A Product Market Fit Show | Startup Podcast for Founders

His 1st startup failed—but his 2nd one hit $100M ARR & a $1.6B valuation. Here's what he learned. | Liran Zvibel, Founder of WEKA

Mistral.vc Season 3 Episode 60

Liran quit a cozy job at IBM to launch Fusic, a TikTok-like app back in 2011. He raised over $10M, acquired tens of thousands of users, and failed.

So he went back to what he knew: deep tech and enterprise. He launched WEKA in 2013 to improve the efficiency of GPUs. He was operating on hard mode: building deep tech and selling to large enterprise customers. It took him 5 years to build a commercially-ready product. In that time, he raised over $35M from strategic investors, since VCs didn't get it. 

Once they launched, they more than doubled every year. And this year, they crossed $100M in ARR. 

Here's how Liran built WEKA and got it off the ground.

Why you should listen:

  • Why deep tech is much harder than normal software startups and always takes much longer.
  • How to get enterprise customers to commit well before your product is ready. 
  • How to leverage strategic investors to get you through the early days when you have no revenue.
  • How Liran was able to get customers to pay 6-figure deals when competitors offered 'similar' products for free.

Keywords
Weka, deep tech, large enterprises, GPUs, OS, product-market fit, funding, strategic investors, POCs, POVs, AI, GPU use case, performance, cost reduction, rapid growth


Timestamps:
(00:00:00) Intro
(00:02:12) Why my first startup failed
(00:08:35) Starting WEKA
(00:15:04) WEKA's First Customer
(00:17:43) The Operating System of CPUs
(00:21:19) The Issues with Deep Tech Companies
(00:26:19) Competing with a Free Product
(00:32:57) Reaching a Couple Million in ARR
(00:36:26) Fundraising
(00:43:19) Finding Product Market Fit
(00:44:08) One Piece of Advice

Send me a message to let me know what you think!

Pablo Srugo (00:00.366)

So today we talk with Liran, the founder and CEO of Weka that just raised $140 million Series E this May at a $1.6 billion valuation. In total, they raised $415 million. This is a very different type of company because this is very much a deep tech company and not just deep tech, it sells to large enterprises. So you can imagine it took a long time to build the first product and get it to market. In fact, Liran started in 2013 and didn't really get the product to market, the GA, until about 2018 -2019. So it took like over five years. And the simplest way I can describe what Weka does is, and I think somebody that really gets it might kill me, but I think for everybody else, the simplest way to think about it, it's almost like this OS that runs on GPUs and helps GPUs run 20 times faster. The simplest way that Liran described it to me is like without Weka, your GPUs are operating at maybe 30 % capacity and with Weka, they're operating at 90%. So you can imagine in today's world why there's so much demand from enterprises for Weka's product. But it took a long time and a lot of money for Liran to get this to market. He raised a $10 million Series A and then a $25 million Series B when he had no revenue. And so we dive deep into what exactly he had to do to get real validation from customers, to know that he was building something that actually would have demand once he built it. And what he had to do to go out and raise all of this money without any actual revenue. We also talk about Liran's prior startup, which raised a bunch of money, but ultimately failed and why the best way to win is to play to your strengths. Welcome to the Product Market Fit Show brought to you by Mistral, a seed stage firm based in Canada. I'm Pablo. I'm a founder turned VC. My goal is to help early stage founders like you find product market fit.

Liran, welcome to the show.

Liran Zvibel (1:47)

Thanks Pablo. Thanks for having me.

Pablo Srugo (1:49)

So, you know, today you're running Weka and you know, that's been a huge, huge success so far.I mean, hundreds of employees, hundreds of millions of dollars raised. Before that, we were chatting and you told me that you started another company called Fuzik, which was kind of like TikTok -ish, except it didn't work nearly as well as Weka or as TikTok. Curious to kind of hear that story and what that was and kind of how that went. 

Liran Zvibel (2:12) 

In 2011, we left IBM after they'd acquired their previous company. When we left IBM, we left it with a huge sense of hubris, basically. Anything we would do is going to turn out being an incredible success.

Pablo Srugo (2:31)

Why was that, by the way? Just things were going really well at IBM?

Liran Zvibel (2:34)

Yeah, we've created a great product before. IBM acquired us for a lot of money. The product within IBM was selling like hotcakes, obviously in the retrospect of WEKA, if you slap the IBM logo on many other products, they would probably sell well. We grew incredibly fast and faster than anything you could imagine working from a smaller company. And we said, hey, we've created such a good product before. We got it to scale such big sizes at IBM. We can do anything. And then mobile was starting, video on the internet was just starting. And we said, hey, let's help teenagers generate content. The cloud was ramping up. We figured out, hey, we could actually let them record themselves singing and dancing alongside video clips from YouTube. And YouTube just realized how they can get the rights to put music on the internet. And obviously it's going to be a huge, huge, huge success. We created this platform, great technology. In seconds, you would upload your video and you could end up seeing yourself singing alongside, dancing alongside the video clip. We raised tons of money in that context, dozens of millions of dollars because anyone who saw that said, hey, Obviously, it's going to be a huge hit. 

Pablo Srugo (4:08)

I mean, there was a lot of trends. You had mobile, you had YouTube obviously taking off. And then so all these kinds of things came together, people figured, this is going to be just as massive as all the other kind of social plays.

Liran Zvibel (4:20)

Yes. And then, you know, the iPhone 4 just came out and you had the front facing camera. You could say, hey, you could generate your own content. But then a few things happened. One, teenagers, especially back then, didn't have the latest phones. So they didn't have front -facing cameras. You usually think about kids yourself. You're not giving them-Maybe now it's more accepted to go buy a kid an expensive phone. I think back in 11, 12, it definitely was not the case. So kids didn't have their own phones. If they did have it, they didn't have the latest model phone. So in order to record the video, you need to have a friend hold an older devices. 

Pablo Srugo (5:08)

Cause back then, I mean, people were still using what- probably some people started using flip phones. Some people probably had blackberries if I remember correctly. 

Liran Zvibel (5:14)

So the quality of the videos recorded, like what we were running in our QA was great because we did have the latest one, but what teenagers had and you know, connectivity we've had in our office was great, but connectivity people had all over. It's difficult to remember, hey, how underwhelming the internet was 15 years ago. So while we've created a product that was a no-brainer and was supposed to be a huge hit, it wasn't. 

Pablo Srugo (5:45) 

How far did you get? Like, I'm just curious, like in terms of usage, did you spend a lot of money kind of in the advertising side and actually got a lot of downloads, but just no usage or was it never, were you never even able to kind of get that, that initial adoption just because the hardware and the internet and everything around it kind of wasn't there yet.

Liran Zvibel (6:04)

So we actually, back then, doing interactive things with music was a huge, huge thing for the music industry because they were all trying to figure out how do they interact with their fans, how do they bring on more adoption. People thought, hey, maybe through the internet there would not be any concerts anymore. So we've had the most incredible reception from the studios. So we were having campaigns back then, Justin Bieber, One Direction, Ariana Grande, these huge, huge, huge, artists. And they would launch their singles, they would launch their video clips with us. We would get millions of people come in, dozens of thousands of recordings they would win prizes like we got them to win concert tickets and whatever but then we couldn't get them to come to come back again the following month so or the following week because the friction of going through like Justin Bieber posts like a short clip on his -back then Twitter- tells you there is a contest, asked you to come. You would figure out your two friends, one's gonna hold the camera, one's gonna help you, you'd get the video, but unless you have a strong need, you wouldn't go through all the friction of generating more content. And once the big contest was over and we got the winner announced, there was very little interest on coming back to these pages.

Pablo Srugo (7:54)

It's incredible how you realize how many things need to align and go right to create like an Instagram, a Snapchat, a YouTube, like these content platforms that also require retention, right? And like daily active usage and consistent creation. Yeah, it's not easy and that's why there's only so few companies that ended up being big winners in those spaces. So I'm curious, like how do you go from a company like that, the sort of company that a non-deep tech VC like me can understand to Weka, something that is a little bit beyond the sort of things that I get and something that is just so technologically complicated, right? Not consumer, it's enterprise. I mean, it's almost everything that Fuzik is not. 

Liran Zvibel (8:39)

Right, so with Fuzik, we basically said, hey, we've created a deep deck company once, we got a lot of reception in the enterprise, but enterprise is difficult. Deep tech is difficult, you have to spend all that energy. Consumer is gonna be so much more straightforward. We've tried that, it was part of the hubris. But you know, when we left IBM, we said we don't wanna do enterprise, we don't wanna do data management storage anymore because we felt the market was broken. It's insanely fragmented. It's a huge, huge market, like $150 billion annually, but hundreds of products that sell and no product scales to more than a few billion dollars annually. So customers end up running any number between 10 and 50 through tons of silos and vendors have to have big complicated portfolios. So we didn't want that. Also the delivery, even though these are software products, most of them have some proprietary hardware component in them, so you're buying software in the hardware box. 

Liran Zvibel (9:54) 

What's an example of that, by the way?

Liran Zvibel (9:56) 

So you can look at companies like NetApp, and NetApp has like four or five different boxes you can buy, whether you're buying their On Top or E-Series or Bunch. If you're looking at even younger companies like Pure Storage, they have the line of Flash Array. which is one set of hardware. They have another line of the flash blade, a different set of hardware. EMC has, don't know how many different hardware platforms EMC has. So all of these companies, no, Kulot Bakard. So all of these companies have these very, very dense portfolios. And even within the portfolio, each product comes on a different proprietary box because they need some different help from different hardware components. On premises, it's just hard, it adds friction. But it also means that none of these products that run very well on premises are gonna run well on the public cloud. Back in 11, and definitely 13 when we started Weka, we thought that cloud is just around the corner. Obviously it took another decade and a worldwide pandemic for enterprise to start adopting the cloud, another, I think, another interesting question of product market fit. But back in 13, we were fortunate to go through three huge, huge tech step functions. One, containers started happening, it was before the days of Kubernetes and Docker, but containers started happening, the ideas of microservices, hey, you can take a big platform chop it down to tiny little pieces and run it across many servers was starting to form. It's a very, very different way of thinking about how do you run a data center. Two, Flash was really happening with a standard called NVMe. There were probably 50 companies that were trying to make NVMe, trying to make Flash natively connected to CPUs. You may remember names like FusionIO, Viriden, File in Memory, Texas Memory Systems. They all had their proprietary approaches and became billion dollar businesses. Then the industry said, wow, that's a great idea, great need. If 50 companies can do it, it's not so difficult. So there was very little tech mode, but there was no standardization. They created a standard that happened to not suck called NVMe. And now the whole industry basically transitioned. When we started Weka, NVMe was a PDF. That PDF made so much sense. And we said, that standard's gonna win. 

Pablo Srugo (12:55) 

And that was a standard, to be clear, for flash memory? 

Liran Zvibel (12:57) 

How do you connect flash to modern CPUs? So people have been using flash in their mobile devices. You've had your disk on keys. So the consumer space have really adopted Flash, but the data center hasn't because there was no reasonable way that was standard to basically attach these Flash devices to modern servers. And NVMe created that way. The third thing that happened was networking. And networking speed, again, got a step function that in essence was probably 100 times in a few years. We took these three very big step changes in what's possible and then we've said, hey, you can actually create a different kind of data management product that doesn't have performance limitation, doesn't have scale limitation, doesn't have the form factor limitation. You can run the exact same thing on premises and the cloud. Basically what we're doing, we… came with a way of virtualizing data like VMware had virtualized compute. Obviously it took us many years to get this product to market because there was a bunch of fundamental computer science, fundamental technology that we had to go through and a lot of customer interactions in between just making sure, we're not drinking our own Kool -Aids. We're building something customers would want. 

Pablo Srugo (14:40) 

And this is a really simple question, but just to make it very kind of tangible, these are all, you know, these kind of three big tech innovations, microservices, flash mentioned networking. When you're talking to a customer, like just give me a sense of who is that customer? know they're large enterprises, but who's an example of that customer? with what you're leveraging these technologies for,

 

What do they get? and then what's kind of the ROI for them? Like what's the business case? How do they think about these purchases?

Liran Zvibel (15:09)

Yeah, so one of our early customers that, know, sort of serendipity that we're talking to before we actually realized, hey, that's going to be a huge, a huge part of our go -to market in the future was Nuance. I don't know if you remember in the early 2Ks, they had the dragon dictation. They were very strong on voice recognition. we're going to them. That's basically the first time that we've witnessed anyone that's using GPUs to actually solve an enterprise problem, an AI problem that was before AI was cool. Obviously they were solving an interesting problem and they were a company. 

Pablo Srugo (15:59) 

In the kind of dictation. like they're doing like text to speech to text, sorry, using and they leverage GPUs to actually get this done. 

Liran Zvibel (16:07)

They did.

Pablo Srugo (16:09)

Decade ago. Wow.

Liran Zvibel (16:09)

And what they were doing was like, they were the pioneers of the pioneers. It's early,

Pablo Srugo (16:15) 

it's hard to realize a decade ago doing that, that is early.

Liran Zvibel (16:20)

Very early. Obviously back then when we were going on one of the first POCs with them, one of the first, like they've been running an alpha. It was not even fully functioning all around product, but they had the hunch that this could help them. It was interesting for us. We were talking with a lot of the big secretive hedge funds on Wall Street because they care about scale and performance and latency.

Pablo Srugo (16:53)

So fundamentally, take Nuance or these hedge funds as a great example. Nuance, they're running, like we said, text-to-speech or speech-to-text kind of dictation services. They need GPUs in their case to process a bunch of this data. And for them, your solution means faster, cheaper, more reliable. These are the sort of the value props that they get.

Liran Zvibel (17:13)

Yes. And they were using some other storage product from IBM. Traditionally, it was an older file system called GPFS. Throughout the years, it has changed names. And they were experiencing operational performance scale issues. And they wanted to see, would using Weka help us solve these problems? 

Pablo Srugo (17:41)

And Weka is just pure software that kind of gets added to whatever stack they already have, or how do they implement your product?

Liran Zvibel (17:49)

Right, so we're on the one hand pure software, on the other hand we do require hardware to run on, so we need these servers with the fast networking and the flash devices. We normally come and replace whatever other data management product you're running, so we don't augment it because the problems we solve are fundamental. It's difficult to solve scale and performance if what you're using doesn't scale any slow. So what we're doing, we're running on standard wreckable servers that have a relatively strong CPU, fast networking. Nowadays it could be one or two 400 gig NICs, a bunch of flash on NVMe devices. Then we're pooling these resources and we're making the equivalent of what otherwise would be an appliance that you're buying. Similarly, we're doing the same thing on cloud where we would be running on instances that have fast networking and flash devices. And then we give the exact same value prop, the exact same feature set on the cloud that you can get on -prem. Also, we have a bunch of features around pushing data between different systems. So we can enable a hybrid cloud, we can enable you to start a workload on -prem and then migrate it to the cloud or start from one cloud and push to another cloud if you can get better pricing on your resources. 

Pablo Srugo (19:40)

Is this, and again, this might be a stupid analogy, but can you think of this kind of like an OS for these CPUs, for these servers, or is that not like the same way Mac OS is for a PC or Windows for a PC or whatever, this is in a similar place in the stack for kind of these GPUs, CPUs, or not really?

Liran Zvibel (19:57)

 It's like an OS. We actually have implemented ourselves a lot of the functionality you would get in an operating system. So we're running our own networking stack, running our own IO stack, we're running our own scheduler. So we can run alongside Linux, but we've actually implemented our own hardware level hypervisor that allows you to curve a portion of that server out where we run. One way we can run is as dedicated clients, the other way we can run is actually you can converge WEKA on the exact same servers you're running your application. That point you're getting even more efficiency. 

Pablo Srugo (20:42)

Walk me through what happens through that POC, because this is the difference. This is where I find at least it most deep tech and other let's say non deep tech startups diverge in the sense that product market fit for non deep tech company is usually not about, you build it? It's more about will they come? Like, can you actually build something people love and so on? In many cases in deep tech, because you get to go kind of so far out, you almost know if you can deliver this sort of improvements on performance, reliability, or whatever it is, they will come. But the question is, you know, can you actually build it? So what happens during that, you know, 2013 to 2016, 17 phase? 

Liran Zvibel (21:25)

Yeah. But by the way, in deep tech there is a big risk on are you solving the real pain points? And I think we're seeing a lot of deep tech companies that come and they either don't solve a pain point that's painful enough or the pain point that they're solving is not scalable enough. I can give you tons of examples of great technologies that didn't make it. And that's, I think, the highest risk because as an engineer, as a tech person, can fall in love with solving an interesting problem, but then you realize that the market actually doesn't think it's an interesting enough problem. For the large enterprise companies, I'm talking about deep tech for enterprise. For IT infrastructure, people have to realize these companies run and they have run with the current stack and making changes is a huge risk. Buying from a no-name company is a huge risk. You may not be there and all of these buyers have made a mistake of picking something that looks like a cool technology from a company that didn't make it, only to go and revert to the tried and true big players that at least you know are gonna be there next year. So you need to pick up not something that you can solve, but something that you think you can solve that has a big enough pain to enough customers that you can show that you actually can scale. And then you need to do it in a way that convinces them that you're a safe bet.

Pablo Srugo (23:21)

So it's a great point like how do you manage to, to validate that upfront? Because again, the differences with non deep tech, you know, it really sucks to go after a pain point that isn't a real pain point, but at least you wasted probably a handful of months, a million dollars building something, MVPing it, didn't work. with deep tech, you don't get to a product that quickly. And so it's really important that upfront you do something, to validate that. like, how, how did you know that if you, if you built it and it worked, you would have buyers?

Liran Zvibel (23:50)

So we went through a bunch of these POCs, POVs very early. We tried to figure out with these customers, hey, where do you suffer the most? We tried to figure out where could we find a repeatable case where customers suffer a lot. For a period we thought maybe hedge funds, the problem that hedge funds, there aren't enough of them. So it's difficult to build a company off of hedge funds unless you charge insane amount of money. But even then you're like a lifestyle business. We went through life science and we probably could have reached similar growth through life science, but we would have had to spend a lot more resource because life science customers are a lot more conservative. It takes longer time to convince them. And then through 17 and then 18, what has happened is this growth of AI problems, GPUs, all of these folks realized, hey, AI with neural networks are really great at solving difficult issues, difficult problems. There is obviously tons of money in solving these problems. All of them have the problem of performance and scale, which was two of the few things that we saw so much better than anyone else. We've started working with one of the most ambitious, full self-driving projects out of the Bay Area, and we've basically aligned ourselves to them.

Pablo Srugo (25:33)

 When was this? 

Liran Zvibel (25:34)

That was, so during the second half of 2017, we proved to them that our value prop, what we can bring them, is so much better than anything they can get out of the incumbents. So they were trying to run on pure storage. Pure was saying back then that they're the best solution for AI. They were a much bigger company than us. And when you're talking about AI, you have this concept of time, time to epoch. How long does it take you to go through one cycle? What we're showing them is that we were able to shorten their time to Epoch from about 14 days to four hours. 

Pablo Srugo (26:22)

at that point it's speed, like it's not just money saved, it's actually you get to develop your product that much faster. 

Liran Zvibel (26:26)

Yes. It's competitive. Yep. And basically if you're running one Epoch after the other and you're consistently so much faster, it means that a full week on the Weka ended up being more output than a year. on the incumbents. At that point, it's a no brainer. You take the risk. you know, they're trying to down negotiate us on price, but we realized we have a big, big edge. Like the incumbents were trying to say, all right, we'll give it to you for free. That's a great tactic. Big companies employ to try and kill new entrants. So if you can make sure that the first dozen customers that want to buy from the new entrants get your product for free, you basically suffocate the new entrant and they don't make it. And a lot of deep tech companies that don't show big enough differentiation between, hey, you get the offer of not paying, the risk is too high. 

Pablo Srugo (27:27)

Free is hard to compete with and there's credibility behind it. yeah, you've really got to-  And this is where I think it's so important, the type of ROI, which which at least for me is really what shines here is like, it's not just we're saving you time, we're saving you money, which, know, okay, that's impactful, but save me money? they're free, you know, that stuff. It's we're helping you do what you do every day that much faster, right? That's an edge that you can't really say no to if you're serious about winning a certain category, right? And that's core to what you do. You kind of have to, as you said, it's a bit of a no brainer you've got to go for. 

Liran Zvibel (28:02)

Yes. And basically when they were buying us instead of taking the free, we knew we've had it. 

Pablo Srugo (28:11) 

And how, by the way, that's another question, like pricing, like how do you even, what is the business model and what's an average kind of ACV given these sort of larger enterprise customers? 

Liran Zvibel (28:22)

So we actually tried innovating around the pricing and business model and what would customers pay for. And that's the one place that we've decided that we're going to be boring. The industry charges for capacity and we end up taking a subscription based on the capacity. So it's an annual subscription based on capacity and the management. We tried performance gain. We tried how many servers use it. We tried overall a bunch of things.

 

It makes the comparison of WEKA to the incumbents too difficult. It changes. It makes it too hard for the customers to imagine how much they would be paying in the long term. And while we probably could have charged more or we could have made it a more acute difference, hey, that's what the value you're getting, this is what you're paying for. We've decided reducing friction is more important than charging for the absolutely right parameter. And we're just charging like the rest of the industry.

Pablo Srugo (29:47)

It's funny, I've seen that many times where companies start up specifically will go out and charge like, especially when they save somebody, know, customer revenue, they'll be like, we'll charge a percent of that revenue so that it's a no brainer. Hey, like we're only going to charge you a percent of what we sell you. And generally speaking, what I've seen in only a handful of examples is that they default ultimately to the industry average. And maybe they do a little pilot to show how much they might save. Then they figure out like here's an annual contract. And then that's just what they charge for the reasons you said, it lowers friction. makes the customer not have to think as hard, makes it easier to predict all these sorts of things that aren't necessarily top of mind when you're a founder, but they're certainly top of mind when you're an enterprise customer. 

Liran Zvibel (30:25)

Yeah. Cause at the of the day, you have your champion. Hopefully you have someone like an executive sponsor, but then you need to also make your way through procurement. And the procurement folks at enterprise companies are not creative folks. And you just want to make sure that any step, like you're proving your value, you're creating the right relationship, you convince them to want to buy. From the time you convince them they want to buy to the time you get the PO, you want to make it as quick as you can because the incumbents, once they figure out they may be losing that K, are going to do anything within their power to go and combat you. So you want to make the time at procurement the quickest that you can. And at that point, you just want to show them, hey you would have paid Pure, you would have paid EMC, you would have paid Neda, this amount of dollars, you're going to take these exact dollars and pay WEKA. 

Pablo Srugo (31:36)

It's a simple thought exercise. And what are we talking about? Like in general, are we talking 100k, a million, 10 million? Like I can imagine these being pretty big customers. 

Liran Zvibel (31:46)

We do have a good amount of deals that are seven or eight figures ARR, but a standard deal is probably, we like deals that are 200, 250 to begin with. And then when you're comparing them to the on -prem, usually you're buying an appliance for three years. So a 200 or 250 is going to be between 600 and 750K. And then you need to buy the hardware. Normally the hardware vendors have 50 % GP, so it's about twice. So we're talking about for an entry level business for us, the customer would pay anything between one and one and a half million dollars overall for the three year appliance they would be buying. we're to 600 to 750, but we're going to count as booking the 200 to 250. 

Pablo Srugo (32:41) 

And how fast, I mean, that's the beauty, like that's the upside and downside of enterprise. It takes a long time. They're really big procurement processes, blah, blah. But when you get them, they really start to pile up and move the needle pretty quickly. How quickly did things take off in like 2018, 2019? Like once you really found this kind of AI GPU use case, how fast did you get 10 customers, You know, a couple million in ARR?

Liran Zvibel (33:08)

So, 18 was still, we didn't call that the GI year. We were still running with a small number of customers, obviously that car maker, we're trying to figure out how do you come up with a good product. I think also for enterprise and deep tech, there is a huge difference between an MVP, a minimum viable product that you can go and convince a customer you're bringing them value to a minimum lovable product. So we waited until we had the product that reduced enough of the friction. So we could start scaling. we really, really started scaling in 2020. And in 18 and 19, we're keeping it handful of customers. didn't really care about how much revenue we're pulling in, but we did care about where are we bringing value? What do we need to solve? How do we integrate? How do you make it simpler for them? And then in 2020, we went through the exercise of scaling up sales, marketing, the other portions to really bring and start scaling ARR and the business. 

Pablo Srugo (33:58)

How sharp was that ramp like in 2020-ish and in those early years? 

Liran Zvibel (34:09)
So we've been doubling year over year for since 2020 until now. We're now eight figure, sorry, nine figure ARRs. We're doing well. 

Pablo Srugo (34:32) 

Congratulations. That's the big, for what it's worth, for an outsider , that's the big milestone. You know, we celebrate a lot of milestones and they're all important. You know, fundraising is one that “Liran: Yes, being a centaur is good.” But yes, being a centaur is the thing, that's for sure. 

Liran Zvibel (34:46)

This year actually we're going to more than double, but that's because, you can track Nvidia, you can track Azure and the other ones, all the AI public company stocks, or every company that actually convinces the market, capital A market, they're an AI company, they grow like crazy. That's because there is so much pull. at this point, no, this year we're going to grow quicker than the plan is and we're beating the model, not because I think we're hiring great people, but just because product market fit is not a yes, no question. Product market fit is a gray scale. When the product market fit is so strong, when the market is basically cooling, you need to have people that are competent and working And then they're basically beating the model.

Pablo Srugo (35:43) 

And it makes sense. if today, companies, large companies can't get their hands on enough of these chips, on enough of GPUs, then they better make the most of the ones they have. then WEKA is just again, a no brainer. 

Liran Zvibel (35:56)

And what we're showing them, and it's actually quite straightforward to show, hey, before you've had WEKA, you were buying these bunch of GPUs, you've paid millions of dollars, sometimes dozens of millions of dollars. They run at 30 % utilization. You bring WCAI in, they run at 80, 90 % utilization. It almost doesn't matter how much you're paying for the WEKA. And by the way, we cost exactly like the other product, but you're getting three times on your $30 million spend for compute. 

Pablo Srugo (36:28)

Yeah, that's truly a top of mind problem for frankly, most enterprises today. Let me ask one question that's just burning in my mind because one of the things when I look at your story, I think, okay, like after 2018, 19, where this kind of AI play with GPUs was starting to become clear. I know it was still pilots, but I understand the story from an investment perspective. Like there's a clear narrative to tell there, but I'm looking at fundraising history and you raised 10 million in 2014. And then the big round that I'm really curious about is $25 million raised in 2016 when given the story you've given me, like things were still pretty unclear then, right? mean, how does that, how did you make that round happen? Because fundamentally with a company like this, if you don't get funded through it, you know, your runway runs out and your revenue is not going to save you. how did you kind of, yeah, tell me just a little bit about that time and that fundraise and how you made that all happen. 

Liran Zvibel (37:22)

We've used an interesting tactic that proved out well. We kept doing these POCs and POVs. So, and while we couldn't show customers are buying WEKA, we were able to get customers to be very excited about the technology we're building. while, and it is somewhat reasonable to go and convince and we did run the round from a deep tech company. Now it's called Celesta, their equivalent now is Walden Catalyst. They invest in deep tech companies. They understand the motion. They know how to go interview their customers. The other trick that has worked very well for us, well, trick, It did take a lot of work. We've actually funded ourselves through a lot of the strategic players. And now some of them are competitors, but throughout the run B and C, we raised money from Qualcomm and Melanox and Nvidia and Micron and Seagate and Western Digital and Hewlett -Packard and Cisco and Hitachi. And they probably forgot the bun. So we had probably the most amount of strategic investors, these folks really understand the market and they can validate that what you're doing is reasonable. so you know, getting a single strategic investor is a risk. Hey, if they don't like you anymore, what does it say about what you're doing? Having to maybe not solve it, but we have dozens. So basically if you can convince the whole industry that holds very smart investors that work for like what we would be calling strategics, that what you're doing is really differentiated and they want to be part of your cap table, then it helps the VCs because now people with deep pockets that understand the market validate you. Also, it helps the early customers. 

Pablo Srugo (39:40)

Yeah, it gives you credibility, I think, across the board. And it's funny, you say that like a lot of founders are wary of taking on strategics because of the signaling risk, because of different terms that they might ask for. So oftentimes they just get zero. the other choice is to go and get 10 of them. 

Liran Zvibel (39:48) 

Yeah. I probably didn't go through all of them. It's like the top of mind ones. 

Pablo Srugo (39:54) 

And by the way, like for them? What was the story to them? I mean, obviously there's a financial piece where they think they're going to earn a return, but like strategically, were they thinking they're going to be partners? Were they thinking what exactly? 

Liran Zvibel (40:08)

So definitely, no. If you're looking at companies like Hewlett-Packard, Cisco, Hitachi, they would be selling the hardware platforms. Hitachi and Hewlett-Packard, they're also storage vendors. If you're looking at the component vendors like Micron, Seagate, Western Digital, Samsung, they want to understand from the forefront, what would be needed in a few years, how do they create differentiated products if they're looking at companies like Mellanox or Nvidia or Qualcomm. They're a strong ecosystem. Obviously, Nvidia invested before they became such a great data center company, but they had the strong hunch that we would be making a big difference in the whole story. Mellanox invested because a part of the reason we could take what we're doing to market is their fast networking. So all of these companies look at these investments as a way of peeking into the future. They all have their CTO offices that have ideas, but investing in startups help them validate. They all want to see which one of them are winning. And their investment census is more around a “Hey are you looking here at the really differentiated technology? Could that be a winning game rather than is that going to wreck on an ARR the fastest?” They're not necessarily, it's not 100 % financial. And that's way it works well for the deep tech companies, or at least for us, it worked very well. 

Pablo Srugo (41:48)

Well, I could see that working because that's really, it's this kind of bootstrap motion, right? It's like, who can you get that really will just get it? So some of these strategic will really just get it. Then you leverage that to get maybe a deep tech investor. Then you leverage that to get the product far enough. At some point you have actual revenues and then all of sudden it was like, this makes total sense, right? Everybody else will accept it. 

Liran Zvibel (42:10) 

And end of the day, if you're running a deep tech company and you have a strong belief that what you're doing is differentiated and you're gonna hit product market fit, the most important thing is don't run out of money. And there was a period at WEKA that the easiest way for me to raise was through these strategies because they understood the story. You go to, to a VC, they, they cannot comprehend the differentiation and what they want to say. You've been on the road for four years, almost no revenue. You don't have any customers. What you're doing doesn't make sense. 

Pablo Srugo (42:47)

I could see that. I mean, I know very few VCs who would get it. And even the ones that who would get it to your earlier point about procurement, they've got to go to their whole partnership and explain it. And there's going to be people in that room that don't get it, no matter how excited the deal lead is. So it's a really tricky kind of dance.

Liran Zvibel (43:05) 

So for a while, I was going through the strategics because I could convince them we're differentiated and they would give us dollars. And sometimes people forget, but actually the dollars are important. They keep you up. 

Pablo Srugo (43:21)

There's no doubt about that. Perfect. Well, listen, Liran, we'll stop it there. I'll ask the two questions that we always end on. The first one, I think I know the answer to, but I'll ask it anyways. In the story of Weka, like when did you feel like you had true product market fit?

Liran Zvibel (43:35)

It happened twice. Once I told the story back in the end of 17, beginning of 18, we're able to convince that car company that they should pay us instead of getting the product for free. We felt, Hey, this is a very, very strong hunch that's gonna be a good direction. And then the next time in 2020, when we're really starting to build the commercial go -to -market team and we're starting to see, you bring on salespeople and you're scaling sales. you can get people that have not been with us and are not founders for an extended period of time, and they can convince complete strangers to give them six, seven figure dollar deals. It's not easy. It wasn't back then, obviously, if you go retract the doubling a few years, they were not huge numbers. But I think if if new employees can convince total strangers to give you a large amount of money, you're onto something. 

Pablo Srugo (44:41)

I think that's right. And actually, that's a lot of times where things break and you will see many companies who get that initial product market fit where it's either founder led or founder with a handful of people they’re really close with. And then when they scale the team, growth doesn't scale and there's something else that breaks there. If you could go back, you know, about a decade now to when you were just starting Weka with one piece of advice for yourself, what might that be?

Liran Zvibel (45:04)

Going through and being VC funded, VCs try to convince you very strongly to go to market and sell earlier. In retrospect, we have tried selling before we had an MVP, before we've had an MLP, I think it's very difficult for the company to realize where it is. I think we've wasted a bunch of resource and efforts back in 17, 18, maybe even 16 to bring on the revenue. If we could have waited longer, so I don't say “don't have the customer interaction”, customer interactions are gold, but

Don't have the interaction in the sense of trying to get the PO. Have the interaction in the sense of trying to figure out what's going to be your MLP, the minimum lovable product, and how do you make the quickest path to something that provides value but reduces friction. Then, when you have something that you think is an MLP, try to really sell it. Anything before is just going to mount frustration on both ends.

Pablo Srugo (46:18)

It's pushing on a strength and it's the fastest way to burn a lot of money. 

Liran Zvibel (46:21)

Yep.

Pablo Srugo (46:22)

Perfect. Well, Liran, that was amazing. Thank you very much for jumping on the show. 

Liran Zvibel (46:26)

Yeah, thank you. Really enjoyed it. So thanks, Pablo.

Pablo Srugo (46:30)

I just gave you content that you liked so much, you actually listened to the end and guess what? You didn't pay a single dollar. Not only that, I didn't even put any ads in your face. So you just got a bunch of content for free and now that I've delivered that value, I'm asking for something in return. Open your app, open Apple Podcasts, open Spotify, open whatever app you use to listen to this and hit that follow button. It's actually gonna help you because it's gonna help you make sure you don't miss out on the next episode, which you liked so much that you listened to the whole thing.

 

People on this episode