This Week in European Tech: Quantum's ChatGPT moment is coming

31:38Absolutely. First, make it work and then make it better and cheaper. Talking about better, maybe not so much cheaper, is the big news we've had coming out from Anthropic this week because they have a new model on the blocks. This week, they shipped Fable and it's a guard railed version of their flagship Mythos model that was released earlier this year. Now, to figure out how fable and mythos are connected and related, we can go a little bit into the roots of the words, because they are closely connected in the sense that fable is from Latin, fabula, a telling, a story, a tale, from fari, which means to speak, whereas mythos is a Greek word that also means word of speech or utterance.

32:24And again, it says something about something. Words, very LLM-centric. Now, in Greek, mythos would often be a sacred story. And over time, myth has sort of become the untamed or raw story, the one for the priesthood, the one that may not be available to the masses. Whereas the fable is the children's version, more tamed with a bit of a moral attached. So all us little people can tolerate and understand it. And that is exactly the approach Anthropic has taken here because they've shipped Mythos to the grownups and then they've shipped Fable to the rest of us. And now we have Fable this week. And what is Fable?

33:05Under the hood, it is essentially the same model as Mythos. But from everything we can see, it is the most powerful model you can buy. On Sweebench, which is a software engineering benchmark that's very popularly used to measure these models, it scores an 80%. versus GPT 5.5's 59, but it is heavily guardrailed. And what does that mean? Well, it means that the model has guardrails so that you can't use it for nefarious stuff. It does bring some side effects. You know, I had a test to do some simple queries about my own health, and I got a biological research warning. Make of that what you want.

33:44And when you do that, the model just shuts down and replaces itself with Opus 4.8, which is also a powerful but much less powerful model. And to some extent, you can ask why is that important and what does it mean? You could argue that it means we've hit an AI glass ceiling, where the models are now so powerful that we handicap them before we can let them into the wild. But just to keep myself honest here, Anthropic, they're saying that 95 % of the sessions they run, they never hit a gate, that the ceiling only binds at the edges. And 95 % sounds like a lot, but it also means that 5 % of all sessions are hitting guardrails.

34:21And if you think about how many sessions take place on these models every day, you can see it is a high number. So it'll be interesting to see how this develops. We'll talk about Fable and Mephas and Europe in context. But before we do that, I'd love to get some reactions for you and from you guys. And Andrew, you've said that the models are still just not reliable. Now, we've seen the strongest model ever here to be shipped is shipping in a handicapped version. What's your take on that? Well, it's interesting that Anthropic built a more powerful model, but they built one they don't fully trust.

34:55I'm not sure where the line lies between trusting the model and trusting the user and just trusting their legal department to not get them in trouble. But yeah, we're running into government's limits, not just technical limits. And the bottleneck may no longer be intelligence, although you've pointed out I don't like that word used for the transformers by LLM models, but confidence in releasing it. So, you know, sometimes now sort of every AI launch sounds less like a product announcement, more like a controlled relief. So, yeah, I'm not sure if AI is getting so much step change smarter now, but it's trying to close the gap between, you know, what companies have built and what they're willing to let the public use.

35:38And that appears to be widening. It is an interesting glass ceiling, isn't it? But Callum, I know that one thing is developing the model, something else is running them, all the inference, all the running costs. And inference is really where the cost lives at scale. I think you're in D-Matrix, which is a company that invests in chips for inference. Do you want to talk about what you think this means from the inference perspective and maybe the industry at large? Yeah, and maybe I would just add one thing. Fable got jailbroken, what, 24 hours after launch. so human creativity and ingenuity seems to still be able to out mark well at least the coders who tried to put guardrails on the model so can you really contain it is and why is everyone not having access to it anyway if it's if it can be done that quickly but yeah the cost is is quite phenomenal and this is maybe only the the early price to get us hooked and isn't the real street price that we'll see as the API or it goes to sort of only usage based.

36:46I think what we're seeing with a lot of these models is that they're still being deployed on the old architecture. Yes, it's the most sort of the forefront GPU models from NVIDIA, but it's not these new, should we say IPUs, so inference focused processes that are very much tailored to try and run these transformer based models on. So I think we are at the start of that and a lot of the cost comes down to well we can only run the massive right. You're only starting to see as these new processes like D matrix. So D matrix simply, they move the compute into the memory. So you don't have the memory wall that involves sort of shuttling the model parameters backwards and forwards.

37:38So they claim and this week, actually, they did a big announcement that they're now in full production for their new Corsair chip. So effectively, what it does is that you can, you can make the running of it far more efficient you can sort of either improve the latency or improve the throughput and you can do it at far lower energy levels so what we're seeing is these model these new chips move into production is other groups so like gimlet labs who are running a lot of new they're trying to orchestrate a lot of these different chips being able to package up different parts of the model processing onto the relevant chip that's required.

38:18So the D matrix chips would do the decoding phase of the inference side and the GPUs would do the initial part. So that's sort of where we're going to move to, but we still have to, rolling out chips is a long business. Even getting them into tape out, even post tape out. So once you've put your first chip into silicon, and if that all goes well, you have to find a random leakage current. You have to wonder why a thermal hotspot is happening in a certain area. So it just takes time to get these next generation, or I wouldn't say next generation, these new paradigms of chips out and useful. And that's maybe when we'll start to see the cost come down, but maybe we won't, because they'll allow us to be able to push the frontier of intelligence even further.

39:10You're seeing that Oriel Networks, they're looking at distributing different, so a UK-based company looking to distribute different parts of the experts and model of experts across different chips. And because they have an optical interconnect, you can sort of bring that together more seamlessly. So I think we're going to start to see innovations in the data center build out that will lead these models to being a lot more performant. but whether that lowers costs or improves intelligence, I'm not sure. I think a lot of the safety discussions sort of assumes these systems are becoming more agentic or self-directed.

39:47But the deeper issue may be that we're still fundamentally their next word prediction machines and they can maintain context within a conversation window, but they don't possess the context in the human sense, the persistent model of the world, causality, intent, truth. And every model seems to raise the question, are we seeing genuine reasoning? or is it just increasingly sophisticated passive matching? And I think the industry often talks as if this sort of intelligence is slowly emerging, but actually many of the failures still look exactly like what you'd expect from a system that doesn't actually understand what it's saying.

40:22You know, the counterpoint, I guess, from the anthropics and the opening eyes would be that understanding is a sort of philosophical distinction. And if a model can build increasingly accurate, you know, depiction or understanding of the world internally and like reason across them, And then perhaps understanding sort of emerges miraculously, you know, at scale. But I think that, you know, hallucinations and that sort of brittleness, I guess, and, you know, poor translation to genuinely novel situations suggest that, as you know, my view is that the architecture itself may have limits. So every new safety layer may be compensating for the same underlying problem that these systems are extraordinarily good, generating very plausible language that sounds intelligent.

41:09But I think we're still debating whether they actually know what they're talking about. A bit like us on this podcast, really. That is a really good question on both. I was struggling on the first one. The second one is absolutely well-established. Well, it's very clear we don't know what we're talking about on the podcast. So that's done. That's not a question to be answered. One of the things you raised, Andrew, was around safety and AI safety. And let's talk about what that means in the context of Europe. In the UK, we have the AI Safety Institute. They've been working with METHAS for a while.

41:42The European Union, they have ANISA, which is the EU agency for cybersecurity. And they were granted access to METHAS only on the 2nd of June. And it was Anthropic that sort of said, listen, you can have it too to play with. Now, the twist here is that last week we talked about CADA. CADA is this proposed new European set of rules that says that European government institutions and agencies shouldn't rely on digital platforms that are not sovereign for important work. And arguably, cybersecurity is important. In fact, by CADA's own definition, it sort of falls into at least level three, if not level three plus.

42:21Now, CADA would basically make the use of methods non-compliant because it's a U.S.-owned, U.S.-run model, and there's no security agency carve-out in the draft. New rules are not in place yet. They are not going to bind until 28, 29 at the earliest, so there's some time to fix the own goal. But this seems exactly like the type of issue that can be created by aggressive regulation. Now, Callum, what are your thoughts on how Europe can best strike the balance between sovereignty on one hand and access to the best tech where we need it on the other? Well, I think in this case, actually, just to your point, that the slowness of the bureaucratic engine and it not being implemented until 28, 29 means we'll probably have open source models of the same sort of caliber as Mythos and the same model size that can maybe be repurposed for European only.

43:20yeah i think it's a common theme that seems to be um seems to be the the view is that it's hamstringing europe but i guess maybe on the the positive side or the way you could think about it you are seeing a lot of ai backlash in the us uh that isn't necessarily being as reflected here maybe that's because the europeans don't necessarily buy into it as much or it's not sort of is shoved down their throats, but some form of showing that there is some control or some sort of people thinking about what could go wrong maybe in the longer term allows the proliferation of that technology a little bit better.

44:02So you think sort of an important part of managing the public perception and helping to drive it in a positive direction. Andrew, what are your thoughts on this? Is it regulatory overreach or is it the right calibrated the right way? I think it's great that there's an institution looking at it, but you can't win the race if you've got the brakes on. And it's as simple as that. And this idea that somehow we can impose these constraints, especially some of the European recommendations, and not have an impact on GDP or our defense capabilities is a bit naive. And unfortunately, the only way you get to dictate the rules is by being a leader and keeping up with the Joneses.

44:51And certainly at the edge, when you're talking about autonomous military systems, you really don't want to be left behind because you're hamstringing yourself. So I'm very, very cautious around some of the safety legislation. I think things are moving so quickly that it's great that we're looking at it. I think the legal frameworks will struggle to keep up they already are and most of all it will be very very concerning if we impose legislation which constrains businesses on on using the best models or even worse preventing us develop the models we need to to not be left behind in 20 years yeah i would say i also completely agree but yeah it's almost you wouldn't be getting this model on the 2nd of June if it had been developed here and the only way it could be developed here is if you were able to sort of go full speed and and not be encumbered at every point so I was kind of saying my point of view wasn't backing the need for safety regulation I'm just thinking what could be the positive of it and the fact that it's not coming into 2829 means it's probably going to be redundant by the time it comes in.

46:08A silver lining indeed. We're so slow, even at regulating, that it's harmless. Okay, Callum, very good. I like the optimism. Now, let's talk about some other homegrown things. Let's talk more about European tech sovereignty, because we have more news about sovereign AI models. We've been talking about a UK sovereign model for a while. And this week, Cosign, a London-based coding agent lab, they announced Lumen Sovereign, and they made a play to release Britain's first sovereign frontier model. Now, that might seem counterintuitive because to date, before the announcement, they had raised just$8 million, which doesn't sound a lot in comparison with the US Frontier Labs or even with Mistral.

46:52And it does look on closer inspection, like what's being built isn't a model from scratch, but rather an open-weighted model that is upcycled on UK soil with more post-training. Now, there's a partnership with the UK government. They're going to get half a million GPU hours from Isambard AI, and they're going to be spending a good chunk of that on AI coding and another good chunk on things like cyber and KYC and AML, plus some clinical trial stuff. The whole idea here is to develop a model that UK corporates and governments can use in air gap situations, on their own premises, inside their own walls, cheaper than the American models and sort of customized to local needs.

47:38They've got some MOUs out with BT, with BAE systems, with NatWest, Lloyds and HSBC. Now, understanding is that as of yet, it's not actual contracts. It's sort of design inputs. So nobody's really bought anything yet. But the idea here is, look, we're going to try and build something here that can be used locally, that has been developed locally, and that isn't solely dependent on the U.S. Now, do we think there is a need for this kind of sovereign AI model? Andrew, you've had views about that in the past. I mean, is this announcement here, is that important, a newsworthy, or is it even the right approach?

48:15Yeah, 100 % that we need sovereign models. Because I've said here and described the situation of AI as being on the journey towards nuclear weapons. And I think if you want to retain yourself as a first class world power, then it's a choice as to whether I have that nuclear capability or not. And you need to start today. So it's a step in the right direction. My biggest concern is whether this consortium approach will work. I don't know the complete ins and outs, I'll be honest. But from what I've read, it remains to be seen whether consortiums are, structures are consistently good at producing frontier innovation.

48:55You know, if you think about all the biggest innovations, you know, Google Search, ChatGT, GPT, AWS, none of those were built by a consortium. You know, frontier innovation tends to come from a small group with very aligned incentives, a huge risk appetite. and then you know once they get going the capital to back it up so i'm sort of cautiously optimistic i might have liked to see a slightly different boulder approach which is sort of to offer tenders tenders to large tenders to you know two or three companies who were trying to replicate the more traditional fully integrated vertical company in the uk like an open air or an anthropic because that is a structure we know works.

49:43It's incredibly hard as it is without sort of reinventing the structure in which that world-class innovation can happen. That doesn't mean it won't work. And it's great news overall because it's bringing to the forefront the importance of having something you control. And I've been the biggest sort of proponent of that happening because we don't want to be sending all our wisdom across the ocean to the US and having wisdom come back, which may or may not be biased, which we don't control and, you know, isn't nation state. Certainly a sovereignty angle there. I mean, Callum, what's your view? Have you already made bets in the Frontier Lab space?

50:24And do you see a need for a UK sovereign model? Well, I think you're seeing the need for, I wouldn't say necessarily a UK sovereign model, but a bargaining chip on the global stage. So obviously America has all the cards, Europe has ASML, which is a pretty powerful bargaining chip. I would say the move to fund quantum, the procure program, is the government trying to do a venture-like bet on an industry to become a bargaining chip within the global sort of geopolitical, the new geopolitical environment that we're in. Do we need a sovereign model? I think we just need models built here or sort of specific companies built here.

51:13Eleven Labs, could that be considered a sovereign model for a very specific application? It is a sort of UK champion in AI. DeepMind could have been a UK sovereign champion. We see others coming out of it, Ineffable Labs. I think the UK has enough AI talent. if you're trying to replicate a mistrower in the uk maybe this program to build a sovereign model is almost like a guaranteed procurement contract that we will help this startup develop something that we need to help seed a wider ecosystem i can see that being the play to your point we've backed multiverse computing in spain that the spanish government it's not a a frontier model creator it does model compression and gives those models sort of frontier like capability at lower power so they're trying to build a champion that's a bargaining chip that no one else can replicate so i would say let's find out what the uk can do really well and double down on that so that we become part of this global ecosystem instead of necessarily replicating it alaf alpha tried to do that in germany and got bought by cohere because it wasn't necessarily they didn't need another sovereign capability there yeah i mean like sort of counterpoint to that i i worry a little bit isn't that sort of isn't it isn't it giving up a little bit like isn't it sort of just kind of accepting that we've lost the race um and we should kind of well we'll we'll do something so that we're sort of a thorn in the side enough that maybe we get some leverage but it doesn't really solve the fundamental problem which is ai is our future and if you don't start now it'll get to the point where actually catching up becomes essentially practically impossible i mean the reason why i think the nuclear um analogy is so interesting is because it's pretty easy in quotes to sort of build some sort of nuclear device but the infrastructure around that for it to be useful actually particularly to deliver it reliably at distance whenever you want at the push of a button is extremely complex extremely expensive and requires you know infinite infinite amounts of vertical integration and delivery systems and nuclear submarines you know etc etc etc and i think that's analogous to um ai that as the technology continues to evolve transformer models aren't probably going away but they will be supplemented or new technology will come along and i think unless you set your sights on on keeping up and being autonomous and having sovereign capability inevitably you just slip automatically to being a begging bowl and accepting whatever you can get with or without that bargaining chip from the US or whoever.

54:12And I think for me, when you look at where this technology is going, that's a big worry. Be ambitious, go bold, be the master of your own destiny. Andrew, we can always count on you for some solid views there. Now we've had some news on quantum and on AI models. We've also had news on AI hardware. This week was London's tech week. We've seen all sorts of announcements come out there. We've seen a government package. We've seen some companies timing their announcements. to this, and specifically the government, the UK government, has announced a£1.1 billion hardware plan. Why 1.1? Maybe I think when you squint and round it, it rounds up to$1.5 billion.

54:54I don't know whether that was the intent. Let's unpack the plan a little bit. There is a push to build a big national supercomputer in Edinburgh. £750 million to decide for that. Should land in 2030. That's around the same time as Callum's new quantum computer. And we'll see if it'll be relevant by then. Inside the package, there's 400 million pounds for chips and 150 of that will buy British inference chips this summer. So one of the things we've been talking about here in the program is that the government should, rather than just handing out grants, should be a customer and actually buy stuff.

55:30Exciting to see that it's the intent here now for the government to buy chips now from British suppliers. Now, the government, as part of the package, also announced that in partnership with the British Business Bank, it will make the largest fund investment it's ever made to Playground Global. A Playground Global is not a typical British venture capital fund. In fact, it's not really a British venture capital fund at all, because it's a fund from Silicon Valley. Now, Callum, I'd be interested in your thoughts on this because you're building and have been building a great investment firm here with Bullhound Capital.

56:07What are your thoughts on the British government effectively partnering with Silicon Valley investor, should we call it talent, to get them in? Is it a helpful addition to the industry where we are short of growth capital or do you feel it's displacing your activities? give us your view i would say that if the bbb were to pick one fund to do this playground global would be one of those sort of the best ones that i could have picked um we their earlier investors in dmatrix um their venture partner sasha is ex sort of um vp of software nvidia and he was instrumental with dematrix in getting their software to the level that it's needed to actually be rolled out onto sort of um into full production so the the one thing that i always hear from a lot of u.s companies or companies redomiciling to the u.s is they can get two times the engineer in the uk for half the price so the talent here is amazing but there's just there is not the quality of engineering leadership that there necessarily is in Silicon Valley.

57:23So I can see that this playground global sort of fully focusing on the UK, the expertise that they can bring will be instrumental for the sort of companies that the BBB wants to back and is becoming a lot more, shall we say, sort of commercially focused on and going after. so I can really see this as being complementary to the overall UK ecosystem if we think about the this is probably one of the best moves for the UK founders they get access to this very experienced group that has the the ability for the Olexes for the fractals to be able to go through their whole sort of tape out and actually building the software stack on top.

58:10So I see this as a very positive move for sort of enticing the Americans to really care about the UK industry, especially with such a strong, great display. So it sounds like it's sort of a relatively positive endorsement and a view that it's a strong partner to bring in. Andrew, what are your thoughts on this? I mean, you've sort of said we should go bold, we should build bigger. is getting and importing some American swashboggling talent. Is that going to be a driver here? First of all, I think the BBB have been instrumental in accelerating the VC sector in the UK. And it's great that they've been given more money and they need to continue doing what they're doing.

58:56Playground Global, completely agree with Callum. Fantastic fund. They're actually investors in universal quantum. We've got some co-investments with them. But I'm immensely frustrated. I'm not going to lie that this money's gone to a non-UK fund. There are a whole bunch of extremely smart fund managers in the UK who could have run this as a UK fund and who do have the ambition. And I just think that it was unnecessary to ship in mummy and daddy from the US as if we don't have people who know what they're doing on our little island already. So yeah, I'm a bit frustrated by it. I'm not going to lie.

59:33And it's similar to what happened with ARIA and 50 Years, which is also an American fund that they partnered. Again, they do a fantastic job, 50 Years, incredibly smart, successful fund. It's nothing about them that they are not great fund managers and not great people and incredibly smart. It's more about there are also smart people and we need to back ourselves and back our own funds and back our own VC ecosystem because that money will be invested into UK companies. The pull will be even greater to flip US if it isn't already just because of need, because of capital and because of market. But the carrying the profits will go back into the US ecosystem or be taxed by the IRS and not the UK.

1:00:20So short answer, I would have liked that to go to a UK fund manager. You know, it's 150 million pound commitment. I'd be interested in your thoughts on this because the Reeves and the government, they get a lot of sort of complaints for not being enough sufficiently business focused. They did make a big additional capital injection into the BBB, I think on the order of several billion pounds. And if you think about it, if this is 150 million out of, let's say, an additional pool of 5 billion and the remainder, let's say 4.567, all go to homegrown managers. is it sensible to get a bit of u.s talent into the mix or would you simply be completely no it all has to be uk and uk only i think the talent comes across anyway right and it's not an altruistic sort of uh choice by a usbc to invest in a uk company it's a purely commercial one and playground global as i said invested very early into universal quantum so when there are great companies, they will invest.

1:01:29And I don't think it needs, you know, essentially the UK sovereign wealth fund equivalent to be backing those managers. That said, I don't know whether the fund is domiciled in the UK and will pay UK tax and the UK tax on the pyramid. Maybe it is. That's great if it is. So look, I'm not saying this is a negative for the ecosystem. It's probably be a net positive, but it just, when I see, you know, I run Dragon Chases, as you know, which is 100 and nearly 200 GPs and LPs representing about 50 billion under management. It's primarily not for profit. And there are a whole bunch of GPs in that organization who would do an amazing job of deploying that capital, you know, in the UK.

1:02:12So I just think we need to be careful about the signaling and the uk and indeed europe's biggest problem that we talk about on this podcast a lot is often our self-confidence it's about backing ourselves and it's about backing our own home grown stuff and people and companies and i think that is what we should be doing and we'll make a few mistakes along the way but you know we have a long heritage of inventing the world and we've got plenty of incredibly smart people like Callum and others who could easily deploy that 150 million. I think that's all fair. I mean, I would say it seems to me that we're almost suggesting that we're not backing ourselves.

1:02:55But if I've got a pool of capital and I'm putting 95 % of it into UK funds and 5 % to a US manager, we say that that's not backing ourselves. It seems to me that it still is. but it seems you're sort of saying it should be 100 % all the way. I was going to say I kind of agree with you Mads in the sense that we can't necessarily make it a judgment on its own. It's sort of look at two years time what has been backed. The BBB's mandate is to support the venture ecosystem and the companies and I think bringing in talent like this to be very focused on the UK is a net benefit But I appreciate Andrew's point that we need to build the sort of confidence locally, but maybe let's just see how the rest of the year or the deployment plays out to how that confidence is built in the UK.

1:03:50I don't doubt they'll do a good job and I'm sure it'll be a great investment for the BBB. But I think there are other managers I know would very much like that money. Good. Well, lots of different views on that. I'm just happy to see more money going into the ecosystem because we need it. And even more preferably, hopefully we can find ways to unlock the pension capital. I think that would be good for both pension savers and also for the economy at large. Now, let's move on because we've had a big Apple week. We're not necessarily going to spend a lot of time on it, but Dan is away this week. And this is his big moment, the moment he's been waiting for, which is that Siri has finally been announced the new Siri with AI, or at least released-ish.

1:04:34It was shown at the Apple Worldwide Developer Conference earlier this week, and it is a ground-up, rebuild, custom Gemini that sits in the back end. So more than a trillion parameters in the most powerful instance. It looks like Apple got a pretty good deal. They're going to pay Google about a billion a year, which was much less than what was quoted from some of the competitors. But it's not just a plain vanilla Gemini. It's a model that will run inside a private Apple cloud. Google will not see the data and everything is being customized and integrated into the Apple ecosystem. So hopefully something to look forward to there for Dan and every other Apple aficionado over the year.

1:05:19Now, we are getting up towards the end of the program here. Just a couple of deals of the week to mention. Andrew, do you want to talk about ISI? It looked like that was something in your wheelhouse. Yeah, I mean, fantastic. Valuation, Series F at 10 billion plus valuation. General Atlantic lead. So world largest commercial satellite operator. Highly profitable, well, in terms of at least a quarter of a billion in revenue, 100 million EBITDA, very, very long contract backlog, I think of over a billion. And yeah, rapidly scaling satellites, 50 to 100 satellites a year by 2028. So another example of how Europe is really punching its weight, I think, if not in space launch, which is a big bugbear of mine, access to space, But certainly in satellite build and capabilities, we're starting to catch up.

1:06:21And I think it's all good. It's good to see for sure. Another deal of the week, Nuer Robotics out of Germany. $1.4 billion raised at a$7 billion valuation. Closed on Wednesday with investors like Tether, Qualcomm, Amazon, NVIDIA, Bosch, Scheffler, and the European Investment Bank in the mix. This is really a humanoid robotics play. They're talking about 6 ,000 units this year already with an order book above 1 billion. They've developed a sophisticated approach to training the models and doing skills transfer to get the robots quickly to a place where they can be useful in production. And David Reger, who is the CEO, he said he really sees this as the last chance for Europe to actually ever produce anything meaningful.

1:07:09again which is a big word to come from a german somebody in the european industrial hardland now callum i don't know if we've already had your deal of the week or whether you have another one in the in the wings i don't necessarily have a deal of the week beyond what's happening today i think it's the week of the ipos right we we almost we've had a quantum one we've had a we've had a we'll have a space one today the only one we're missing is a fusion one to almost have the trifecta of future technologies that are coming to the public market. So I just would say that this is the week that it seems a lot of things are changing.

1:07:49Yeah, that was very exciting. The future is coming home. And I would just add on the neuro robotics and people almost overlook that Europe is the home of robotic deployment. So I can see that being a very overlooked area that Europe could be quite significant at. Everyone deploys ABB, Bosch around the world for precision manufacturing. So people make a lot of noise about China building these cool robots and America pushing the forefront of intelligence. But Europe is pretty good at deploying these things. No, 100%. I mean, sadly, you could say we've sold KUKA robotics and ABB robotics were sold to SoftBank.

1:08:32So we've been selling some of the crown jewels, but hopefully they're a new kind of the next generation of European robotics pioneers can come through. And yeah, it's good to see. In terms of what's up in the week ahead. So next week, SpaceX will start trading. It's not going to be in any indices yet. The S &P blocked it. They're saying it's going to take longer before they'll let SpaceX on board. NASDAQ 100 SpaceX should enter in July. So next week will be active money only. It'll be interesting to see how it does. Monday, we've got Bending Spoons. They're going to open their book for the listing.

1:09:10Tuesday, Strasbourg is going to rubber stamp the AI Act loosening. We spoke about it a couple of episodes ago. And at the time, we said a deal would likely come in May. It did about six days after our episode was taped. Now it's finally getting rubber-stamped. Now, in the episode 89, we said that the parliament would initially block the loosening of the AI Act. The whole idea here is to try and water down some of the provisions to not hamstring European enterprise too much in the adoption of AI. It wasn't parliament that blocked it. It was actually the council, so the group of sort of governments.

1:09:50So mea culpa, we got that one wrong. But it was blocked at the time. It has now been unblocked. It was Meritz from Germany who unblocked it, so good on him. And what actually changes is that a lot of the rules that were initially part of the AI Act are going to be pushed back. Some of the high-risk compliance has been pushed from this August to December 27. Embedded AI all the way back to 28. and machines have been excluded completely. The only thing that is still in is MedTech. MedTech was left behind, which I do think is a bit of an issue for the European economy. But let's see how that plays out.

1:10:28Now, in terms of the scope, initially, it's just the AI Act, the GDPR and the cookie omnibus is still stuck in committee. Hopefully they can get that sorted as well. So that's going to be middle of next week. Thursday, we're going to see Accenture earnings, which is going to be the cleanest read on enterprise AI service demand. Always a good bellwether to watch. That is it for the week ahead. Callum, Andrew, it's been a pleasure and a gift having you here on the program today. Hope you've enjoyed us as much as I have. To the listeners, thank you very much for tuning in and see you on the next one.

1:11:03Thanks, Matt. Thanks, man.

1:11:13We'll see you next time!