I was like, you know what? Let me let me be bossy. And I was like, I start becoming very inspired because suddenly, instead of like trying to figure out a product that other people's may like, I was building something that I found cool, that I thought would be useful for me, that I would want to use. So, that was the birth of Pulsia. LA has the best Korean food in the in the States. When I went to Korea, I was like I had some Korean barbecue. I was like it's better in LA.

[00:01:54]>> How does it feel when you're back in LA?

[00:01:55]>> Feels good. It's It feels like like I feel like the pressure is going down.

[00:02:01]Feels more mellow.

[00:02:02]>> Does LA feel more like home to you than SF does still or?

[00:02:05]>> Yeah, 100%. I really love SF. It grew on me. It's a city associated with like stress and work. LA associated with like you know a part of my life where there was like some things love stories like amazing work experiences amazing friends parties like just life stuff I don't associate it with like work work you know LA is like it's vibe it's like good vibes [laughter] is a really good buddy of mine we used to work together uh with Travis Kalanik at Cloud Kitchens. He became an investor of Kosia which is cool.

[00:02:46]>> I met Ben uh I think now it's like wow a while ago. It was like 2019 almost 7 years in fact more than seven years. But what I was really impressed about him is the fact that he had this unique talent that I never seen before of being like very good with numbers like a very strong quantitative problem solver but at the same time like having this kind of like edgy creative like out of the box vibe and the fact that like he was creating this like machine they were just being able to absorb anything that like a founder would put in this machine and just creating a business behind it.

[00:03:21]So, I thought it was a great idea. Reach out back to him, ended up investing a little bit, and yeah, I've been enjoying seeing him on this adventure.

[00:03:28]>> Anyway, after that, like I went back to Paris, decided to start locking in my Paris apartment, building it.

[00:03:37][music] Fleewood down. [music] >> Clearly I spending time here and having built Pia here I can feel that like there's something about this beauty and this love for community and this love for people and this love for for beautiful things and interesting things and great culture etc. that is embedded in foria in a way right [music] ever since I started I clearly knew that like it you know entropic models and open eye models were so expensive but I was like it doesn't matter because end of the day we just should figure out um if if customers really want the product and [music] want this AI for the use case of building a And when I realized that like there was such a big potential with Possia, staying in Paris would would hurt the project, I needed to go to San Francisco because it's the only place where people would understand.

[00:04:49]>> I'm [music] feeling my heart beating [singing] with my body. [music] I love this feeling.

[00:05:03][music] >> [singing] >> Yeah. So, I've, you know, I moved to SF like a little bit ago. Um, the park has been going quite a bit. Uh, the marketing stunt is really working and, you know, True Ventures is my preede investor. They've been convincing me to move to SF. Yeah, I mean, it's it's very exciting. Um, I I have to figure out where I work, but like it's kind of nice to be able to have an office and be with other people, even though they're investors. like they're they've been very supportive and like there's a lot of like um people in the team that used to be operating, you know, like um in different companies and so they're giving me great advice. I think the cost is getting a little bit out of control, but for now I'm really just focusing on, you know, maintaining the growth, making customers happy. I need to deal with like, you know, AI, customer support. I've been building and it's a lot of work, but very exciting.

[00:05:52]And that's when I started to like [music] work at out of the True Ventures office, which was like my one of my prece investors. They had an office in in in SF. And honestly, like they were a big reason why I moved to SF. Like they were always telling me like, Ben, like you need to move to SF. Like this is where you should be. Yeah, Paris is cute and yeah, we all love Paris and yes, Paris is beautiful, but like you need to come to SF to build. So I've been preparing this stunt where I'm going to have Pulsia raised around. And so on top of that, I I was thinking about, okay, like what what is the toolkit on the stuff you need to do when you're raising a fundraising round, you have to have conversations with investors. You also have to have a data room like investors will ask you like, "Hey, show me the demo of the product and like what are what is the data like how much are you growing, etc." So, so we created with Posia this like live dashboard where it shows PIA working on every company and shows all the numbers and all the growth in public. I'm very excited to launch it. I think I'm probably going to launch it like in a week or so. Like I need to finish a few things and I need to get the product right. But um I think I should just go for it.

[00:06:57][music] >> I went from 500 users to 5,000 [music] users.

[00:07:09]The marketing stunt of like having Pulsia [music] raised the running round really worked really well. I had like a ton of people, you know, email me wanting to to to invest. I was starting to gain traction on Twitter and I started to get a ton of new users, right? A ton of users. And obviously the bill start going up and up [music] [music] and AI cost went up to 500,000.

[00:07:55]In the month of March, I went from 500 users to 5,000 paying users. So, it was a crazy month. A lot of the infrastructure I had built like was starting to crumble. And so, at the time, you know, like scaling so fast, it was kind of crazy. like you know like the customer support agent started to like have to deal with like a lot more support because people were starting to build more and more diverse things and it was getting really weird you know at the time like I started implementing meta ads which is another feature but the long story short is like you know adding all those all all those automations created more and more AI compute because again every time something is happening in Pulsia it's an AI agent that is most of the time using opus or set which are very expensive models uh to do task and so the bills start going up and up and so then that's when I was like you You know what? Like there's all this investor interest. I would be stupid not to take this money.

[00:08:44]This money can be used for compute [music] so that I can continue bearing the pain of this AI compute cost. You know, investors are were like amazed by like the big vision I have for the product, you know, to create a product that enables anyone to be a founder. And so raising that money enabled me to not have to worry about cost so much for a second. [music] >> [music] [music] >> cash doesn't solve anything for your product. In this case, it helped me some with some breathing room. That's when I started working with Sapium [music] a ton. Signed a deal with Sapium and like for them to be my exclusive infrastructure for [music] agents. They are building infrastructure to enable any company that wants to build autonomous agents [music] to be able to have an infrastructure that works perfectly. And so I remember in April Elen was like, "Hey dude, like I was looking at like your AI bill. Uh it's trending towards going to a million million dollar and plus a month. Like this is going crazy." Obviously I was like, "Yeah, yeah, I know, but like you know I just raised money and like uh for now it's fine." And he was like, "Yeah, yeah, I don't think it's fine."

[00:10:01][laughter] And I was like, "Yeah, you're right.

[00:10:04]You're right. It's not fine. Okay, cool.

[00:10:05]What should we do?" Right? And so, nobody complains about AI costs until the bill gets too high.

[00:10:30]Yeah. So you know obviously we have this like huge bill with entropic even by charging customers a fair amount of money like there's so you know there's so much cost that come into play. I tried to break even initially but actually people run longer tasks and like ask more complex stuff to to the to Pulsia and so it ends up being so expensive and so I had this million dollar bill last month actually was more like 1.2 million. Thankfully, I'm able to afford it for now, but it's not a [music] sustainable business model. And so, it becomes so important for the business to be able to give this level of AI service to our customers without charging them like 300 bucks a month or 500 bucks a month. I want Pulsia to be used by billions of people. Like, I want Pulsia to be to become an iconic company. To do so, I need to reduce prices to have more people being able to experience it. But I cannot do that using Entropic. If I go from like 5,000 customers to like 50,000 customers, uh, in three months I will have spent all the funding, right? So clearly I need to figure out something before I scale.

[00:11:32]As I was talking to Elen, Elon was like, "Hey, dude, like what we could do is we could rent GPUs and we can start using open source models." And as I started digging, I was like, "That's crazy.

[00:11:43]Like, okay, so we can rent a GPU." with Elon, we decided to like, you know, go to the data centers to see like what we were renting, you know, to see the the racks.

[00:11:58]Most of my of users, they asking simple actually rather simple things and they have rather simple code bases and you can simplify a lot of that stuff. And so now the question is like how how unexpensively can I serve them so that I can give them a very good price so that more people can use it.

[00:12:23]What do you think the models are are open source models are going? Like do you think they're going to stop converging with closed models or do you think it's going to be behind >> a never- ending race?

[00:12:35]>> Yeah. And I don't see any sign of it just going out. I think it's still even under realized.

[00:12:46]The future is going to be quite interesting. I think we're really at the inflection point now where things will accelerate. We've seen the new frontier models and it feels [music] like there's never a ceiling. So, I'm very optimistic with what's coming. I think we're barely scratching the surface of what's possible and we're just getting started.

[00:13:04]So the next 10 years are going to be extremely uh compelling and I really believe that we'll see the equivalent of thousand years of progress in the next 10 years.

[00:13:12]>> And it was something that like you know took us pretty much 2 months to configure but in June my AI bill dropped to pretty much 100k which is a game changer because now it means I can be profitable. It means I can offer better pricing for my customers. I can make it more affordable. I can offer a free version of Pulsia which going to help with anyone being able to try PSIA and become a founder. And yeah, and that's like that's why I'm so excited about it.

[00:13:39][music] >> So I start to get into the habit, you know, as I move to SF to meet people in person because, you know, a lot of the customers are in the US. Really great to meet [music] customers in person.

[00:13:49]>> It blows my mind how easy it is. Like I just put that very simple idea in there and the next thing you know, [music and singing] it was just like all there. It was a great framework.

[00:13:58]I'm trying to make it affordable by using AI that are not like the most expensive like open AI tropic ones cuz that's like so expensive.

[00:14:05]>> So what what are you building with?

[00:14:07]>> Actually I could show you. It's a >> Yeah, that's >> so it's called Arya takes all the information too. So it's for uh if you're using for the sleeping uh aid, if you're using to for uh financial aid or even with the the mind reset, what it does is it factors in everything just for you. when I did Arya and end up to where I was like, well, if people keep asking questions, of course, it's going to cost me more than what I'm making because naturally, of course, everybody's going to use, you know, more free mode.

[00:14:33]>> So, have a limit of, you know, five credits.

[00:14:36]>> The way Polia works is that I work with a lot of different companies and I try to pick the ones that do the best job so that they can so that the customer doesn't have to think >> about what to use. They just >> cuz you don't care about which company it is. You just want it to work.

[00:14:49]>> Exactly.

[00:14:49]>> You just want it to work, right? You know, I started to like understanding a lot that like, you know, people they were building one idea, but then they wanted to build more ideas and like I think they wanted to be more uncapped on their ability to build more and often times yes they want to build more but like they don't have unlimited money on the bank account. Making a free version of Pulse and also making just BIA more affordable was very important to me. As I realized that like I could actually reduce the cost drastically of AI, I started being like, okay, actually if I can figure out a way to get cheap abundant intelligence, then I can make Psia free.

[00:15:31]Pulsia is like the end state of AI. If I'm able to go through the pain of having customers saying, "Oh, it doesn't work perfectly. You told me I was like going to build my business as I sleep and like there's a bug. I'm taking on all this pain of like all this like sometimes negative backlash. When I pull this off and when PIA actually is stable and actually delivers the synonym of AI could be pulsia. AI is pulsia. It's like an AI that does thing for me that is like my employee my co-founder my my assistant or whatever you call it. It's like an AI that does action for me. Like if you ask strategy today like how do I market my startup? Shajibit is going to be amazing at at answering. Okay, so you can do meta ads, a call out reach, you can I hear an influencer, you can start a YouTube channel. Possia is gonna be like, well, let me start a meta ads. Let me do some call outreach. Let me let me create a YouTube channel for you. Right?

[00:16:23]This is what PIA can do. And obviously, that's what AI should be.

[00:16:29][music] >> [music]