Node to Node

Dr. Claire Thorne:
Hi both.
Professor Simon McIntosh-Smith:
Hi.
Dr. Claire Thorne:
Hello and welcome. I'm Dr. Claire Thorne from Deep Science Ventures and this is Node to Node, conversations from the frontier of UK supercomputing. The video podcast that delves into AI supercomputing and the UK's capabilities. We'll be covering all things compute. Sovereignty of course, but also AI as an infrastructure for science, for powering breakthroughs in science, and asking the question around who gets access to its power and who does it benefit? Node to Node is brought to you by the Bristol Centre for Supercomputing, part of the University of B ristol, which built and operates Isambard-AI. It's the flagship compute of the UK's AI research resource funded by the UK government's Department for Science, Innovation and Technology, and it's providing researchers and SMEs with free access to advanced compute to train and test their models.
Isambard-AI is the UK's most powerful and the world's 13th most powerful supercomputer. In this special episode to mark the launch of Isambard-AI one year ago, we'll be sharing some examples of how AI is already changing lives across the UK for the better. I'm joined today by Professor Simon McIntosh-Smith, the founder and director of The Centre and really the godfather of Isambard-AI. Welcome, Simon.
Professor Simon McIntosh-Smith:
Hello.
Dr. Claire Thorne:
And also Emma Hindley, MBE from the UK Government's Department of Science, Innovation and Technology. Emma is deputy director there, responsible for public compute and the AI research resource. Welcome, Emma.
Emma Hindley:
Hello.
Dr. Claire Thorne:
Simon, firstly, what do we mean by AI supercomputing?
Professor Simon McIntosh-Smith:
Well, AI is this amazing phenomenon where we need massive amounts of computing power and it's really why AI has started to be a real phenomenon today. People can run some AI on their own gadgets, on a phone or on a laptop, but actually to do the really breakthrough parts of AI, for example, training a large language model, we need phenomenal amounts of computing power. That's something you'd never be able to do at home or on one of your own devices. So that's what a supercomputer is, something that's way beyond. Probably thousands, maybe tens, even hundreds of thousands of times more powerful than anything you might have at home or in your department at work or in a university.
Dr. Claire Thorne:
And when we talk about more powerful, are we talking speed, speed of calculations?
Professor Simon McIntosh-Smith:
It is largely down to speed, and in fact, the supercomputer we're talking about today isn't about AI. It's unbelievably mind-bogglingly powerful. Just to give you some idea, if you've got everyone on earth to do one little mathematical calculation every second, it would take everybody on earth, all eight billion people, doing something every second, 24 hours a day, 365 days a year. It would take them 80 years to do the same as what Isambard could do in just one second.
Dr. Claire Thorne:
A lifetime's work for all humanity. I love it. And Simon, getting to this point with Isambard-AI, one year on, one year of operations. I mean, it's a first of its kind mega project for the UK. How did you get here? What did it take?
Professor Simon McIntosh-Smith:
It's taken a lot of hard work. Some crazy ideas. I mean, it was a bit of a crazy idea right at the very beginning and a fantastic team of people. And really it's the team that's made this possible. We've been very lucky. We've been able to build a very, very strong team at the Bristol Centre for Supercomputing, but we were remarkably well-supported by University of Bristol. We've got some brilliant partners. So HP actually built the supercomputer for us. NVIDIA provided the kind of core components of it, which are the GPUs that we use. Even the people who built our module data centre for us, Contour, they all did remarkable jobs. We challenged everyone with a very ambitious vision of what we wanted to achieve. This one of the world's biggest supercomputers and built in record time. And everybody was up for the challenge and we hit lots of little snags, but people kind of helped us overcome that to deliver it in record time.
Dr. Claire Thorne:
And there's something special about the way Isambard-AI is powered as well, isn't it, it's energy source. Can you tell us a bit about that?
Professor Simon McIntosh-Smith:
We really care about sustainability of what we're doing. We don't want to provide a large supercomputer at any cost to our neighbours and to the planet. So designed the most energy-efficient computer that we could. And in fact, when we launched Isambard-AI, it was the second most energy-efficient supercomputer that had ever been built anywhere on earth. But even the way we power it, so we pay a little bit more for a tariff that's entirely UK renewable energy. So it's only solar, wind, hydro or battery. That's where all of our energy comes from.
Dr. Claire Thorne:
Yeah, thank you. When I think about data centres and supercomputers, there's two things that spring to mind automatically. One is how is it powered? The big energy question. And the other thing is sovereignty and who gets access, and we'll come onto that in a bit. Simon, what does this mean? What does Isembard AI mean for the UK?
Professor Simon McIntosh-Smith:
Before we had it, there were lots of things that we just could not do as a country for AI. So we couldn't train a large language model. Any of the big uses of AI, we just couldn't do it. We didn't have the way to do it in the UK. Once we had Isambard-AI, lots of very important, very valuable, strategically important things for us as a country became possible again. So it really was a significant threshold that we've crossed over with this new capability.
Dr. Claire Thorne:
You've crossed the threshold, but you've also built this incredible user base. Can you tell us about who's using Isambard-AI today?
Professor Simon McIntosh-Smith:
Yes. So most of the users come from AI research and development, so many of them might be academics, but actually lots of our users also now in companies, especially in small and medium-sized enterprises, SMEs and startups, lots of very exciting AI startups in the UK, and many of them have been awarded time on Isambard and now they're kind of racing ahead doing things that they just couldn't have done otherwise.
Dr. Claire Thorne:
You mentioned before just earlier there about the pace of which Isambard was built and operationalized. It's incredible, but also I think the pace at which you built this user base and researchers and SMEs are adopting it. So is it 4,000 users doing a thousand projects at the moment across the UK? Is that right?
Professor Simon McIntosh-Smith:
That's right. Up to 4,000 use already, even though we only opened just about a year ago. So that's a remarkable ramp up. It's way faster than I think any of us were expecting, but I think it really shows how much demand there was. It was pent-up in the UK waiting for something like this to come along.
Dr. Claire Thorne:
What does it mean for the sorts of questions that we're able to answer? And maybe you could point to a use case to give us a flavour of what Isambard-AI is doing.
Professor Simon McIntosh-Smith:
Yeah, there are so many exciting things. We're often talking amongst our team and every week we're hearing about a new project that's starting up where the hairs on the back of our neck literally stand up. They're just incredible. I've been in this sort of area for decades and there's always been lots of exciting stuff, but this really feels next level what's going on now with AI. People doing lots of projects that relate to medicine and healthcare, people looking for new drugs that will either protect you from disease or treat disease. We'll say a bit more about those a little bit later on. But one of my personal favourites, actually an AI use case in agriculture, this is not maybe one of the first areas you might think of when you think of AI as all kind of high-tech and shiny.
There's a project actually for some colleagues of mine at the University of Bristol using pretty cheap cameras, video cameras dotted around a farm, effectively cheap webcams, and using them to basically keep track of a herd of dairy cattle and keeping an eye on their health and how they're doing because it turns out cows are actually quite secretive when they're not very well. They kind of hide it from everyone else so they don't get predated. So it's actually quite hard to tell when cows aren't very well, but these cameras have been trained through the AI to spot when a cow actually might be a bit unwell so that the farmer can be alerted and they can be treated quicker than they might otherwise have been. So it'd be good for the health of the cows, good for the economics of the farm.
Dr. Claire Thorne:
In a really non-invasive way for them as well and a way that could be potentially scaled to other farmers and other communities, yeah.
Professor Simon McIntosh-Smith:
Exactly.
Dr. Claire Thorne:
Fantastic. Thank you. Emma, if I can come to you, why was building Isambard-AI and getting it up and running such a priority for UK government?
Emma Hindley:
Well, I think we've already heard a little bit about the kind of use cases and the science that can be done on Isambard-AI that was just not possible to do in the UK before. And so there's two reasons really why this is a real priority for the government. Firstly, accelerating British science and accelerating those kind of breakthroughs that can make a real difference to the way that people live and operate a real positive change for people's lives. And then also the growth that comes with that. So once these discoveries are made, we've just heard about how agriculture can be improved through the use of this AI technology. There's also new materials that can develop new products and there are businesses that then spin out of that or there are efficiencies that come from that. And so we've got both scientific advancement and economic growth from this amazing technology.
Dr. Claire Thorne:
What I'm hearing is that this is really a signal with intent for the UK. The fact that the UK couldn't do this before, it didn't have the capability, it's invested in it, it's got it up and running with record time. Now here we are and we can do all these amazing new things. So it's almost like a signal of the UK's technological leadership as well.
Emma Hindley:
That's right. And the UK I think for a long time has had huge strengths in science. Our academic institutions are very, very strong and excellent. And so being able to provide this sort of technology that just gives a lift to all of those sectors is fantastic.
Dr. Claire Thorne:
Yeah, that's fab. Thank you. I think for me, Isambard-AI is both critical national infrastructure and its scientific infrastructure, which is what you talked about there. It's both of those things. And of course to do both of those things, we all often talk about sovereignty and how do we do this in a safe, secure, reliable way. And I think maybe two or three years ago we weren't even talking about sovereignty and now it's the number one priority. It's everything. It's the economy, it's inclusion, it's everything. So Simon, is Isambard-AI, is it sovereign AI?
Professor Simon McIntosh-Smith:
Absolutely, and that really was one of the driving forces behind creating this capability. I think the geopolitical situations obviously got a lot more complicated over the last few years and things that we didn't really have to worry about in the past we do have to take seriously now. Even just in the last few weeks have been some incidents where access to some of the latest AI models was suddenly switched off. Now if you were relying on that for your business or you're relying on that for some government-critical operations, that could be really disastrous. So I think governments all over the world are looking at this now and thinking, "You know what? If AI is really as important as we think it is, we probably need the ability to be able to produce that AI ourselves and run that AI ourselves as well."
Dr. Claire Thorne:
Simon, reflecting on Isambard's first year of operations, what has been the most exciting outcome for you, the most exciting output personally?
Professor Simon McIntosh-Smith:
Well, firstly, I can't believe it's already been a year. I mean, it feels like weeks rather than months or a whole year. It's gone incredibly quickly. And initially we've got a five-year mission, a bit like Star Trek, five-year mission, and we're already 20% of the way through, which is just incredible. But so many things have already happened. I think right at the very beginning, one of our first users was from UCL, a project called UK LLM, and they were training one of the UK's first homegrown large language models, and that meant they could do things like building the best support for British languages. So things like Welsh and Scot's Gaelic, all sorts of languages that might be overlooked by very large international large language models. And they do a brilliant job of that now. But also things like English case law, another British case law, which again, don't feature. They're not quite so important to LLMs that were trained elsewhere.
And that project was waiting to be able to have access to a proper AI supercomputer, and as soon as we switched to Isambard on, they were jumping on it and filling it up with doing this, and that was fantastic. I think even at the beginning before we were officially open, we were making a big difference for a very important project.
Dr. Claire Thorne:
That's great. It's such a great example of that project of how you can use AI technologies or AI infrastructure to support and sustain cultural diversity across the UK is fab. Do you think Isambard-AI will win a Nobel Prize?
Professor Simon McIntosh-Smith:
That's a great question. It's a very sort of meta question about AI itself, isn't it? One day eventually maybe they will, but I think initially what's much more likely is they'll support somebody else getting a Nobel Prize. I think that's quite likely. We're already seeing people use AIs very much as AI scientists. So they're almost co-pilots with you on your journey. In fact, our own team is doing this. We're using it a lot ourselves developing the code that's running Isambard way more than even we expected a year or two ago. So those technologies are maturing incredibly quickly and what we'll be sitting here talking about in one or two years will probably blow our minds. So I think it's only a matter of time.
Dr. Claire Thorne:
That's the bit I'm personally interested in is what does this do to the nature of science itself, how we do science, what does it mean to be a scientist, and what does it mean to work alongside these tools. What worries you as well from the first year of operations, what you've seen in application areas perhaps?
Professor Simon McIntosh-Smith:
Something that we're actually seeing a lot more of is exploits about latent bugs that have been in this kind of software we all run on supercomputers. So in the past, previous smaller iterations of Isambard, we've only seen these come up very, very rarely. They're almost coming out weekly now. It's possible to use the power of AI to really search for bugs that have actually been in the code for decades in some cases, but no one's ever spotted them and turned them into an exploit that you could use to actually gain access to a supercomputer that you shouldn't have, maybe do some damage that you shouldn't have. So we're having to be really on the case watching out for these things and reacting way faster than we've ever had to before. So that's something that worries me in particular.
Dr. Claire Thorne:
Thanks, Simon. Emma, Simon talked a little bit about the farming application, the farming project supporting cow health and also UK LLM around minority languages. I know you've seen a lot of the case study videos, you enjoy them. Which ones resonate with you?
Emma Hindley:
So I think the one that immediately leaped out to me was a company called Prima Mente who are looking at how you can create models that can help you to understand or help doctors to understand how the body is developing through illness processes and try and predict those or prevent those earlier. So in particular, at the moment they're looking at the brain and looking at ways that you can predict and prevent Alzheimer's or how people can live their lives in ways that delay the onset of Alzheimer's and doing that at a level that has such greater detail than has been possible before. So you can look across a huge range of different cell processes using these enormous data sets on Isambard-AI and then spot patterns, spot interactions in a way that would never previously have been possible.
And that I find particularly exciting because actually before I joined the civil service, I did a PhD in neuroscience and I was looking at... I was doing fundamental research, but with an eventual plan that that would've helped people to detect Alzheimer's more quickly. And I spent three years advancing knowledge by a fraction of what Isambard-AI and Prima Mente can do in an hour. So I mean, it's just incredible the transformation.
Dr. Claire Thorne:
I think this is the common thread that keeps going back in all of these application areas, all of these projects and use cases, is that it's the speed at which some of these scientific questions can get solved now.
Emma Hindley:
And the scale. I think they're scared of the data and so the detail that you can get into it, you're not looking at one marker, two markers. You can look at a combination of 20, 100, 1,000 markers and then see things that previously wouldn't have been possible to do.
Dr. Claire Thorne:
For me, when I look at all the projects all on the YouTube channel for BriCS, there's two that really resonated with me and they both happen to be from AC, so the AI Security Institute, UK Government's Security Institute. One was around data poisoning in the pre-training of LLMs, and the other one was around developing AI personas in pre-training. And the first one resonated because I thought it was a really lovely example of this public-private trilateral that's required. So it was Anthropic and Alan Turing Institute and the AI Security Institute, and I thought it was a great demonstration of that. And I thought what was shocking about that one is that it's much easier to poison these LLMs, these commercial LMs than we all originally thought. So I wonder if you wanted to speak to that one, Simon.
Professor Simon McIntosh-Smith:
Yeah. Again, remarkable work and something that was quite a surprising result for everyone, and I think that's when something has real value. I think we may have all assumed that it was quite hard, you had to work really hard and maybe change a lot of the data going into a large language model to poison them, and the AC discovered it's really tiny, just hundreds of documents and that's it, which is something that's really useful to know because it means we can guard against it, we can build in different guardrails, we can be aware of it, which I think is also very, very important.
Dr. Claire Thorne:
Lots of issues of trust and risk and resilience all wrapped up into that one use case. And on the other one, the one around developing AI personas, so how AI should act in good or bad ways at the pre-training stage I thought was a really great example of ensuring that AIs can cooperate with humans and that they don't go off on their own fully autonomous. Emma, did you reflect on that one as well?
Emma Hindley:
Yeah. So in this example, AI was trained with examples of other AIs and how they behaved. So for example, if you tell an AI during the training stage that terminator is the way that AI should act, you find that your AI tends towards less cooperative, less helpful to humans ways of operating, whereas if you show an AI examples of where AI is helpful to humans and works along the same ethical frameworks, then you have an AI that has a personality that is more along those lines. And so knowing that actually AI can develop personalities in this way I think has huge implications for how you both train AI and how you interact with AI to make sure that it stays on a path that is a more helpful and a real tool for humans.
So I think just that kind of... I don't know that anybody... Maybe they do, tell me if I'm wrong, but I don't know that anybody really understands how these kind of effects come out. And so knowing that those effects are there then enable you to think about in a deeper way about how do you train the models to make sure they're coming out with the kind of personalities and tendencies that we want them to see even when that is unexpected and unpredictable.
Dr. Claire Thorne:
What struck me is I do a lot of work on inclusive education pathways and opening access opportunity for all, all of that stuff, and what struck me is that in that work, we know the importance of role models and we know the importance of doing things early on in the pipeline, and it was the same things we were learning in this project here and the importance of putting positive role models, not the terminators in the pre-training phase and doing it before it's too late. So I thought that was a really interesting parallel. And the other thing about that project I guess is that we know that there's this... We're all endeavouring to build safe and inclusive and responsible AI, but actually neither safe nor responsible AI is possible without inclusive AI. So you need projects like this to be building that in at an early stage. So I love those two. Simon, did you want to point to any of the other use cases? Maybe some around healthcare.
Professor Simon McIntosh-Smith:
I think this is actually the largest area we're seeing in general on Isambard. It's possibly an area where there's so much rich data that we really haven't been able to extract all the learning and wisdom and knowledge from where AI is now finding things that we'd never seen. We've got projects looking at heart health, trying to understand what's going on at the cellular level, trying to come up with new drugs to treat heart disease, lots of projects looking at developing new drugs that will treat all sorts of different diseases. I think this is an area where over the next five, six, seven years, we'll see all sorts of remarkable breakthroughs that we'll be looking back at, going, "Do you remember when in the past we had to worry about these sorts of things and now actually that's been resolved?"
Dr. Claire Thorne:
Yeah, I think in the healthcare space, how the UK is uniquely positioned is that we have the NHS, so we have a source of data that is kind of unparalleled, and so I suppose there's a role for Isambard-AI here as being a UK supercomputer that is trustworthy and safe and reliable and responsible to deal with that data. Are you seeing that in your project cases?
Professor Simon McIntosh-Smith:
Absolutely. And we also actually get a lot of feedback from our international peers. So when we go and meet our equivalents in America or Europe or Asia, they all say they're really jealous of the kind of rich data that we have in the UK from the NHS. There is really nothing of that scale, of that longevity anywhere else in the world. So that really is a unique sort of national asset that we've got that we can really benefit from.
Dr. Claire Thorne:
And Emma, the use case you mentioned before Prima Mente on Alzheimer's research, that's a great example, isn't it, of using NHS data but in a trusted way through Isambard-AI?
Emma Hindley:
Yeah. So the project is being done in collaboration with clinical doctors in I think quite a lot of different hospitals across the country where they're taking blood tests and able to look at markers within the blood. And so that data is obviously sensitive. People don't want the health information to be shared or accessed. And so the fact that Isambard-AI has this level of security and safety for that data is really important that enables these projects to go forward and create these breakthroughs that could create huge change for the lives of people who do not then go on to develop dementia because we're able to use that data and deal with that data in a kind of secure and safe way.
Dr. Claire Thorne:
And we are seeing that with Nightingale AI as well, Simon, is that right?
Professor Simon McIntosh-Smith:
That was another one of the very early, very large projects actually led by Imperial in London. Again, bringing together data at scales we've never really seen before and combining it in ways that just hadn't been possible to come up with new breakthroughs. And again, even those projects, they're already making breakthroughs. They feel like they were very much at the beginning, and as they really get going and they really start to exploit the data they've got and the techniques that are now possible, they're going to do things we haven't even imagined yet.
Dr. Claire Thorne:
Yeah. So in terms of projects that involve NHS data sets, how is this really going to benefit individuals? And maybe Simon, you go first.
Professor Simon McIntosh-Smith:
I think for the first time we're now going to be able to really take the very rich data that we've got in the UK and learn things from it that can benefit individuals. So we'll be able to come up with new ways of understanding what's really wrong with someone and coming up with treatments that'll be really optimised for them and that might be-
Dr. Claire Thorne:
So really personalised healthcare.
Professor Simon McIntosh-Smith:
Really personalised. It might be a new kind of cancer treatment. It might be a new kind of drug that would treat them.
Dr. Claire Thorne:
Yeah. Thanks, Simon. What about you, Emma? How are you imagining that the NHS dataset as an asset can really benefit individuals?
Emma Hindley:
So I think that the NHS dataset allows you to look at disease in a much more holistic way. So previously much research would focus on a particular biomarker and if you've got this biomarker in your blood, what might that mean? And now instead of looking one-to-one, we can look at a far broader range of things that you might not even have... Scientists might not even have made that connection themselves yet and AI will be able to spot those patterns.
Dr. Claire Thorne:
But still at an individual level.
Emma Hindley:
At an individual... So you can use those patterns to then predict what's going to happen to that individual. If you can look at that variation within the population and understand it, you can then really personalise treatments and choose the chemotherapy drug that gives you the fewest side effects and that attacks the tumour more effectively so we can really narrow in on the individual and give them a personalised treatment that would not have been possible without that broader understanding of how all of those elements of someone's individual biochemistry are interacting.
Dr. Claire Thorne:
So personalised pathways, personalised drugs and treatment plans, and also perhaps interventions that happen at the right moments.
Emma Hindley:
Yeah, I think much earlier treatment becomes possible because AI is able to spot things that might not be visible to a human radiologist or a scientist looking at it and can see patterns much, much more quickly than we might be able to. So I think that it's an enormous tool that can support NHS doctors as they are making their treatment plans to have just a much more rich and rounded piece of information about that patient that they've got in front of them at the time.
Dr. Claire Thorne:
And we're already seeing this not just at an individual patient level, but we're really seeing this with applications around drug discovery, aren't we, Simon?
Professor Simon McIntosh-Smith:
We are. And I think there's been pretty much a revolution in how drugs are now being discovered. I mean, the phrase in silico where much more of the work's now being done in simulation on computers are now more and more driven by AI where AI again can consider many more options and all sorts of things that human experts might not have considered because they might have been a crazy idea. AI could just think of all of them.
Dr. Claire Thorne:
These virtual wet labs where perhaps they maybe not have considered it, but also maybe didn't have the facilities or the funding to be able to do it.
Professor Simon McIntosh-Smith:
Yeah, some of them even physical wet labs where an AI is now deciding what the lab is actually going to do and AI driving a lab.
Dr. Claire Thorne:
Emma, if we zoom out a bit in terms of public health society, what benefits does Isambard-AI enable?
Emma Hindley:
So I think it's enabling us to spot public health problems and issues much, much earlier. So there is research being done into respiratory illnesses. There's research being done into child mortality. There's an enormous amount of work that's being done. And even into non-health elements that affect human health, like pollution, so that we can predict much more accurately what's likely to be happening. So we are able to... Based on some of the information and research that's coming out of Isambard-AI and similar AI research, we're able to target interventions much earlier and much more accurately that then prevent people from getting into the later stages of illness or even from getting that illness at all.
Dr. Claire Thorne:
Yeah. I want to look forward and see what's next for Isambard, but let's just pause a minute and think about when did your journey start with AI and supercomputing, Simon?
Professor Simon McIntosh-Smith:
Gosh, so I think I first learned about AI in about 1989 when I was doing my computer science degree decades and decades ago. At that point-
Dr. Claire Thorne:
That's mad. It's pre-Google.
Professor Simon McIntosh-Smith:
It's pre-internet, yeah, all that sort of stuff. But it was very much a mature field even then, but there wasn't the scale of computing that we have now nowhere near, and nor was there the scale of the data. So those are the two things that really brought about the AI revolution that we're seeing now. And then we started building Isambard supercomputers at a much smaller scale more than 10 years ago now, and that's been long-term government funding, exploring the art of the possible so that when we came to the point where something of the much larger scale that Isambard-AI required, we were ready to go and do that very quickly.
Dr. Claire Thorne:
I think that's something that we should really celebrate here, this long-term thinking, this long-term commitment from UK government and from other partners as well into building this kind of infrastructure. So as you say, it's like it's a decade... More than a decade long journey. Emma, can you give us a sense of the current UK AI landscape? Because yes, there's a long-term commitment, but for me, almost every day, every week, I'm seeing a stream of new strategies or new investment announcements, new roadmaps, et cetera, announced. So it's a busy landscape. What is it looking like at the moment?
Emma Hindley:
It is busy. I think that it is really clear that this is a generation-defining technology and that this could be revolutionary, and so it's really important that we are able to take advantage of that incredible transformation that this enables. My own job is to do with expanding access to the AI research resource of which Isambard-AI is our biggest part. So we are looking at expanding that to bring online a cloud system and building another supercomputer in the next few years, but another really big machine to be able to do a lot of these future workloads. So like Simon mentioned with the duplicate AI scientists to assist. So we're going to be trying to build machines that are more able to support those type of workloads to let more and more scientists and businesses take advantage of this.
Then more broadly on the strategic level, we're thinking about how do we encourage the growth of AI and usage of AI through businesses through the UK economy to drive growth in the sector? So that includes things like AI growth zones, particular areas around the country where the government is trying to attract in private sector investors and businesses who are using AI to generate the data centre connections, the expertise to drive AI usage. We also have the AI hardware action plan, which was announced in June 26th, which is looking at how do we support British chip design and other British companies in particular who are working in the space of designing the hardware that goes around supporting a computer of this sort.
And so there are a whole lot of plans to support those businesses from the very earliest stages all the way up through to the point of testing their chips in what's called the Scaling Inference Lab within ARIA to help make sure that the British companies who have really fantastic ideas, including chips that use far, far less energy, chips that are far quicker, that use light instead of electricity to communicate across the chip itself, that these ideas are able to progress and turn into the kind of game-changing companies that they really could be.
Dr. Claire Thorne:
It felt like the Bletchley Park AI Summit, the AI Safety Summit in '23 was a big turning point. It's when the seeds of all of these things started really, and it's when the Isambard-AI announcement was made and development obviously happened straight after. You mentioned the Scaling Inference Lab at ARIA, the Advanced Research Invention Agency, the AI Growth Zones, five of those across the UK. It was announced in '25. We've had the D6 AI for science strategy more recently. Obviously the AI Opportunities Action Plan before that, the AI Security Institute that we mentioned a few times, there's the incubator for AI inside government as well. With all of those things happening or in motion, how central is Isambard-AI to that?
Emma Hindley:
Really critical. Really critical. So I think Isambard-AI was the first real test case of building a supercomputer of this scale in the UK. And the fact that we've seen this incredible research coming through already... We started off doing analysis of the value that the computer has created and it gives a huge value back, and that research was done a few months ago when it wasn't completely full yet. We weren't using the full capability because as Simon says, this has grown so rapidly. And so we're already seeing fantastic science coming out of it, economic growth coming out of it. And so I think this has really proven what everybody thought would happen. Now we're seeing it happen, and so we're seeing how important this is for the scientific community.
Dr. Claire Thorne:
Simon, what's next for Isambard-AI?
Professor Simon McIntosh-Smith:
Well, we don't rest on our laurels, that's for sure. So we've actually been very busy doing a whole bunch of things that are going to come next. The service itself continues to evolve and grow. So for example, we're about to add something called an inference service. So this is the kind of AI service which is the thing that will enable that kind of AI scientist that will sit alongside you. We're able to serve tokens out on a range of different models to a whole range of different users. So that's something. We've already got in sort of prototype form now, but that will roll out quite soon. We're about to add a very large, incredibly fast data storage facility called the Bristol AI Data Facility, or BRAID, that's actually been arriving today. So that will be deployed a bit later this year in 2026. So that'll be very exciting. And we've also got ideas and plans, what we'd like to do next, what would come after Isambard-AI at sort of even larger scale.
Dr. Claire Thorne:
Because this version of Isambard is not the first version, right? There's been four versions before that, is that right?
Professor Simon McIntosh-Smith:
Yeah, this is the fourth one. That's right. So we've always got ideas for... We're always learning. Every time we do one, we learn and we know how we could do it bigger, better, faster, lower risk, more cost-effectively next time. So we learn with each one that we do.
Dr. Claire Thorne:
Could Isambard-AI version five, 5.0, be the world's most powerful?
Professor Simon McIntosh-Smith:
It could be. I mean, that's largely a function of how much money you're willing to spend, quite frankly.
Dr. Claire Thorne:
We're looking at Emma here.
Professor Simon McIntosh-Smith:
It needs power. Actually, the fastest supercomputer in the world right now in the middle of 2026 is in China and it uses about 40 megawatts of power. 40, four-zero. That's an awful lot of power. We actually have a route to getting 50 megawatts right next door to Isambard-AI. So we could build something that was even bigger, one big computer that would do that. Whether that's something we really want to do or not is quite a big question. But I think at least we have that capability. We could deploy that in lots of different ways to the benefit of the UK.
Dr. Claire Thorne:
You have the power, you have a potential site for a bigger Isambard.
Professor Simon McIntosh-Smith:
We've got the site right next door to where Isambard is today.
Dr. Claire Thorne:
So what are the other limiting factors? You mentioned funding, of course. Anything else?
Professor Simon McIntosh-Smith:
And people. And we've been very fortunate. The Bristol Centre for Supercomputing, we've grown from really nothing to will be about 35 people in the summer of '26 and we'll keep growing slowly, but we've been developing ways of building these facilities and operating them very, very efficiently in terms of the team that we've got, exploiting lots of techniques from the cloud space actually. So we don't need a massive team to keep growing, keep adding more facilities to the portfolio that we've got today.
Dr. Claire Thorne:
And you mentioned the inference service that you're going to be launching soon. Can you tell us a little bit more about that? I'm really curious.
Professor Simon McIntosh-Smith:
Yeah, you're hearing all sorts of interesting new terms coming recently. I heard someone use the phrase tokenomics recently, which is as we're using AI more and more, if you've used anything like ChatGPT or Claude, every time you type a prompt and you're getting a response back, that's sending what's called tokens. Your query is broken up into small pieces sent across the internet and the reply comes back in the same way. And it's how quickly you can generate those tokens and how cheaply you can generate them, because if they're expensive, it limits how you can use AI. So we're looking at being able to do that from Isambard itself. In fact, even one of the sovereign AI projects that's been funded was looking into... A company called Doubleword. Looking at how you can make that faster and cheaper running on Isambard, and they've already got some great results. So that's the sort of thing we're hoping to support in the near future.
Dr. Claire Thorne:
Fab, thank you. And before you mentioned this idea of AI co-scientists, so what does Isambard version 5.0 or beyond mean for the nature of doing science itself?
Professor Simon McIntosh-Smith:
It would mean that you'd be able to deploy those sorts of capabilities to far more people in the UK. So more of our researchers, more of our companies, so they'd all be able to really take advantage of this new capability as it really matures over the next couple of years.
Dr. Claire Thorne:
So essentially you have scientists working alongside AI co-scientists?
Professor Simon McIntosh-Smith:
That's right. Who the AI co-scientists would have at its disposal every paper that's ever been published, access to do experiments that you've never even thought of, ability to write code that you would never have the time to do yourself and run that code for you. So it's a remarkable step change in what's possible for scientists when they have access to this capability.
Dr. Claire Thorne:
Yeah, absolutely. Thank you. Emma, just picking up on that point around when they have access to this capability, who has access at the moment and how do we scale this so Isembard-AI is in the hands of every UK scientist.
Emma Hindley:
So this is what we very much like to do at the moment. So who has access first of all? Anyone who is wanting to do research. So we've got lots of academics, we have industry in partnership with universities, we've got small businesses, but we have regular calls for research on Isembard-AI and on the other AI research resource machines where people can apply to put forward their project and ask for a certain amount of computing time and power.
Dr. Claire Thorne:
Yep, GPU hours.
Emma Hindley:
GPU hours, exactly. So it's open to all different... I mean, we've heard the different use cases. Some of them are looking at fundamental research on AI itself. Some are looking at life sciences, some are looking at physical sciences. There's a whole range of use cases and we're very open to all of those. This is not about advancing specific areas, this is about advancing science.
Dr. Claire Thorne:
So it's free GPU hours. It's open to researchers, but also those in private sector in SMEs as well, is that right?
Emma Hindley:
Yes. Depending on the different call route that's being applied through. We have a separate channel for sovereign AI companies to come through where the government is working very closely with those companies to support. But if anyone who is listening is interested, then go onto the AI research resource page on gov.uk and there are application links there. So we're very excited to get more and more people on. And as I mentioned earlier, this is something that we are planning to expand in the future and create more and more access to scientific AI across the UK.
Dr. Claire Thorne:
Simon, how is Isambard-AI doing at the moment? Is it oversubscribed? Is there capacity?
Professor Simon McIntosh-Smith:
Yeah, it's actually full all the time now, which is great. That's a really good sign. And all of the calls are very heavily oversubscribed. So far more people apply, far more time is asked for than we have, and that's great. That shows there really is the demand there, but also that we need to keep working and make sure that we build out more capacity so that we're not artificially constraining the benefit of this technology to the whole country.
Dr. Claire Thorne:
Yeah, absolutely. So thank you both. Before we go, I've got some surprise quickfire questions for you. Simon, in this race, who wins?
Professor Simon McIntosh-Smith:
We do.
Dr. Claire Thorne:
Who is we?
Professor Simon McIntosh-Smith:
I'd say the UK. And actually we are doing really well nationally. We were lagging behind quite a lot, but with these recent investments, the UK is being genuinely competitive on the international field, which is very exciting.
Dr. Claire Thorne:
Super. I was going to ask about the international field, but also the domestic field because there are other supercomputers in the UK, but with a different flavour and with a different focus. Is that right?
Professor Simon McIntosh-Smith:
That's right. Usually focus in different areas. Might be general purpose, simulation for high-performance computing, but Isambard's actually one of the only supercomputers in the world optimised and run specifically for AI.
Dr. Claire Thorne:
Claim to fame. Simon, what's missing? What's the UK missing? What should it be doing now to invest for the future?
Professor Simon McIntosh-Smith:
Something we find very difficult is really long-term planning as in not just five years, but 10 years, 15 years, 20 years. And some of our international peers are very, very good at that. Japan does a very good job and they get quite a lot of benefit from being able to take that longer-term view. We've always found that quite difficult in the UK.
Dr. Claire Thorne:
Yeah, thank you. Emma, I wanted to come to you. Personally, what are you watching and keeping an eye on at the moment in this space, or perhaps who are you learning from or who should we be celebrating? Is that a tough one?
Emma Hindley:
Yes and no. So what am I watching in this space, at the moment thinking about what are the next systems that we're going to be building? What I'm really watching a lot of is what are the trends in the types of work that are being done on these systems, so trying to predict out in five years' time in a field that is developing so unbelievably quickly, how do we guess now what scientists are going to want in five years so that that system that we build is going to build... Four years. So that system that we're going to build is useful, the most useful thing that we can have at that moment.
And who am I learning from? People like Simon. So there are a lot of real experts who have been working in this field for 20-plus, 30-plus years, and what I've found is that in this kind of... There is a real community among scientists who are working in this kind of AI and supercomputing space and everyone has been incredibly generous and collaborative with their time and knowledge and not just me to each other, and I think that is something that's really lovely about this world.
Dr. Claire Thorne:
Simon, if Emma's learning from you, who are you learning from?
Professor Simon McIntosh-Smith:
There are so many people, and the AI space is full of fascinating people who we may not have come across before in any of our previous areas. I think AI enables so many more people who maybe they've not been able to write some code and run a simulation because it just wasn't their expertise, but now they can with the help of an AI. So meeting such a broad range of new people from different areas of science in the UK, which has been very exciting.
Dr. Claire Thorne:
And who would you want to celebrate in this space?
Professor Simon McIntosh-Smith:
We've actually got a brilliant technical leader in the Bristol Centre for Supercomputing called Sadaf Alam, and she is absolutely fantastic. She is one of the world's visionaries in how to design and build supercomputers. We were so lucky to get her in Bristol and she really is the brains behind what we've built with Isambard-AI.
Dr. Claire Thorne:
The godmother of Isambard-AI. I love it. Simon, what has Isambard-AI not yet done? What has it found difficult to do?
Professor Simon McIntosh-Smith:
We're still growing very much in the early stage of the software stack and there's certain things that already work really well and there are other bits we're still working on that aren't quite where we want them to be yet. It's partly the scale of the thing that we built was bigger than anyone had done this type of supercomputer before, but it's actually getting there quite far. So another year or so, I think we'll be very happy that we've got everything where we want it to be and we're going in the right direction at the moment.
Dr. Claire Thorne:
And to both of you, maybe not advanced AI, but how are you using AI in your everyday lives? I'll come to you first, Emma.
Emma Hindley:
Pretty basic ways, but I recently moved house and redid my kitchen and I asked AI to show me what it was going to look like if I put different handles on the cupboards or different countertops. So I'm terrible at imagining what something is going to look like, and that allowed me to actually choose something that I liked because it showed me visually what it was. I take meeting notes with it. I write scrappy meeting notes of what I think is important and it will turn it into something professional that I'm able to then share with actions where it has worked out the right dates. These are really quite simple applications, but incredibly time-saving.
Dr. Claire Thorne:
And what about you, Simon? How are you using AI at the moment?
Professor Simon McIntosh-Smith:
I actually used it for some data analysis just in the last week and I had a lot of historical data, I was using it to try and predict some trends and I was using one of the AIs to compose all that data into one big dataset, and even before I asked it, the AI said, "Oh, I've noticed there's a bit of an issue with the data in the middle here. I'm going to fix that for you." And it wrote some code and it ran the code and it fixed the data before I even asked it to do that before I even knew it was a problem. And at that point I thought, "Wow. This is really impressive." And that's today, what's it going to be like another year or two?
Dr. Claire Thorne:
And last one for you, Simon, what is Isambard-AI doing right now?
Professor Simon McIntosh-Smith:
It's doing hundreds and hundreds of different things right now. It's full up. It's busy beavering away. We've onboarded... Most recently, all the new things have tended to come from the new sovereign AI projects, which are very exciting. Lots of them are very interesting, innovative AI startups based in and around London, and they're really starting to power up now. So that's been the exciting new thing that's come on in the last few weeks.
Dr. Claire Thorne:
Simon, you've got some props that you brought in today to show us, sort of here's one I made earlier. What have you got?
Professor Simon McIntosh-Smith:
Yeah, I have. I think it's always nice when you can actually see something real. So I've brought two things in. The first one I'll show you both is this is an actual GPU from Isambard. So we always talk about AI supercomputers are based on these GPUs or that used to stand for graphics process units. These ones are actually made by Nvidia. And this is one of the Grace Hopper superchips that we've got inside Isambard. There's 5,280 of these all inside the main supercomputer and that's what the AI is actually running on. These are incredible things that have been manufactured. Some of the most complicated things that humankind has ever made.
Dr. Claire Thorne:
Super heavy.
Professor Simon McIntosh-Smith:
They're much heavier than they look because they're very, very dense.
Dr. Claire Thorne:
And that enables your footprint as a site to be quite small, does it?
Professor Simon McIntosh-Smith:
It does. In the past, a super computer like Isambard might've been the size of a football pitch, but now it's actually in a room that's about 12 by 12 metres, and that's it. It's incredibly compact for what they do.
Dr. Claire Thorne:
And how much does one of those chips cost?
Professor Simon McIntosh-Smith:
One of those, and this is sort of the gold rush behind AI now, but one of these would've cost about 25,000 pounds. So they're incredibly valuable things to get your hands on. Yeah.
Dr. Claire Thorne:
Thanks to DSIT.
Emma Hindley:
You're welcome.
Professor Simon McIntosh-Smith:
So that's the GPU. And then of course, one of the things that we've used a lot, you've probably seen all the little Lego mini figures that we have around. Lego crops up quite a lot-
Dr. Claire Thorne:
If you look closely, they're in the use case videos, aren't they?
Professor Simon McIntosh-Smith:
They do. They're little Easter eggs that you can sort of spot. But he tends to crop up in lots of different ways. So that's a different version of a slightly bigger Lego Isambard that he comes around with us and goes to various things as well. And of course it ties in with the name being BriCS for the Bristol Centre for Supercomputing as well-
Dr. Claire Thorne:
And also a lovely metaphor because the supercomputer is modular in construction, right? You've taken the principles of Lego.
Professor Simon McIntosh-Smith:
That's right. And I grew up with Lego as a kid and it always made me think about how to solve problems and build solutions to engineering challenges, and it sort of carries on that really big Lego, if you like, for building modular data centres today.
Dr. Claire Thorne:
The dream job for you.
Professor Simon McIntosh-Smith:
Yes.
Dr. Claire Thorne:
Super. Thank you both. So I feel like we've covered a lot. We've talked about the wider landscape, including sovereignty and sovereign capability, SME access, not just for researchers and academics, and also the nature of research itself and what this means for it. Thank you to you, Simon, the founder and director of the British Centre for Supercomputing at the University of Bristol, and our special guest, Emma, Deputy Director for Public Compute and AI Research Resource at DSIT. Thanks, Emma.
Emma Hindley:
Thank you for having me.
Dr. Claire Thorne:
If you're curious, do check out the use case videos and all the projects come to life on the YouTube channel. This is @BRICS-UOB. So that's B-R-I-C-S-U-O-B, and also on Insta, Bristol Centre for Supercomputing. If you're a researcher or an SME, apply for grants, essentially free GPU time to run your large scale projects through the regular calls that are managed by UKRI. If you've already accessed Isambard-AI to tackle some of humanity's biggest challenges, we'd love to hear from you. And if you're curious about Deep Science Ventures, you can check us out on our website and our socials. We're a UK venture studio and we build deep tech companies from scratch. AI is core to how we do that. It's core to our venture building process.