232. How British AI is inventing new materials

0:10Hello and welcome to The Rest is Money with me Steph McGovern and with me Robert Peston. We had Lord Patrick Vallance on last week, Minister for Science, obviously former Chief Scientific Advisor to the government. And one of the things we were talking to him about, Robert, was how you start science businesses, tech businesses in the UK, how you set them up, how you scale them up and how you keep them. Yeah. So obviously what we wanted then to do is to go from what you might call government policy into what it's like on the ground in the UK to create a business that's potentially world leading.

0:45And so today, we are talking to an absolutely fascinating individual who has created genuinely an AI business that if it succeeds, will change so much of British manufacturing, British industry, even the climate. Yeah, so this is Dr. Chad Edwards, to give him his full title. So he is a co-founder and CEO of Cusp AI. He's a chemist by trade and I guess call him a deep tech entrepreneur. He had big roles at Cambridge Quantum Computing, Google's Open AI team as well. And last year, he set up Cusp AI with Professor Max Welling with the aim of using AI to design new materials for things like clean energy, semiconductors, water purification, etc.

1:39So we wanted to get him on to talk about what his business is doing and what it's been like setting it up and getting the funding he needed. So this is our interview with Chad Edwards. Chad, the reason we've got you in is because your business, even though it's quite a young business, is absolutely gripping because it's really pushing out the frontiers of how we create new materials. and materials are the building blocks of literally our entire world. So tell us about how you are applying artificial intelligence to effectively create new substances. Yeah, absolutely. I mean, we're a 20-month-old company, so still, as you say, Robert, very young in our journey, but things have moved extremely fast in our space.

2:28And I believe AI has enormous potential to revolutionize the way that we think about materials discovery. And so for those listening, I guess one of the most intuitive kind of descriptions we can use here is many of us would have witnessed in our own homes. We used to live in an era where you would generate or write text and you would look on Google's images to try and find the best image that would work for your presentation or whatever kind of project you were working on. And today we all know we've now moved to a new era, which is the generative AI era, where you can now generate entirely new images of exactly the properties that you kind of want.

3:01And the same analogy is now true in the material science world. For decades, scientists across the world, including myself at one point in my career, spent vast amounts of hours in a lab trying to discover new materials and come up with new hypotheses. And now we're moving to the materials on demand here, where you can truly specify the properties of the material you're looking for, and it generates it entirely on demand. So yeah, I think it's an incredibly exciting time to be in the field. And I think this has the potential to be one of the biggest impacts of AI. We're all very familiar with large language models and other models that we're using day to day for automation and writing emails, etc.

3:36But I think this is fundamentally going to change the world as we know it. So is it, to simplify it, and I know it's not as simple as this, Chad, but is it a case of, you know, you're saying, right, I want a material that's going to help purify water, but also be able to do all these other things. These are the properties I need, what are the different things that might mix together to create that? Is it that? That's exactly it. Yeah. Yeah. The perfect pitch, Steph. So yeah, we ask our users or our partners to specify the properties that they're looking for. It could be conductivity rates. It could be stability, synthesizability.

4:13We plug that into our engine. You can think of our technology as a search engine for materials. You plug that into our engine and it then uses the same underlying technologies we use for image and video generation to generate entirely new materials on the fly. Of course, one of the key differences in our space versus image generation space is you can look at an image and say that looks reasonable and sensible. That person has the right number of heads and toes. But in the materials world, you have to develop a whole technology stack to understand what the potential properties are, because you can't determine them just by looking at them.

4:41So then when it does, when the AI, your setup does work this out and it says, actually, this is a material that would do all these things. Does it give you like an ingredients list for it so that you then go away and make it or does it take that step of actually creating it as well? Yeah so we work really closely with labs across the world both commercial labs academic labs and other government labs to actually synthesize the materials as well so of course we have to ground our work in reality which of course is experiments and so we work with various labs to run experiments. Those experiments then generate data and that data is then fed back into our models.

5:16And the models are continuously learning both from computational data and real world experimental data. So your digital engine will come up with an ingredients list for your compound. You'll then get a lab to build the actual or to make the compound. You'll then test the compound. And then who owns the compound? Do you own the compound? Is this your intellectual property? Yeah, part of the IP is retained by us as a company. That's kind of a core part of our business model. And, you know, just to put this in examples, as well for people you know you've you've done some big partnerships haven't you with like Hyundai and Meta and things like that what what is the is it you're creating for example new materials for cars or how does it how does it work with partners like that yeah so we sit down with teams in some of the corporations that you've mentioned we co-engineer a prompt that goes into our engine that specifies the properties that they're looking for for their specific application so one example that you gave their stuff was Hyundai the the automotive company there we're working with them on future fuel cells for hydrogen powered cars.

6:18And so they're looking for new materials that enable them to have much more effective hydrogen power cars. And then with Meta, many of you will be familiar with Meta as a social media network. So many of you will be thinking, what do Meta have an interest in material science? It's kind of counterintuitive. But Meta have a small group in the FAIR group dedicated to material science. And And so we've collaborated with them since actually day zero of founding the company. And a couple of months ago, we released the world's largest data set for carbon capture materials in collaboration with Meta. And another one that kind of is in the headlines a lot at the moment is PFAS, so forever chemicals.

6:55These are chemicals that typically have lined things like our frying pans at homes and have now found their way into our waterways and, in fact, into our blood systems. And by very definition, they're forever chemicals, so they were designed to never break down. And now we have huge concentrations of them across the world. And so we're working with a very large water chemical company in Finland, designing membranes that can actually selectively remove these PFAS molecules from water. And in terms of carbon capture, are you quite close to these materials actually being used? Yeah. So, I mean, as a company, we're 20 months old.

7:31So we've had to build the team, build the technology and get it up and running. We do now have our first set of materials from a carbon capture project. So we've gone all the way from building our technology to generating entirely new materials, synthesizing them and then testing them. And do they work? I would say at the moment they're still undergoing rigorous testing. So we're still TBD, but the early results look pretty promising. I think for us, this wasn't a case of developing the world's best material in one shot. It was a case of getting our platform up and running and running through the process end to end.

8:00And of course, there'll be subsequent iterations of this project that enable us to get to state of the art, which I think we can do quite quickly. It is so incredible hearing these kind of real world examples of what you're doing. So tell us what it was like starting the business. Like, did you just have this idea? Because I know you've got, obviously, you're a chemist, but you've also got experience, haven't you, with quantum computing and work with Google's OpenAI team. So, you know, you're immersed in those two worlds. Was it just obvious to you that there was a need for this and there was a business idea there?

8:32Yeah, I think there was kind of a eureka moment, maybe about just over two and a half years ago. We saw, I saw a publication that came out actually from Microsoft. It was a model called Mattergen. And it was the first application where we'd seen these generative models actually be applied to material science. And for me, that was a big eye opener to say, well, now the world of AI is really going to transform how we think about materials discovery. And we can now move to this materials on demand era that we are working towards. and so yeah I just recognized that I had to be a part of those technological developments and recognized that it's had the potential to have a really big impact on the world I have two very young children at home and I often think to myself what will I say to them when they're when they're my age and I really want to be able to look back at my two children and say I didn't just sit on this on this plane recognizing the wreck that we're walking towards but actually I tried to do something about that and I think the technology that we now have available to us has the potential to change that trajectory and so yeah that was another really big kind of personal motivation for me starting the company.

9:32So tell me this Eureka moment, was it like, you know, like Archimedes in the bath or whatever? Was it, you know, what, what did, and then what did you then do when you, after having this Eureka moment? Because, you know, we talk a lot on this podcast about setting up businesses and then, you know, how you do that and then how you scale it up and everything else. So, so what happened after that moment? Yeah, I think, I think for me is I had a bit of a cheat because I've been through this once before I did it in my former company. So I helped build a company called originally Cambridge Quantum Computing, which then subsequently became a company called Continuum after a merger with Honeywell.

10:07And that saw the team go from when I joined, it was 12 people. And then by the end, it was 550. So it was kind of an incredible journey going from a small team to a very, very large team. And so I kind of knew the playbook in terms of the types of steps that you'd have to take to start building a foundational company like that. So the first thing I did when I realized that there was going to be a really big kind of transformation here is started to look for a technical co-founder. And kind of on the grapevine, I heard that a very famous machine learning researcher had just resigned from his role at Microsoft.

10:39And so I picked up the phone to him. I met him in a former life a couple of times. And I said, hey, Max, what are you? His name is Max Welling. And he's now my co-founder. I said, hey, Max, what are you doing next? And he said, well, Chad, I've got job offers to join the leadership teams of some of the world's biggest technology corporations, but I'm done with big tech. It's too slow and I really want to build another startup. And by the way, I've become very interested in material science. And he'd become interested in material science because he'd been working on Mattergem, which was the thing that I'd seen in the news that then triggered me to do this.

11:08And what's Mattergem again? Just tell us what Mattergem is. It's a generative model for generating materials, but it was the very first one that was kind of published from Microsoft. And so yeah, he and I joined forces in October, 2023 and then both well i left my my former company in in april 24 uh to start both working on it full-time and then we did a pretty big funding round we did a 30 million dollar seed round um and that again was catalyzed by the network that i had in it from my former from my former life um which at the time was the biggest in europe i think for a seed fund as funding round um of course the world has changed drastically since then and we now see increasingly even larger seed rounds being raised um and then not so long back we did 100 million Series A, which was led by tier one US investors, Temasek from Singapore, NVIDIA, and a bunch of other good strategic investors as well.

11:56So we want to talk about the development of the business, but since you are talking about fundraising, we've just been having a conversation with the government science minister, Patrick Vallance. One of the things we were discussing is not enough finance for businesses like yours is provided by UK institutions. And it is striking. You've just said that, you know, basically the big chunk of money that you've raised and$100 million is quite a lot of money. That came from a Singapore institution and American institutions, not British institutions. Why do you think companies like yours so often have to raise their money from abroad?

12:41I think there's a couple of angles to it. I think the first one is the availability of capital elsewhere feels much larger. There are much larger pockets of capital across the world, whether it's in the US or Singapore. We've got massive pots of capital here. They're just not being deployed to businesses like yours. And that's my second point, which is, I think, just to go through the real life example, as we were raising our Series A, I would be talking to European investors and it'd be a Friday afternoon. And then the European investors would be saying, let's regroup on Monday once we've got our leadership team back together.

13:13The US investors will work over the weekend and by the time it gets Sunday evening, you've got term sheets from US investors. And so there's a pace difference. There's a hunger. There's a hunger, yeah. And they will fight for deals. And if you're a competitive deal, they will fly to the UK to come and meet you almost there and then. So yeah, there's a real mentality difference, I think, in the way that investors operate across the world. So do you think, therefore, because that's really fascinating that the way they even treat the deal. It's not just about the money, it's kind of the love they give you in the process as well and that relationship build.

13:48Do you think the UK is a good place to be a tech entrepreneur then? And if not, what would make it better? Yeah, I mean, I think every country across the world, there's always things that can be done better. But I think if I think about the UK in my last journey, we built an extremely successful company. I mean, Continuum is now valued at the last fundraiser at 10 billion and will undergo some kind of IPO in the coming 12 months or so. That's public information. But yeah, with my current company, I think we're an AI for science company and the UK is incredibly strong when it comes to science and technology.

14:23You look at some of the biggest developments that have ever happened over the cause of humanity, from the development of the steam engine to the jet engine, And all of these things were developed here in the UK by British scientists. And so we're spoiled, I would say, for talent here in the UK. And when you're building a company of the kind of nature that we're building, talent is really the most important aspect. Everything out of capital and everything else, talent is the most important aspect. And so that provides a really good fertile ground for companies like us to get incorporated and go.

14:54But there is, I suppose, a sort of patriotic aspect to all of this. You made this powerful point that, you know, you want to be able to say to your children in 10, 20 years time that you've contributed to human welfare, that you've, you know, taken off, taken us away from a path of, you know, whether it's climate change, destruction or, you know, various other looming problems. But one of the things that we slightly agonize about, bring our hands about on this podcast regularly is just the number of great British companies that then end up being owned and controlled by overseas institutions or overseas businesses, which means that so much of the wealth that gets created as these businesses go from small to large actually flows outside of the UK and doesn't generate enough tax revenue or doesn't generate the tax revenue in the UK that it could generate and it doesn't create necessarily all the employment that it could generate.

15:57So this issue of where the finance comes from sort of does matter, doesn't it? I think so. I mean, very often in these conversations we get drawn into the capital. And of course, when you're building a company, there are many components that you have to get right to build a successful company. Capital is an important one. And the source of that capital is also important. As you're raising or building a young company, there's a signal by the types of investors that you raise from. So if you raise from a tier one Silicon Valley investor, the likelihood is that you're probably going to have a more successful company just by having that very investor associated with the company because subsequent funding rounds become de-risked for other investors if they see a certain brand of a fund is already invested in a company.

16:36So that's a dynamic that we have to kind of balance as founders as well. And probably another reason that we see the most successful companies going to raise outside of the UK is because the tier one US investors are in fact based in the US. But I think as capital aside, I think the other really important thing that we're kind of missing here in the UK, and now starting to get more of, which is why I'm really optimistic for it, is actually really good examples of companies that can go out and do this. And once you get a certain company that gets to a certain stage, that then starts to leech talent who then are inspired by the journey that they've been on.

17:11And they then have the knowledge to go away and do their own company. And they know that, like I have, I've been through this once before. And once you know it, you can go again. And I think we now have a few really good examples of this in the UK. Arm would be one that comes to mind, Eleven Labs, Synthesia, Continuum, my past company. So these companies will start to create talent that now kind of know how to play the game. And I guess DeepMind has been one of the most influential technology companies in British history. There are endless kind of fantastic companies that have now spun out of DeepMind as a result of the journey that they've been on.

17:45And they were acquired by Google. And people said, should they have been acquired by Google? But the net impact on the UK has been phenomenal. So the British investor Saul Klein has actually just written a paper, which actually Patrick Vannis just again brought up, saying two things. One is that he thinks that certainly in terms of Europe, the UK is by far the best place to be doing the kind of thing that you're doing. I just wonder if you agreed with that. He also thinks, however, which is also a remarkably positive thing to say, that he believes that the changes that are going on here in terms of the development of a whole range of technologically advanced companies, not just in AI and not just in quantum, but across life sciences and the rest.

18:30He thinks that the conditions here are as good as they've been since the late 19th century. Do you think his optimism is well-founded? Yeah, Sol's actually an investor in our company. He's backed us from the very beginning of our journey. So, yeah, a lot of what kind of Sol advocates for resonates very strongly with our journey as well. But maybe in some further context, so my co-founder, Max Welling, is a Dutch machine learning researcher. So he's based in Amsterdam. So in some ways, we're kind of an Anglo-Dutch company, and we can benefit from both the Dutch ecosystem and the British ecosystem.

19:02You're the new Royal Dutch Shell. Chad I was just going to ask you a bit about um you you're clearly as you say you know business minded learnt loads from the Cambridge quantum computing business you were at was that but you started in as a chemist didn't you so was was the business stuff innate was it taught as you were doing all the chemistry like what where where's that come from because I think one of the problems we have is we often silo people, don't we? And go, right, you're a creative person, you know, go and do your creative stuff, but don't think about the business side. Or you're a scientist, right?

19:37You're going to discover things, but just focus on the pure science. Don't think about the commercial side of things. So was this something driven by you, the interest in the business side of it? Or was it something you were taught or brought into? Yeah, I mean, I started my career as a chemist, did a master's in chemistry, then a PhD in chemistry. And then at the end of my PhD, decided I wanted to go off and do something more commercial. And so I went to business school and did an MBA at the Manchester Business School. And that gave me a really good grounding in all things business. And really, it just taught me that business is largely about relationships and strategy.

20:12And I saw a really nice kind of convergence of my skill set, bringing together my PhD in chemistry and now the commercial acumen. And then went on to join my first company, which was Cambridge Quantum Computing as the commercial co-founder. When I joined, it was a a bunch of very smart people in an attic of a university building. They had no products and they had no customers. And my job description was quite literally figure out how the hell we make money from this thing. And so, yeah, it was kind of an incredible journey. And I just learned a lot by doing and managed to form relationships with some of the world's biggest companies like the Totals, the Roches, the BMWs, and actually found that there aren't many people that are able to translate or sit at this kind of technical commercial interface and translate in both directions.

20:56So on the one hand, you need people that can translate deeply technical topics like AI, like quantum computing, to big companies like BPs and the BMWs, et cetera, of this world. And on the other side, you need deeply technical people to understand what the real world commercial potential of that technology is. And so that translation is really important. And I still today don't think there are many people who are able to bridge that gap. And we need to encourage more people to kind of sit at that interface because it's really, I think, the key to making a successful company. When I founded CUSP, I said we have to be commercial from as early on as we possibly can.

21:31So we've always centered ourselves around customer problems and real world problems as opposed to doing these things in isolation, which I think risks you becoming kind of more of an academic exercise. Can I move us on to a slightly more general question because it'll Help us to understand more about what you do and what, in a sense, businesses in your bit of the AI world do. Because we are living through this artificial intelligence industrial social revolution. People tend to sort of generalize a bit about AI. Explain to us the difference between your kind of artificial intelligence project and, you know, an open AI or a Gemini.

22:12the large language model generative AI businesses projects that are on everybody's minds at the moment? Yeah, I think there's a kind of technological answer here, and then there's a more societally focused answer that I could give. So if we start with the technology, actually, fundamentally, we do rely also on large language models inside of our tech platform. So we have the benefit of having the last 50 to 100 years worth of scientific literature that have been published by researchers all across the world. And now we're in a regime where you can start to use the last 100 years worth of... And that's all in your database.

22:48You can now extract that using large language models and you can query all of the scientific literature almost immediately using these models. Now, of course, that's not available to everyone. You have to form relationships with these types of companies to access that data. It's kind of locked down. But technological-wise, we do use that same technology in certain domains. And then in the generative side, we do also, as I mentioned at the beginning, we do also use generative models. So these are the type that take a prompt and then generate entirely new materials. So technology-wise, I would say we are using very similar approaches.

23:22There are some nuances, of course, because we have to then run physical simulations to understand what the potential properties of those materials may be, which don't happen in a large language model. Large language models are, by definition, language. which they predict next token. And we have now designed models and built models that are kind of chemically or physically aware of the material science world. So we've gone on and built an entirely kind of proprietary stack dedicated to that. Is this closer to what people call world models? In a way, yes. I mean, yeah, world models are kind of more depictions of places like this room, for example, and how we would walk and navigate in a real world context.

24:01Ours is kind of more what we like to call physical AI. So this is the being of matter and physics. Jensen at NVIDIA is very pro-physical AI right now. He talks a lot about humanoids and self-driving cars and robots, etc., robotic labs. And so I see this as the next era of AI. We've been through the large language model. We're now moving into the physical AI era. And the UK didn't have a position in the large language. But we didn't have a horse in the race for the large language model race. There are now, of course, two dominant players in the world, I would say, and there's Mistral here in Europe.

24:36But I think we have an opportunity to be at the front of this kind of AI for physical world. And I think that plays exactly into the strengths of the UK. And when you're, you know, we mentioned some of your partnerships, customers, whatever you want to call them earlier on, you know, for example, in the automotive sector. are you how are you finding it in terms of um working with sectors and their understanding or even willingness to take on ai like i imagine there are probably somewhere you could see materials ai generated material solutions for them but they're a bit like whoa whoa whoa that's not how we do things here are you finding that and if you are are there particular sectors where it's trickier to to work with them than others yeah so maybe on just answer the second part of Robert's previous question, which was around the impact and how we differ from the LLM world.

25:26I think very often we see in the news now that AI is going to displace jobs and take jobs. And maybe in certain spaces, maybe customer service and other areas like that, that is in fact the case. And you can now see it day to day, even more junior players at legal firms arguably are being displaced by large language models. In our world, it's not so much the case. were actually a tool for enablement, a tool to accelerate scientists and researchers to do their work. And so I don't see the same level of kind of societal threat from the work that we do relative to some of the other technological advances.