Loading summary
Transcript41 lines
- [00:01]
A
Enterprise software. The first rule cannot do it in vacuum. That means you need an enterprise design partner who's going to tell you five different ways this is not working out, no matter how good the AI model is.
- [00:14]
B
Welcome back to another episode of Builders. As always, this show is brought to you by Frontlines IO, Silicon Valley's leading B2B podcast production studio. If you're bringing technology to market and want to learn from your peers, we have a library of more than 1200 interviews with Venture backed founders and marketers. Where they talk, all things go to market. And of course, if you want to launch your own podcast, we offer podcasts as a service to more than 80 tech startups. The idea there is very simple. You show up and host and we do everything else. Now, with all that said, let's jump in today's episode. Today our guest is Hardik Kabaria, co founder and CEO of VINCI4D. Hardik, welcome to the show.
- [00:53]
A
Thanks for having me, Brett.
- [00:54]
B
Of course. So let's imagine that you're getting on a flight, you sit down with someone, they're chatting next to you and they say, so what do you do? How do you answer that question? What do you do?
- [01:02]
A
We build foundation model for the physical world. Everything around us is physical. That's how human interact with the world. It's your phone, it's your watch, it's the car and the flight you are sitting in. These are all physical parts operating in the world that is governed by laws of physics in the universe. And as humanity is trying to build better and better experiences for ourselves, it's important to build better parts and can only happen if we better understand the physics in the universe. And we build an AI that helps the engineering team of the world understand the physics of the parts they're designing.
- [01:37]
B
And maybe for the audience here, just to visualize what that actually means in the context of a customer, you don't have to name the customer. You can if you want to, but maybe just talk us through a case study or a customer success story that you have. And again, totally fine if you want to make it anonymous.
- [01:50]
A
Yeah, anonymous is the way our customers like us to be. But it's easy to describe it. So today when we build a foundation model for the physical world, and I'm sure you've heard that so many, many, many times, we've actually done it and we have shipped a product, but it comes by picking a problem. So the first thing we picked is like, okay, universe has many different physics. Let's pick one. Picked heat transfer and Then if you think about what is the most interesting challenge in hardware today, designing power efficient manufacturable semiconductor components. You have that we can all have the AI inferences and then we can use AI in all of our day to day life. But when we do inference, there is a chip running that code in a data center and yes, it's a heat transfer problem. So we picked heat transfer in semiconductor electronics. And so now today our users, our engineering team at the top tier semiconductor electronics companies and they use our AI powered software to ask the questions like for this memory or this asic, how is the heat going to transfer through? Can I make it more efficient so I can run faster workloads with these cooling constraints and so on. That's how the users are using our tool that is deployed for that.
- [03:09]
B
So how do you settle on heat transfer? I feel like a lot of the founders I know when you have technology that can be applied to many different markets, one of the most difficult things to do is to walk in and say, okay, this is where we're going to focus. Because when you focus on one area you're kind of saying no to everything else. What was that process like to be in the position to say heat transfer? That's where we're going to focus?
- [03:27]
A
Yeah, I mean I do believe in beachhead style market building strategy for startups. So you have to pick a problem and you want to pick a problem that allows you to expand your footprint. There are so many industry verticals in hardware, from medical devices to semiconductor to consumer products, and there are so many different physics. So these are the two axes I was thinking about in my head and we wanted to pick one where the problem is sort of your hair on fire. And the second is that the world really cared about solving it really fast and you were creating lot more parts in the same amount of time. How many new semiconductor chips are created per year and launched versus how many new aircrafts are launched per year? I don't know exactly. So if we are creating a Ferrari, our users better care about taking it for a lap. That's a metric to keep in mind. And the second is how important that physics is for a performance metric for those products. This is where I got really attracted to heat transfer problems in semiconductor and electronics. And the other part is complexity. So what is the most complex parts human manufacture today? Nanometer to centimeter is seven orders of magnitude of features in semiconductor industry. That is like resolving human hair on New York. That's how complex we can produce today with an extremely high yield so we are already capable of producing such parts, but the design processes are left behind. So that created an opening. So I always like to ask ourselves, like, okay, even if Vinci didn't exist, is the world going to solve this problem? And if the answer is emphatic, yes, okay, that, that's the opening, that could be very small, but we can run our train through that. So that was the mentality that we went through. And of course then semiconductor itself are at the heart of everything, like your phones, laptops. And so many chips are in the car. Everything that is inferring in AI has a chip. Everything that is training an AI has a chip. So it naturally creates a nice supply chain motion that we are starting to see. Our semiconductor users are introducing us to the board users which are there downstream. But you have a natural problem because hardware is system of systems. It only succeeds if everything succeeds. In a data center, there are so many people providing a part, but if only works, if the whole system works. When me or you want to run
- [05:53]
B
an inference, you're about three years into the journey. At what point into that journey did you decide on heat transfers? How long did that take and how much iteration was involved to get there
- [06:01]
A
before we had a name for the company?
- [06:03]
B
Oh, wow.
- [06:04]
A
We had been thinking about where to start for quite some time because for me, physics is the language of the universe. Yes, we could solve heat transfer equation, we could solve elasticity, electromagnetics, fluids. It's very important to decide where to start. And that pretty much governs the journey for quite some time, as we both know. So it was the first decision we made before we had a name for the company.
- [06:28]
B
When you think about the competitive landscape, you don't have to name competitors, but maybe just think about the buckets of competition. What do those buckets look like?
- [06:36]
A
Yeah, so physics is an important part of design process, been around for so many years since at least 50s and 60s. So there is a bucket where you have a legacy tools being used by extremely competent engineer in an episodic way. So you do physics once in a while at a particular phase change or a gate of a program with a very high caliber talent taking a long time to do things right. So that is an existing play that is out there, but it is hugely limited by the number of people who can do these things right. They would have PhD in mechanical engineering or electrical engineering to even use these tools. And that is a very small population creating physics analysis for $4 trillion economy. Then comes the next wave of technology which I call situation specific. So hey, I can create models that are specific to, you know, the wind tunnel experiment of an aircraft or a car or it works for specific things. And the place the last bucket is, I have a foundation model for physics starting with one and now two. That just works. You know why? No matter the company, Apple, asml, tsmc, or my next door startup with one person, nobody invents physics. Physics just exists. So nobody owns physics. So we understand the physics because physicists before us have done a good job of characterization. If we teach that physics to the model ground sum first principle, then you know what it does works out of the box. Because the heat equation is not different inside the firewall of Apple. It's exactly the same as it is outside of the firewall of Apple. So if we teach grounds of real physics, that is the bucket that just works out of the box and create a completely different value. Not only it enables more people in the organization to perform analysis, but it actually enables them to do higher fidelity physics. We can talk about that like it just levels up the game. That's the third bucket. And of course, that's the bucket we are playing.
- [08:41]
B
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 I.O. podcast. Now back to today's episode. I mean, I hate using these words, but are you almost democratizing physics then? Or is it physics as a service? Like, is that how you would think about this model?
- [09:14]
A
So I'll go to something even basic. So let's say me and you, we both have used Waymo, right? And when you use Waymo, what do you think you're using?
- [09:24]
B
Filipino drivers behind the scenes.
- [09:27]
A
Exactly. Hey, I will take that. Right? But at the end, it is a transportation infrastructure. It's going to change how we travel from point A to point B. Even if it is Filipino driver, somebody else remote driving the car. For me and you, it's completely abstracted. We don't have to have insurance. We don't have to interact with the person who's driving, like Uber. It's a completely different abstraction layer that is now going to be judged as an infrastructure throughput. How many rides did they give? How many times somebody had to intervene. So it's an infrastructure that has been created by a completely new technology powered by AI. Where are we headed? Physics as an infrastructure. Infrastructures are judged on throughput quality. How much data did you process? It actually does not care about users, it cares about the travel. So in our case, they care about how many physics questions were answered. What is the quality of the answer, how high fidelity they were right? Because at the end, compute is a commodity and physics today is not at all a commodity. Infrastructure makes it a commodity. We are building physics infrastructure that makes it available to anybody who wants to tap into it is a compute capacity running. It's not limited by human capacity. It is available at a compute capacity. So that's what we are building. So of course it democratizes it. That is the job it does, but it does something more. It actually is judged on the throughput capacity.
- [10:50]
B
And how do you think about building a business model and a pricing model for this? Because I would have to imagine that this is not just amazing product innovation here. I have to imagine there's heavy business model innovation happening as well.
- [11:02]
A
And that's happening across the ecosystem because of the AI is transforming the way humans work. Right. So that's just happening and it's happening across enterprises where not just the physics work or hardware engineering design work, it's transforming how we do things with software. Right. And how do we judge that? We are pricing it like companies ahead of us. Whether it's anthropic or OpenAI, cloud cursor, you name it, they are all pricing it based on the usage and creating an implicit corollary to, you know, hey, correlation to the value created usage is some sort of an inkling to the value created. So that transformation we can adopt. Definitely we are adopting. So we also charge based on usage, not user. That is also in line with the infrastructure story. Nobody cares when you adopt a new database or a new data system. You don't care how many people are going to use it. It should just work. It's there and it should work. And our goal is the same. So it's a usage based. Does that create a longer conversation with an upmarket procurement teams? Absolutely.
- [12:07]
B
As you're having sales discussions, is there a specific moment where you see there's a pattern in terms of where like the light bulb turns on and people just really get it and understand what you're saying?
- [12:16]
A
It starts by thinking what the hardware economy is. I'm going to take a step back and then I take you there. So there are very few hardware companies that scale. But there are a lot of hardware programs. So if we think about the number of programs these companies run, that's exploding. That's the exponential curve. The number of employees is not really exploding. So the hardware engineering pool is sort of playing flattish. Very slow growth, but the number of unique programs is growing. So what we do is we try and create an aha moment for a program to say, okay, for my program, whether it's ASIC or memory or something, I can do things that I couldn't do before in an extremely small amount of time. When I use Winchy's physics infrastructure, what it does is that allows them to create better decisions. Higher fidelity data, easily available, just enables them to make a better decision. And when they get that, that's an aha moment. But now how do we see that? So your question was, how will winchi detect in our sales motion that that just happened? And the way we see it is that they stop comparing Vinci and to something they were doing before. They just put Winchy snapshots into the way they talk to their upstream team, downstream team, heck even upstream and downstream partner. So the Winchy is a snapshot going out in the slide. We call this the moment of authority. Because when an engineer puts that and signs their name, he and she is saying, I trust this enough, saying this is the right way to do things. So you go from an assistance to authority. When that transition happens, we know we turned a corner into a completely different category of discussion that does enable us to have a larger discussion with the procurement team or the leadership of the hardware engineering company.
- [14:08]
B
One of the trade offs I think of being so innovative is that oftentimes there's not existing demand for a market to grab and say no, no, no, look for us instead for you. Do you have to go out and create this demand entirely? Then like is there any existing demand? Is there existing search, is there existing line items and budgets or you know, they're making decisions about okay, let's go and find a vendor for X, Y, Z or do you have to go out and just create all of this demand in the first place?
- [14:33]
A
So physics is important. That is well understood by hardware engineering companies and their leaders. Because the executives, they don't just care about reducing the cost of their organization. If they have to choose, they care about shipping a better product faster. If they have to choose between reducing the cost versus shipping a better product faster, they will always choose the latter. And they understand it's performance driven, including consumer products. Consumer product, heat up. We all hear it in the news, of course, that traverses all the way back to, okay, is the chip heating up? Is it, Am I able to do data rate transfer faster? These are all physical performance things. So it's important in the design cycle. That's well understood. Is there an existing line item for spending money on physics software? Yes, it exists. So that's good. That works in our favor. But we are creating a completely different way to interact with the physics with a different business model. Now. That's the reason why it's a longer discussion makes sense.
- [15:30]
B
So there is some existing demand and you're capturing that demand and saying, no, no, what you're actually looking for is what we do.
- [15:36]
A
And that is the heart of the conversation. So yes, there is demand. It's well understood. We do get inbounds. But helping them understand how you actually take advantage of it to create a better product faster, which is what everybody's getting paid for in those organization is the work. But we go on that journey together because we also don't believe enterprise products can be created in vacuum. So yes, we are a frontier AI model company, but we are also. Rubber has met the road. This is an enterprise software company. And enterprise software, the first rule cannot do it in vacuum. That means you need an enterprise design partner who's going to tell you five different ways this is not working out no matter how good the AI model is. Right. So that's where we really integrate and understand the workflow and make sure that our product is useful day in, day out.
- [16:25]
B
This show is brought to you by the Global Talent company, 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% less than their US and European counterparts. To book a free consultation, visit GlobalTalent co. And what's the thinking? If, you know, let's say OpenAI were to come out and say, okay, now we're gonna do this similar to what they just did with the figma, a new version of figma effectively. And I think, yeah, someone else did it for Canva. Like, do you have any concerns there that they're gonna go out and launch something similar within their existing platforms? Or is that just not possible for
- [17:05]
A
them to do so because we have done it, it's stupid for me to say it's impossible. Somebody else. That's statement one. But I'll tell you what we had to go through so then we can understand. So first off, this is a physics foundation model. It's not A language model. It's not built on anything that is a language model, whether it's a closed source model or an open source model. We are the chatgpt of physics building it ground up. There are choices that entail. So now you're not predicting the next token, is not predicting the next word or a string or a data type that can be represented as a string. You are predicting a high accuracy temperature field, a high accuracy displacement and velocity and pressure. It's a completely different usage and grounds up architecture. Is it based on transformer? Yeah, it is based on transformer, but now you have to build that whole stack up all the way to the tiny tweaks that we have done into the model. GPU kernels, we are coding all the way into the kernel, so it's a completely different stack. So if somebody were to go down this route, they would have to traverse the journey, which is a significantly different problem than adopting a language model to it. So that is the journey we have had. We cared about the problem quite a bit. So if we could use the language model, actually we would have used it. This is a reverse. We don't care about being an AI model before the enterprise product. We care about enterprise product so much, but our bar is non negotiable. It has to work out of the box. It has to be judged as an infrastructure. How many queries did you process? How often do you process? What level of complexity of queries did you process? Like we have had people who have run queries that lasted seven days. So yes, that means we are creating a completely different product powered by a grounds up first principle physics model. And that shows up in a way the hardware engineering teams care about. For example, you ask a language question and I ask language question, we are going to get different answers. Physics, we're going to get the same answer. So ground up, we are deterministic. So there are so much differences in the expectations of the physics infrastructure versus language infrastructure that and we have baked those things ground up to create that experiences for our customers and for you.
- [19:15]
B
How important is it to be the evangelist here and to be the voice in the industry and educate the market about the specific problems that your technology can solve. Do you view that as a level 10 priority of evangelism is here or is it secondary? What does that look like for you internally?
- [19:31]
A
So for us, we are customers and users first company. So we care about engaging with the people who have real problems and they are trying to solve the problem to meet the deadline, create and ship a better product. Faster to do that. Do we have to help them understand our technology and create transparency? Absolutely. And we do that. And those engagements happen at different levels sometimes. Like you say, it's the role of, you know, being out there and talking and engaging with them at conferences. But what I would say is a lot more important is in depth whiteboard session. That's where we really discover their needs, their problems, and how we can, you know, create a product that really serves them. So I would say it's not just being out there on conferences, but having an in depth conversation is a lot more important, but it's the whole spectrum. So it's not one is more level 10 priority versus others. They are both a priority, but we really care about. Like, let's get into the meat of it.
- [20:25]
B
I know we've only spent about 20 minutes together, but one thing I can tell from the call so far is that you're very gifted at communicating. And what I find often when I speak with technologists is I can speak with them for an hour and I walk away having no idea what they do. There's some questions in my mind of exactly how things work at the company and the technology. But overall you're just incredibly gifted at explaining it in very, very simple terms and helping myself and others in the audience of course, understand it for you. Was that a journey to get there? Are you just naturally gifted at explaining it or what happened behind the scenes? To be able to sit here on a call and explain very, very complicated technology and physics to someone like me?
- [21:00]
A
I mean, to me it's the practice, I believe that we all can teach ourselves any skill if we put our mind to whether it's physics, whether it's math, AI model, building a team. You got to be passionate about what you're doing. I want to solve physics problem for the world. I am so passionate about my that goal. Everything else is a job that I got to do and I'm really happy learning it. And I got good teachers.
- [21:22]
B
And final question for you, let's just talk about the big picture vision. So you seem like a guy who has a big picture vision, so I don't even want to limit you to just three years so you can go out five years, 10 years, 20 years, however far you want to go. What is the big picture vision here that you're after?
- [21:35]
A
So physics is like a compilation layer of hardware. You draw something because you want to make that thing. So the first thing you want to know is how does it perform in the real world? Right. That's equal to saying, I wrote a code, does it compile, does it even run? Like if it doesn't run, it's kind of text, not a code. Then the next phase in software that happened, okay, I'm going to do 80% of the job, you'll do the last 20%, you'll squeeze it. But to do that you have to understand that this code can be compiled on all the different architectures. And analogy is like you want to understand all the physical performance of your memory or CPU GPU board and now you want to change the design. You cannot change the design if you don't have a verification method. We are the verification method. Of course there is so much pull to get to the next which is our engineering team would create 80% of the job and you finish the last 20%, squeeze out the performance, create a layout, send it to contract manufacturer. Then comes the third phase. You write the product specifically and the software feature comes out. We are already at that state. That is the stage we'll be for hardware too. You'll write a product spec, you'll say, I want to create next gen CPU like this. It needs to be produced on a 3 nanometer node. This is the place where I will work with the vendors. This is this type of things. These are my partners. I'm not going to choose the materials you should get. I'm going to tell you my partners, you figure out the spec sheet, help me make the choice. And here are my customers requirement, they want to run these type of workloads and my CPU is going to get integrated into the board. That going to go into the transformers, not the AI transformer, the real transformer, the energy grid. And they are going to be exposed to extreme environment from minus 50 Celsius to whatever, 120 Celsius. So now you make sure that it runs and cools on all of those requirements. And can you help me create a GDS file that I can send it to my manufacturer? That's the AI for hardware. That's the vision we are going after. Not just for CPU for all the hardware parts, but we will get there in steps and we won't wait for eight years to have a model. We'll ship it.
- [23:35]
B
Amazing. Love it. Love what you're building. A lot of admiration for founders and teams that go out and really tackle big heavy, hard problems. And that's certainly a big heavy, hard problem that you're after. So thank you so much for taking the time. You're welcome back anytime and I'll definitely be following along before we wrap for anyone that does want to follow along with you. Where else should they go besides the website of course.
- [23:53]
A
Getwinchy AI reach out to us. We are interested in hearing from all the hardware teams that are running ambitious programs and they want to figure out how they should tap into the new age technology and of course if we can be of service and assistance. Definitely interested in interacting.
- [24:09]
B
An ambitious company for ambitious teams. That's what you guys are all about. Love it. Thanks so much for taking the time.
- [24:14]
A
Thanks, Brad.
- [24:17]
B
Well, that's all for today's episode of Builders, brought to you you by the Frontlines. If you want more amazing content like this, visit Frontlines IO where you'll find the library of more than 1500 interviews with founders, marketers and other GTM leaders, where we unpack the tactical lessons from their journey. And of course, as always, if you do want to launch your own podcast, we'd love to have a conversation with you. Visit Frontlines IO podcast as a service. Mention that you listen. Mention you love the show and we'll give you a 10% discount. Thanks for listening. We'll catch you on the next episode.
- [24:48]
A
Sam.