Alex Higginbottom, Founder at Zenithon AI: BREAKING: Zenithon AI raised $10m

0:00Hello and welcome back to Scaling Europe show. I'm Sam Johnson. Today on the show we've got an amazing founder that's just announced, a$10 million round to solve some of the hardest problems in physics. Alex, thank you for joining me. How are you doing today?

0:10Alex Higginbottom:Amazing. Yeah, thanks. Happy to be here. Great news. A lot of excitement. You're building Xenophon. For those who don't know us, tell me, what is it? So Xenophon is a frontier AI lab building world models for physics. So, you know, we've seen how impactful LLMs have been for knowledge work. Now world models for vision are beginning to have an impact on certain sectors as well. But really, we haven't seen this for physics. And so this is exactly what we're building to try and unlock some of the most important technologies of the 21st century. And what made you do this? You abandoned a PhD to, I guess, build this company.

0:51Tell me a bit about that.

0:53Alex Higginbottom:Yeah, so I mean, my background is in physics. I've always been really drawn to how can we make a positive impact, but also obviously just sci-fi like any physicist, I guess. And fusion energy is exactly that. It's both incredible potential, but also incredibly cool technology. And I worked for three years in trying to commercialize this, working with about 20 different fusion companies on, you know, how can we both accelerate the development of this, but do it in a way that is relevant and will have an impact on the economy. And I realized that, you know, the core iteration loop, you know, this build, measure, learn iteration loop, which is essential for any startup, which is so slow and so painful.

1:38Alex Higginbottom:And so that drove me to do a PhD in AI for plasma physics to see if I could speed it up. but you know um i got halfway through and i was just very frustrated about you know i was just kind of working away in my in my basement uh in cambridge and not really having an impact in the world and so i was able to raise our pre-seed uh dropped out and and hired my favorite um researchers and and was very fortunate to have abby my co-founder join as well and was there a moment in time where was it like an epiphany moment where you're like this is not it i need to be like sort of more in an arena building or was it a sort of like a slow gradual build-up that maybe you you kind of crept up on you that actually you could probably have more value by building a company yeah it's a really good question i think it was like it was there was definitely several epiphanies I think um you know partly actually working in the fusion industry you know we've seen everyone here's fusion they think it's 30 years away but in the last like 5-10 years we've had about 80 fusion companies start and they've made more progress I would argue in just a few years than you know governments and academia did in decades and so that was extremely inspirational and now you know with seeing um the progress unlocked by ai um at deep mind you know an alpha fold and um you know open ai and and others you know i was just so excited about how can we take this kind of startup velocity this kind of incredible technology that is ai and solve some of the most important problems and as if you yeah i was gonna say fusion it sounds like that's kind of like your where you've come from are there are there approaches to fusion or are there companies kind of out of those 80 that you think are that you've got your money on or that you're particularly excited by that's a good question um yeah i think there are some that have very exciting approaches and they have the money to do it so you know we we can't work with everyone right now we don't have the bandwidth so we have been quite selective in in who we're working with and trying to work with the ones that we you know really believe in in the long term i think can you maybe you can't show those names, can you talk about, you know, within Fusion, there are different schools of thought and even within those schools of thoughts and approaches, there are even more sort of different, are there approaches where your technology makes more sense?

4:07Is it with Tokamaks or whatever, like, you know, are there particular simulations or approaches where your technology is being used, you know, at the moment? So,

4:17Alex Higginbottom:yeah, this is something that we've been really, from day one, in fact, much broader than Fusion, but within Fusion, we've wanted to be as comprehensive as possible in covering being a true foundation model for plasma. So that means being useful for Tokamaks, but also laser fusion, every kind of confinement type. And that's really important because as I said earlier, we don't know who the winner is going to be really. So we need to be able to work with all of them. And this is, you know, we'll be launching our first world model next week and this is the first time that um any like this is really quite a quantum leap i would say in in uh the field of ai for in general but also in infusion it's the first time that we've had different confinement types in one model and you know we're going much much broader than that still we're building world models for extreme physics overall so this obviously has massive applications in semiconductor manufacturing and aerospace and beyond which we are um just as excited about as we are fusion and you you spoke about you know your own frustration and that these simulations can take such a long time you almost like program it you get them up and running and then you kind of have to just like leave them to work that i imagine that's very painful as like an academic or researcher even more so for kind of building these companies how short can you bring that that time frame down to what can be the real impact for somebody who could be using this technology yes that's really right now our key value proposition and um if you just to kind of set the stage here you're building a fusion reactor and there are literally trillions of different ways to put it together you know different shapes different temperatures different you know magnetic field shapes all of this kind of thing and right now a single simulation on a high performance computer takes 10 hours to multiple days so how can you ever build something optimal if you're so if you're trying to search trillions of combinations but you can only search you know like it's like picking up random grains of sand on the beach um whereas what we've developed gives you the same results but in milliseconds so it's a million times faster wow that's crazy that's not even like that that's not yeah that's like a not even an order of magnitude that's crazy that's so much faster and so where are you at in your journey right so you know i know you're you're releasing this model next week how closely are you working with certain companies are they starting to see value from it now what does that look like in terms of like product development and real value being used but being realized by end users yeah so this is you know a really really hard problem and this model we're releasing next week is really the first step and just like other you know frontier labs will be releasing new iterations of the model every month every couple of months and it'll just get better and better a broader and broader capability you know i mentioned the speed is the first value proposition the second one increasingly will be training directly on experimental data and there's this idea of closing the sim to real gap and then you won't just have something which is a million times faster but also more accurate to reality and And that ties into a bigger vision that we have of the company, which is, you know, maybe one day, 10 years, maybe less, we won't need to build five practice starships.

7:42Alex Higginbottom:We can do a million in a world model and then build the final one straight away. That's really the kind of North Star that maybe we'll never hit, but that's really what we were going for. um yeah and then in terms of um uh sorry what and then and then what what was there other points to your question it was yeah as well what stage of the product development or like are there users out there using some of the things that you've built already yeah great question yeah of course um so yeah this first year for company you know we we've been hiring like the a team you know um building our product, our initial product, which was a huge effort.

8:23Alex Higginbottom:You know, some of the best researchers in the world, massive conviction, extremely long hours to build this. But still, yeah, we've been fortunate to work with some of the leading fusion companies. We can't share exactly who that is right now, but we will be sharing that very shortly. And already they are getting value from it. They are using it in their day-to-day workflows. And also we'll be announcing shortly various academic partnerships that we have with leading institutions as well. And they've played a strong role in shaping what is the most useful for advancing the state of the art. Amazing.

9:00And let's talk about the round of seraphim. You've raised$10 million. You're still pretty relatively early on in the company's journey or history. Bact is backing you, Luna, Seraphim, MMC. What gave this roster of investors the conviction to invest?

9:16Alex Higginbottom:I think we really are tackling a generational problem, is the first thing. If we can achieve our mission, the world will change. And we've seen throughout history, you know, Moore's law, complexity is increasing. And you can't, the tools of today or of yesterday will not keep up the rate of innovation in fusion and semiconductors in aerospace in other fields going into the 21st century as these systems get more complex so we need something new yesterday and we can do that so I think that is truly a mission which is important but also obviously economically significant we also I think have been able to pull together a team that is genuinely world-leading so our chief scientist lou um he invented many of the biggest breakthroughs in ai for physics in the last five years our team of researchers from um you know meta nvidia um anthropic so i think and and then and then the early proof points that we've been able to demonstrate technically with our model and with our partners i think um yeah brought everyone together amazing and and you've got this money you've obviously built the team how are you going to be deploying that capital we're going to be spending your time on to kind of get you from where you are today to where you want to be in six 12 months well a lot of it is getting poured straight into compute into the cloud providers so and that's both for generating data so we own a lot of the data stack ourself and we can cover a huge amount of trapped space with with yeah a lot of capital poured into data and training models.

11:04Alex Higginbottom:We have very large models on the scale of multiple billions of parameters. So that requires a lot of compute. And then also building out our team both in London and in San Francisco near our customers. You know, it's such an essential part of any sort of startup spending is now on compute. What does that mean about your own fundraising journey, especially when you're now building you know sort of frontier AI big models that are going to require as you say lots of compute how do you plan for fundraising in the future you know are there things that you need to unlock other ways of predicting how much more compute you're going to need in 6 or 12 or 18 months and how are you planning for that for the company yeah that's a really great question so we've spent quite a lot of time this summer actually um trying to figure out the scaling laws for both on data and on the model side.

11:56Alex Higginbottom:And we've demonstrated now scaling laws that we have quite a lot of trust in that we've been able to track and kind of blind test. And that is how we're doing a lot of our planning going forward. So, you know, much like what the LLMs do, although maybe still an order of magnitude discrepancy. And how have you found the transition? You know, this is, I guess, like your first job in many respects, right coming out of studying or coming out of being an academic or researcher how have you found that transition from you know being in a lab or in a university environment to now running a company yeah i mean it's it's a great question i think you know i think it's um i guess elon has spoke a lot about how you know physics is a great preparation for uh problem solving in life and it doesn't really matter where that is and i i think i've definitely found that i think you know i'm very proud of the culture we have at the company where we can bring um quite a rational you know problem solving approach to all parts of business and that could be you know that even like gtm commercialization and marketing is another problem to solve so um of course there's been a massive learning curve massive ramp you know every day i think i'm at my kind of cognitive limit uh and that you know that that is uh has its um challenges but but overall you know i think it's this feels like my calling and wherever you had to spend your time you've mentioned that you've been building this uh amazing team of some of the best talent in the world you've obviously been doing this fund raise you obviously you are you know you have the research background the academic background yourself to do the work how are you spending your time are you doing some of the the training the building are you you've been laser focused on talent and the team has fundraising been a huge suck on your time how have you balanced it all yeah i spend a lot of time on people on you know hiring and making sure every hire is someone that you know this kind of paypal math your idea of i would i could imagine myself working for them or they are much much better than me in this kind of specific you know in whatever their role is that is something that's so important and yeah yeah i will continue to be spending a lot of time on that um i'm still doing a little bit of technical work you know over the summer since you know since we um especially since i'm not fundraising anymore um i've had a little bit more time so actually i'll be first author on a on a paper shortly but i think that might be the last for a while um and then yeah now we're really entering this phase of um focusing on getting our product which we think is extremely impactful into the hands of customers so i'm now spending most of my time with with customers and traveling to them and and spending time with them and learning the problems that they care about how have you found that part of the role i guess you know selling the product to potential customers but also selling the mission and vision to potential investors and you know you've obviously got this very clear strong view on the world and a very clear strong view on the value that you can provide to the product that you're building how have you found you know getting people on side and trying to share that vision in that mission with investors and with customers yeah so i think obviously what we're doing is to some extent it's kind of digging new ground uh it which is obviously a massive benefit for us massive strength you know market creation and doing something totally new but it does mean there is a little bit of you know kind of education that needs to happen with investors with customers uh so definitely that's been one of the things there's been a a learning curve ramp is how do you um you know communicate this in a way that people can understand but then also kind of ramp that up to especially with an investor for example you know maybe over time you want them to have the best understanding possible and we've been really lucky with our investors that generally they're very kind of frontier investors that have a good technical understanding of what we do, which we would care about.

16:06That's amazing. And you've got some great, great investors. So now what's next? What is the immediate thing? You've got this model released. You've got some exciting news coming out. What is the next priority for you?

16:22Alex Higginbottom:Well, as I said, we're still going to be always improving the model. So, you know, there's obviously this is a massive step for us and for the world with this model that we're releasing next week. But it'll get better the next month, get better the next month, increasing the scope, increasing the fidelity of what we can do. A huge goal is closing this into real gap as well. That's a kind of a large objective of ours. So that is one branch. And then now we really are in a great position to get this in the hands of customers. So, you know, working with a select few customers in Fusion and in semiconductors in particular on how this can accelerate their workflows today.

17:08And what does it look like for semiconductors? I think with Fusion, I kind of have a clear view of the impact that that can make. To your point around that, like, so many different sort of parameters or data points that can be slightly adjusted and build slightly adjusted. What does that look like for semiconductors? What is the use case for them?

17:23Alex Higginbottom:Yeah, so that's a great question. We've seen a lot of progress in AI for design of semiconductors. So, you know, this is, NVIDIA have done some work here and Synopsys and some very strong startups. But really, you know, we have not seen that much work on the manufacturing side, which is just as big. It's got something like a, you know, a multi-trillion dollar market cap. But, you know, companies like CTSMC, Applied Materials, Land Research, like all of these massive manufacturing companies. but they have an equally hard problem almost as we have in fusion where um etching out these wafers and constructing these wafers is getting incredibly complicated you know the cto of lime research gave a speech a couple of years ago where you know we're now a hundred trillion different ways that you can make a wafer and plasma is a huge part of that so if we want to continue to have better chips, you know, scale AI, then we need much, much, much better ways of building these chips.

18:30Alex Higginbottom:And that's where we're really excited. And if this goes well, what would be the larger impact on the, I guess, the entire supply chain, the entire market? Would it be cheaper, faster? What does that look like? Well, you know, Moore's law is bending. And our mission is to keep Moore's law or the equivalent of Moore's Law going throughout the rest of the century. We want to continue the pace of exponential innovation that humanity has been on. And to do that, we need new tools. So that's really our mission. Amazing. Well, look, thank you so much for taking the time to chat with me. This is amazing news.

19:08Super exciting. I love to see sort of like frontier technology being burnt out of Europe. It's about like Europe has actually got such a big advantage, especially with our huge sort of industrial history data sets. phenomenal talent and so look thank you very much and congratulations thank you so much so