James Wise on why the Sovereign AI Unit criticism is wrong – and the one company he wishes he’d backed

0:01Hello. Astute listeners will be able to tell that I'm not Amy Lewin. Instead, it's your new host, John Thornhill. I'll be taking over these interview podcasts while Amy's on maternity leave. And as you'd expect from Amy, she delivered to deadline and is the proud mother of a very cute boy called Wilbur. If you don't know me, I'm the co-founder of Sifted, the media organisation which covers European tech and makes this podcast. I also write a weekly column for both Sifted and the Financial Times, where I'm innovation editor. So on with the pod. Joining me today is James Wise, partner at Storrid VC Balderton and chair of the Sovereign AI Unit, a new£500 million initiative launched by the UK government.

0:45As its name suggests, the Sovereign AI's aim is to anchor promising AI companies to this country. The unit was announced last year as part of the AI Opportunities Action Plan, authored by Entrepreneur First's Matt Clifford in his role as AI advisor to the Prime Minister. It has recently backed former deep-mind scientist David Silver's AI lab Ineffable Intelligence in a record-breaking$1.1 billion seed round, which valued the startup at $5.1 billion. And James started his career in venture 15 years ago when he helped launch one of the UK's largest social venture funds, the Social Business Trust.

1:22Since joining Balderton in 2012, he has led investments in second-hand shopping platform Depop, AI agent maker Convergence, bank payment fintech GoCardless, and AI Frontier Labs, Priat Labs, which was acquired by German software Titan SAP at the beginning of this month. James, great to have you on the pod. Thank you for inviting me. Well, let's start with the news. I'd love to hear about Pryor. What does Pryor do and what do you think of this deal? Pryor Labs build tabular foundation models and many listeners I'm sure are aware of LLMs, large language models, which are great at generating responses to questions based in language.

2:03Tabular foundation models take a similar technique but apply it to big data sets. So think about spreadsheets, databases, structured data. If you've used a large language model to ask it a question, to do basic maths on some of these data sets, it's not particularly good. They're not really specialized for that kind of use case. So tabular foundation models, and Pryolabs build the best ones in the world, allow you to do things like statistical analysis, regression, predictions on this structured data, but in a one-shot way, in a very quick way, and in a way which means that an individual like myself who doesn't have huge background in machine learning can do in seconds what used to take weeks of work for a trained data scientist.

2:42And this is a very quick deal from formation of the company to acquisition. I mean, it only raised a 9 million euro funding around in February 2025, I believe. Why did it all happen so fast? Yeah, well, the company was formed in February. We were the first and only investors from the venture side. So I was the only board member in the history of the company. but it was, you know, we invested, and the reason why it's so successful is because it was based on the work of Professor Frank Hutter, who'd been working in this field for decades. So he invented a lot of the methods used traditionally in what's called AutoML.

3:17He was well-known globally. So obviously, because of his expertise in the space and because of the team he built, they were very quickly able to ship the best models in the world. And so they led on all the benchmarks, and because of that, were also very quickly able to build a great community of developers around them and get a lot of commercial interest. So 18 months is very quick turnaround from seed round to, you know, billion euro M &A. But, you know, they've executed very fast and they did 10 years worth of work in that time. Can you share any more about the terms and how much the exit was?

3:52So we can't share a lot right now. This is obviously going through a regulatory process. So we expect that to complete in a couple of months. But as SAP made public, this is over a billion euros they'll be investing in this company, and a vast majority of that is going to the acquisition. Right. Okay. I believe that you met the co-founders before the company even existed. Is that common? It's increasingly common. We're moving to a world where the people with deep expertise in AI, whether that's, and we have the same experience at Convergence, by the way, and there's this experience at prior labs, but they're incredibly rare, people at the frontier, but incredibly important in pushing this technology forward.

4:30A lot of them come from research. A lot of them have been either in academia, in their own institutions, or in some of the other frontier labs, whether that's DeepMind, which is famous for its research, OpenAI, or Anthropic. Because they're so steeped in research, they can build these products very quickly. And so they're able to raise money literally before they start a company, because they know what they're going to do. We're moving from a world where in lots of areas, the startup ecosystem was around fast iteration and learning. And so very quickly still, but over time, you built a product, you learn what worked and you scaled it to a world where you can ship a product very quickly and scale it much quicker.

5:10And so increasingly, we are seeing these deals, which are basically formation investments, investments in the company at formation. And a lot of these companies are selling out really quite quickly. We've had a whole string of very promising AI companies being sold. We've had Silo AI in Finland, SANA Labs in Sweden, faculty in the UK, and our prior labs in Germany. And when Eric Schmidt spoke at the Sifted Summit last year, he urged European founders to be more ambitious, not to sell out for a billion, but aim for a 10 billion, 100 billion. Is that not possible in Europe? Well, I think it certainly is.

5:42And, you know, you listed some fantastic companies there. But I'd counter with Wave, which is raising over a billion dollars this year, has raised a billion dollars this year. ineffable intelligence that you just mentioned raised a billion dollars this year. 11 labs, I think, will raise roughly that amount this year. So I think there's plenty of examples of frontier labs in Europe who are scaling rapidly. You know, the difficult decision for founders in AI is to compete. Often, if you're building a model, you need access to a huge amount of compute and a huge amount of distribution. And some founders will say, well, for me to complete this mission, for me to transform the industry with my particular type of AI, in PryLab's case, tabular data, tabular foundation models, it makes sense to partner early with a company of the scale of SAP.

6:25And certainly at Balderton, we saw that with Convergence, which joined with Salesforce, and Pryolabs, which will join with SAP. But it's not the only experience of scaling AI in Europe. Let's move on to the sovereign AI unit. Why did you get involved? Well, first of all, I think, as many listeners will share, you know, I feel passionately that this moment, this technological moment we're in, in AI is going to be transformational. It's going to be transformational. It's obviously transformational to the work we do at Boulderton. But far more importantly, it's going to be transformational for the economic and national well-being of us all across the world, but specifically those of us in Britain.

7:03And so being able to be part of that in any way where I could use the skills I have, which is setting up venture funds and doing early stage investments, I was keen to participate in. On top of that, this government, building on the work of the prior government, have made it clear that AI is a priority. So lots of people go into government work for different reasons, but this really felt like an area where the government wanted to be successful. They really cared about it as well. Certainly the minister I've been working with closely for AI and also the Secretary of State and Liz Kendall feel that way.

7:34And so I saw this opportunity to both use my skills to be helpful, but also in a way where it felt like the government was really behind it. And how did it work? who approached you from the government? So I have done some government advisory work with previous governments, and I specifically was on something called the Industrial Development Advisory Board, which is a rather grand title, but to oversee large amounts of taxpayer funding into different industries. Often it's to grill the people suggesting that and pull back some of the taxpayer funding. And to be honest with you, I was complaining about some of the government processes in how we dealt with the private industry when we came to funding.

8:13And the Minister Freyai, Kanishka, was kind enough to listen to my complaints and say, well, why don't you do something about it? I will teach you for complaining. And now on paper, you only work five hours a week, I believe, as chair of the unit. Is that really the case? And how do you spend that time? Well, you know, I don't think we can measure the hours, but it's a lot more than five. You know, my role over the last four or five months since I've been chair has been really to set it up. and that means establishing a legal entity. It means establishing the processes. It means hiring a team.

8:41And it means setting a lot of precedent because what we're doing is pretty unique within central government at least. And of course, on top of all of that, reaching out to founders and sort of helping with that part of the job. That's taken up a lot of time over the last four or five months. The good news is we now have a great team in place. We now have our processes set up. And so I can step back and be one of the more traditional chair types who turns up a couple of hours a month and provides some hand-wavy direction. I guess one of the questions about the fund would be its relatively small size.

9:12Do you think that you can move the dial with£500 million? I was in Silicon Valley recently, and just OpenAI and Anthropic between them have raised$150 billion this year. How ambitious can you be? Yeah, and look, in the UK alone, we've had over£8 billion, I think, £6 billion and change, invested in the last quarter, right? So we recognize that£500 million is an entry ticket to this market. It's not nothing. And I have to reiterate that. You know, I spent a lot of time with founders up and down the country talking about sovereign AI and a£5 million check is transformative. So I don't want to be too cute about it.

9:49We recognize it's an entry point. It's an entry point both to other parts of government, which can provide more capital, British Bank, National Wealth Fund and others. But more importantly, it's an entry point to many areas of support the government can provide, which go well beyond cash. And if you are one of those breakout labs, as we heard about the ineffable intelligence round, you mentioned already, you may not worry about raising capital, but you do worry about access to compute. You do worry about making sure that you are accessing data, which is relevant to the kind of models you need.

10:19You do want support around building talent, both here and internationally. You may want support selling to government, right, through traditional procurement processes. All those things are worth a lot more than the 500 million we'll be investing. And we hope this is just the starting point of what we're able to do. You mentioned ineffable. Tell us about it. Why did you back David Silva? Well, David Silva is one of the most important researchers and practitioners of AI in the last decade. He's a professor at UCL, but he worked at DeepMind very closely with Sir Demis on their multi-agent work on AlphaGo, which is one of their big breakthroughs, and to a lesson on AlphaFold.

10:54his approach though to ai is fundamentally different to what we've seen work so far with open ai and ropic and others um so rather than building large language models on huge amounts of human generated data uh they're taking a different approach in apple um they're basically training agents through experience um and if that's right if that works not only will it mean that the AI models that we all use get better, which is great. And we are able to hopefully access even cheaper models. But also, it means that we're going to have a very, very competitive frontier lab here in the UK. Now, it may not work.

11:35The risks we're taking at Sovereign AI are the same risks an early stage venture capitalist takes. We expect failure in lots of the cases. But if it does work, the upside here for Britain and for all of us is so immense. And David's Silver is so unique in his capabilities, we felt it was one of the right bets to make, as it were. It's quite striking in Europe how labs are making different types of bets. I mean, Silicon Valley is very much an LLM pill, isn't it? It's all in on scaling these large language models. Europe, there are several companies now, AMI Labs run by Jan LeCun in France, ineffable here.

12:11We've had recursive setting up in Europe as well, and unlikely as well. Is that a real opportunity for Europe, do you think, to explore these different avenues to getting to AGI? Yeah, I would list a lot of other unique types of model architecture and unique types of models in different areas. It is an opportunity. You know, I think that the large language model companies, and obviously you've got to include DeepMind in that conversation, are building phenomenal products, right? And they're all transforming the way that we do our work. However, they are now in a race. They're in a race for capabilities.

12:45We know what the scaling curve looks like around capabilities. It means huge amounts of data, huge amounts of compute, a lot of energy. And they're so focused on winning that race. I think it means they can often overlook some of the other areas of innovation that are happening in AI. and whether that's something a breakthrough in life sciences, whether it's a breakthrough in robotics or it's a fundamentally different architecture, there is now opportunity to compete in that space. And Europe has, it's more than its fair share of brilliant researchers and is increasingly able to access capital to take advantage of some of those opportunities.

13:19Okay, tell us about the second investment you made on Calisum. So Calisum is a team which Cambridge students went through ARIA, went through Entrepreneur First, so have got great experience with the frontier early stage investors in the UK and beyond. And they're building a software platform to allow you to access heterogeneous compute. So that means that you don't have to be completely reliant on one compute stack. Let's say it's just TPU at Google or it's just, you know, NVIDIA's GPU in other clouds. It allows you to balance different usage across chips. If we're in a world where actually it gets quite difficult to access compute, a compute crunch, which a lot of people think we're already in, it might get worse, then you do need a solution where you can access different types of chips at different types of speeds, different types of times, and color-similar enabler.

14:07And you mentioned there ARIA. You've also kind of co-invested with the British Business Bank as well. Is that kind of really the kind of nexus of new capability in Britain, those kind of three organizations that you just talk? Well, certainly. And look, we do different things, right? ARIA is really focused on grant making, on really early stage research. Sovereign AI is about accelerating early stage successes. So we're not here to do market failure. We're here to focus on successes. But we're doing that in a commercial way. So we're on the cap table. We're hoping to get taxpayer returns. And you're right, British Business Bank, they have a much broader mandate.

14:41They're going to do much broader investing than just AI and have done historically. But they can scale in terms of the capital they deploy to hundreds of millions of pounds in single checks. So we work very closely with those two organizations. Now, the Units Venture team is led by Josephine Kant, an investor with experience at Google and YC. And she told Sifted not so long ago that when she consulted people about taking the job, they told her that she would be absolutely crazy to take the position. Have you been surprised by the caliber of the people attracted to the government and prepared to make these crazy decisions?

15:16Yeah, and Josephine is a great example of that. She's someone who has spent her career at Y Combinator, at Google, setting up a seed stage venture fund in the US and gave up that highly lucrative, potentially, career to go and work in government, where obviously she's taken a salary cut and she has taken on all of the wonderful restrictions that you get working in the public sector. So, and we've advertised for investment partners, for a managing partner, and for lots of different roles within this AI fund. And it's astonishing the quality of people who are willing to give up their time and effort to come and be part of this.

15:55And one of the reasons why that's been possible is because the government have allowed us to set up a fully independent investment committee inside of government. It's a really hard thing for them to do. It's a very hard thing for us to have achieved, to whom else have you. There's been a lot of negotiation about how that works. But because we've been able to give people that freedom, I think people feel it is something which is going to make a difference. It's not just a government announcement, a splash of the front pages that never goes anywhere. So hopefully we've got that right. Hopefully we'll give people the right independence.

16:25And as a result, we've attracted some brilliant things. And how do you address the potential conflict of interest issue between what you're doing at the Sovereign AI unit at Baldurton? Yeah, and obviously this is a huge part of my onboarding and all of my investments that I've done both at Baldurton but also personally are all public. But I'm not involved in any particular investment decision within the Sovereign AI unit. So I don't sit on the investment committee. I don't involve myself in any of those decisions. I'm there to set the strategy, set the direction to hire the people. And therefore, there isn't a direct conflict of interest.

16:58But I recognize that there could be a perceived one. So we've tried to be as transparent about everything as possible. Okay. And now there have been several entrepreneurs and investors who have gone into the government. Kanishka Narayan, former VC, is now the UK's AI minister. We had Matt Clifford from Entrepreneurs First being the PM's AI advisor for a while. Alex DePledge, founder of Ressi, who became the chancellor's entrepreneurship advisor. How valuable is it to inject this kind of private sector experience into government? Well, look, it's been really welcomed. I think that's a great sign.

17:30And, you know, as you mentioned, it's been across governments. The Conservative government previously brought in Matt and he was kept on. They also brought in Ian Hogarth, who helped set up AC. And you mentioned Alex, who's joined more recently to work with the Chancellor. and you know in my few months of doing this job now uh the experience is really appreciated you know i think that the civil service get a lot of criticism uh that's part of the role right transparency and accountability from the press and from the public is necessary um but i think there's a recognition that certainly when it comes to frontier technologies and venture which is a relatively new asset class uh when it comes to the grand scale of things and a relatively small one still but a growing one uh the government doesn't necessarily have that talent and and capability internally and so they're building it but they're building it alongside people they perceive to be experts in the space so far i think it's worked fairly well i think ac has been a great success i think alex's work on uh reforming some of the regulations around what we're building here in the uk what can be built here in the uk has been has been really positive and i think the ai action plan had a lot of really positive points as well um from matt so overall there's some good momentum But ultimately, the government can't just be reliant on bringing in external talent.

18:46It needs to build these capabilities internally as well. And I hope at Sovereign AI, we contribute to some of that. Does it surprise you that this government is not shouting more loudly about the success of tech? I mean, as you were talking earlier, some astonishing companies being formed in Britain and Europe more generally. There's a lot of money now going into the sector. It's a great news story for Britain. Why is the government not shouting this from the rafters? Well, I think anyone who follows me on LinkedIn will feel that I'm certainly shouting it from the rafters and that, to the Secretary of State's credit, she gave a great speech about AI at RUCI last week.

19:22She was integral to us getting coverage of our launch. And so I think lots of people in the government are talking positively about AI. The Chancellor was very supportive of our launch as well. The Prime Minister has talked about AI a number of times, has a lot about it. however there's lots of challenges facing the UK at the moment and I think sometimes it's hard for the government to get a hearing on some of the positive stuff I think it's a great shame because we are in a moment where Britain should be proud and excited about what it's got when it comes to AI I don't think we've done a good enough job of explaining it to the public and I worry that political opportunists will see that and as we're seeing in the US to some degree now there will be a movement to say, well, what does this do for Britain?

20:08Let's talk it down. Let's try and stop it. And there may be votes in that for some people. And so you're right to say, well, I actually think the government's trying. We need to do a much better job of getting this positive message across. Could you ever see yourself quitting the venture industry and going into politics full time? I'm doing enough work at the moment in government to know that I'm probably a focus on the day job for now. Now, one of the functions of the unit is not only writing checks of up to kind of 10 million pounds or so for AI startups, but perhaps more importantly, offering portfolio companies access to the government.

20:43What does that actually mean? Well, it's a very broad offer, but to try and make that feel more concrete. So, you know, very specifically, we're putting out calls around procurement. So we put out a call around strategic development of automated labs. We've put out a call around strategic development and new data sets. OpenBind, which is a data set around protein and enzyme binding, is being developed, and that's public. And we're putting out more calls like that to say, okay, the UK government has access to compute through places like Isambard, which is a known supercomputer out in Bristol. It has access to huge data sets, which are really valuable to making AI better, which the civil service want to use.

21:26They want to use this AI, so they need to train it on the right data sets. And it also has a lot of institutional knowledge and it buys a lot of stuff, right? Hundreds of billions of pounds go through the procurement processes every year in the UK. And so what the Sovereign AI is doing is working with entrepreneurs to help navigate those opportunities. You know, I think historically, in particular around procurement, multiple different types of government have failed to reform the processes such that small businesses, insurgent technologies can be adopted in everything from defence through to your local councils.

21:58And so part of our job at Sov AI is to help entrepreneurs navigate that. And do you think the government can be doing a lot more of procurement? I saw a chart the other day that the top nine government, sorry, nine of the top 10 government AI contracts have been awarded to American companies. Yeah. And look, that probably reflects the fact that we need more AI companies here in the UK to compete with them and build better products. And that's what our focus has been you know i think um a mistake would be and obviously there is some political discussion of this at the moment but a mistake would be to say the uk should have worse software and technology because we only want to buy things which are built here on british shores what i want to be the case is that british technology companies can compete and out compete a lot of those us or it's not just about us companies about international companies and so Brits get access to the best technology and they happen to be British.

22:50That's the approach we're taking at Sovereign AI. Often we will make sure, and obviously our focus is on British sovereign companies, but often I think we do overly focus on large companies rather than smaller ones when it comes to assessing whether or not a company can deliver for government. It isn't the case because you're a small number of people that you can't build competitive products. I invest in startups. The whole point of doing that is to say small, determined teams can build better products. So there can be a shift in procurement where we say we're more open-minded about smaller companies.

23:25And I think that would help British companies compete because a lot of them can't operate yet at the scale that American competitors do. But I'm certainly not against working internationally with the best companies. I just want to make sure more of them are British. There's been a lot of criticism of Palantir in particular, who bring up contracts with the NHS and the Ministry of Defence and other government departments as well. Is there really no British AI company that can do what Palantir does? Well, if you look at the recent history of NHS software contracting and the absolute debacle that we lived through in the 2000s on this, unfortunately, a US company did have to come in and provide a much better service.

24:06So whether it's specifically within the healthcare system and the NHS or education system or other areas of government, I think there should be global competition to serve the British people. I want the best companies in the world to do that. Very clearly, Palantir has some products which are the best in the world at what they do. And the way to change that is to invest in British companies here and work very closely with them to build better software. it would be incredibly short-sighted to say Brits shouldn't have access to the best software because we don't like some of the shareholders on the cap table of these companies.

24:39And even more short-sighted to then say, right, we're going to ban foreign technology from Britain because the knock-on effects would be horrendous, both for Brits who use these services, but also I think for our AI companies here, because almost certainly they would get banned elsewhere. so the right approach to Palantir being very good is to build companies in Britain which are better okay to that end how do promising British AI companies get you on their cap table well at Sovereign AI we are very public about our criteria they're all on our website about how to apply also on our website and you know we've got a great investment team they can reach out to directly.