64. The Hidden Power Behind the AI Boom

0:00Welcome to The Difference Engine, the show for tech founders, investors and innovators.

0:09so what's coming up today we'll be learning from the past to find out what the future holds for category design and we'll take a hard look at europe's unicorns and what they tell us about exclusion but first we're peeking behind the wizard's curtain and revealing the quiet power behind the ai boom it seems not a single day goes by literally a day without some sort of AI acquisition or merger. I don't know how their comms teams figure out if they all get together in a secret WhatsApp group and say, you do yours on Tuesday, I'll do mine on Wednesday, etc. It seems that model companies, LLM companies, generative startups, or agent frameworks are all the rage.

0:53I can't help but feel that the long-term winners in all of this are being overlooked. And I'm sorry to be boring, but we're talking data management. and there's a lot of category design work still to be done there to define just where their value lies and how they show it to the world. Yeah, well, we've talked about this ad infinitum, probably over a few pints, listeners. But I have to say I completely agree with what Paul's talking about that and everybody's chasing the algorithmic gold. But if we know anything from the history of tech, the real value lies in the picks and shovels, enabling successive boom-to-bust gold rushes.

1:32Now, data management, I think, in my opinion, certainly, is the quiet yet infinitely powerful infrastructure layer, the one that actually determines whether AI works in production or not. Yeah. Just this week, we were playing about with a number of AI tools. We had some breakthroughs as a team live in the meeting when someone made an AI suggestion, and a few of us thought, that's impossible, and then we were proven wrong in that very meeting. So the industry has forgotten, and we learned very quickly in that meeting, that the AI is only as good as the data behind it. And some of these models, the large language models like Lama, Mistrell, Falcon, DeepSeek, yada, yada, hundreds of more, they seem to be in a zero-sum race where there's going to be winners and losers.

2:22But it's distracting everybody. We don't think the differentiators is the model anymore. It's the data quality, the governance. if you like the lineage, and that's the domain of companies like Calibra, Alation, Informatica, and Databricks. To stick my M &A hat on here, firmly pulling it down over my head, look at the valuations. AI firms are valued at massive multiples of forward revenue, even when they're burning cash at an enormous rate. Meanwhile, the data management companies, many of them profitable, recurring revenue-based beautiful businesses from from any valuation point of view sit at a fraction of those you know really inverted common sexy valuations um there's a there's a huge huge mismatch ultimately between perception of impact and structural importance i know i'd rather be the owner of an oil well than the owner of something manufactured with plastic and you know it's a cliche to say data is the new oil but it's wild and there are firms that that make um ai models and there are firms that make that make that those ai models viable starve them of the data what have you got not a lot and these are the companies that solve the really dull issues like discoverability data duplication extraction loading integration compliance These are the real headaches that are going to come back.

3:49And they're probably the reason that the MIT report talks about the vast amount of failure, the failure rate. Yeah, and something I don't think the MIT report went into too much detail on was regulation. And I think that regulation is about to supercharge demand. If you look at the EU AI Act and DORA in financial services, enterprises have to prove that their data is traceable, explainable and bias controlled. That means more spend on governance and lineage tools. In effect, compliance is becoming the growth engine for the data management section. To be fair, that's a very European viewpoint. And a lot of free market economists, mostly in North America, would disagree with that.

4:38But I do agree with it. And it's typical that the best solutions, you can think of the mini and the iPhone, they come from having restrictions, restraint and constraints in their design. and regulation is a bit like that so you know if you regulate something and say what it can't do the very best category designers the very best designers of any kind will realize that that is what's going to supercharge their success so regulation therefore might not be the enemy of innovation it might be the catalyst certainly for infrastructure investment we're seeing that already every chief digital officer every cto has to have a proper data backbone and these days they might be thinking about having another one than AWS.

5:26But they need to have a proper data backbone before they can even start extracting the value from their Gen AI and other AI applications. So oddly, these guys, the people that are the buyers, the recipients, may have a lot more power in the future of AI than the media. Yeah. And if you want to see a great example, sports fans of how regulation drives innovation look no further than Formula One racing. And if you look at the level of innovation that's taking place in race cars, telemetry systems, pit organisations and so on, and then maybe compare that to something like NASCAR, you might see quite a difference.

6:09Jalopies going around an oval, that's what comes to mind. That's really exciting, isn't it? I mean, that's down there with cricket in the excitement stakes. you know it's so weird that ai gets the glamour but it's data management that is actually the thing which is indispensable um you know these firms not to find a point to have sticky revenue high renewal rates and mission critical status if one of their platform goes down the entire data fabric of anything from a bank to a to a hospital or entire health system collapses and the ongoing implementation of AI makes their role even more critical.

6:52And yet they're undervalued because they don't have that AI sizzle. But I think, I do think, I do think it's changing. And what I think we're seeing is an AI-driven M &A wave forming where one where buyers move down the stack to enable the imagined AI applications everybody's getting so excited about to enable them to really deliver value. Yeah, I think the smart money's starting to move, right? Databricks bought, and I hope I pronounce this right, Arceon, Archeon, I hope it's not Arceon. That'd be a bummer, wouldn't it? That'd be a bummer, no. And Salesforce bought Informatica, it's much easier to say.

7:34These are signals that people understand that the devil, maybe the dollar, is in the data. You know, a lot of hyperscalers will need these players, so we may see some interesting M &A action there as well. Right, and this is the point where I start to talk about history. So, you know, we should know our history and act on it. It's the same, you know, think about it. It's the same playbook we saw with cloud computing. The early money went into the app developers, but over time, the infrastructure layer, And in that case, it's AWS, Microsoft Azure, Google's GCP captured the real margin and the market power, just as Cisco did in the underlying internet build out in previous years.

8:19Right. And these are the parallels I think that it's easy to miss, right? In AI, controlling the data layer equals essentially controlling the AI economy. Front end models and applications may come, they will go. But the data platforms are the connective tissue. And whoever owns that layer won't just enable the performance of their business. They will ultimately have a stranglehold on the performance of entire economies. And they own the customer relationships, not so much LLMs, which are very switchable. And, of course, the recurring revenue. What are we all in business for? To make a profit, ultimately.

9:01So from an investment point of view, all those lovely data-layer SaaS companies are still a more defensible bet. And predictable cash flows, real customers, positive margins, still a huge, huge upside as AI adoption scales. and its real use cases, the really, really important use cases, which we know from previous technology waves aren't the ones you thought they were going to be at the beginning. Let's see what happens when they start to emerge and then need that data layer to support it. Yeah, and I think this computing enabling infrastructure, which really could do with a couple of categories being built.

9:44I think it could. Yeah, we're here. We're available for higher. Once the hype cools and people realize that valuations really are in the wrong place right now, and maybe some of these data layer companies will see the valuations rise, it's sort of inevitable. Right. So maybe the smartest move right now isn't to chase the next ChatGPT clone. It's to invest in the infrastructure categories that make ChatGPT et al. possible. Yeah, so I think we're all reaching for our keyboards right now, phoning up or broker or whatever one does, reaching for their phone to make some bets. This enabling infrastructure is going to be the categories that matter, we think, down the line.

10:27It's, if you like, the quiet power beneath the AI boom. Yeah, I mean, you know, in every tech revolution, you know, infrastructure has won in the end. But there's no reason to believe that things aren't going to be any different this time. that the enabling infrastructure is in the categories that make up data management. We'll do it here first. Are we on the right track? Subscribe, drop us a review, and tell us what you think. Is the AI frenzy overblown? Are we looking in the wrong place? And does control of the data layer equal control over the AI economy?

11:07right let's dive into europe's uh unicorn scene those shiny billion pound dollar euro startups and potentially category leaders are everywhere everyone likes to talk about them but we're asking the tricky question is a success story proof of meritocracy or a sign our ecosystem is still today a gated community. Right, because the data this year is, well, let's just say familiar. 16 new unicorns across Europe, 27 founders in total, and get this, every single one of them is male. All X, Y, no change there. And when you dig deeper, the pattern we've seen gets even clearer. Average age, 32. Almost half of them, 92%, are uni graduates, and more than half have a master's or a PhD.

11:56Yeah, right. And, you know, plus a not surprisingly high concentration of elite institutions. 16 % went to Oxford University, 12 % to KTH in Stockholm, another 12 % to the Instituto Superior Technical in Lisbon. So we're talking about a very specific academic pipeline for unicorn founders. And you might say, fine, but it's not just education. The disappointment continues. the corporate pedigrees are what you'd expect. Where are their alma maters? Deloitte, Goldman Sachs, DocuSign, Adobe, all big name firms that are very safe bets for early careers. And they all seem to be incubating, by the look of this list, very similar types of unicorns.

12:45And working there gives you a lot of experience. It also gives you quite a high level of remuneration, which de-risks the next generation of unicorn builders. Now, you might be thinking we're sounding a little bit disappointing of the same old, same old. I think we need to point out it isn't bad per se. I mean, these are people who've clearly earned their stripes, they've worked to get to the top, they've seen how to scale and watched an execution that really works. But four of these even came straight from academia, The team is behind Oxford Ionix and Organox. So it's not just tech and finance bros here.

13:23There's deep tech too. That's pretty heartening. And there are some interesting outliers, in fairness. There's four solo founders this year from Tide, N8N, SANA and Nothing. I think they're the phone people. That's the name of the company, obviously. And an outlier in that the founder is actually, I think, of Chinese origin. And of course, Daniel Ek. You may be listening to us on Spotify. the Spotify founder who's responsible for a huge European founder cohort and now has his second unicorn with the health testing startup Nico. So, yeah, there's a little bit of giving back, a little bit of flywheel going on and some entrepreneurial stamina there.

14:06Yeah, I can't help but admire that. But still, all male. In this year, 2025, you'd think, with all the drive for inclusion, would start seeing a shift by now. Yeah, maybe. It's a difficult one for us, and we're tackling it despite that. It raises the uncomfortable question, is this a merit-based free market situation? Is it coincidence that the people who are bright enough and most willing and able to take on the risk of building billion-dollar companies are all so similar? Or are we seeing the effects of a closed loop, a white boys club formed around access, privilege, and networks? Ultimately, I'll lean towards a bit of both.

14:43You can't ultimately fake a unicorn, although they'd argue that some are still trying. I can think of a couple which I won't name. If you look at this, when that funnel is so narrow at the top, all male, mostly white, mostly elite university graduates and company alumni, you have to ask who's not getting into the funnel. Yeah, and policies around inclusiveness and diversity evidently, at least to date, haven't really moved the needle, at least not at the unicorn level. Yeah, I mean, they just appear to be scratching the surface. So, you know, disappointing. You know, maybe good intentions are not the answer.

15:22This has to change. We can't carry on, you know, having, you know, male elites running everything because, frankly, it's just not good for innovation. But, you know, maybe, maybe, we've thought about this a lot because it's something we care deeply about and have done throughout our careers. and maybe there is a bright light coming through you know the the next wave of ai and decentralization could change that now according to the latest uk figures it's manchester that's actually leading london by some way now in creating ai startups so we think that's a that's a good sign you know after all because after all if you think about it the barrier to entry and building something massive is now shifting.

16:06You don't need 200 engineers and a pile of VC money from guys in Patagonia gilets from day one anymore. Yeah, and we've said this on previous episodes and previous segments, a two-person or even a one-person AI startup could conceivably become a unicorn now, and that opens the door for lots of different kinds of founders, backgrounds, motivation, geographies at all. To be realistic about this, it's only going to happen at scale if existing investors and networks are ready to back them and new ones, new networks with very, very different priorities like BoardWave come forward. Otherwise, we'll just automate the same old bias with better tools.

16:51Yeah. Europe's sex scene needs to change, we think, and can it evolve past this echo chamber? can we build a truly diverse ecosystem? As you say, BoardWave is one of the organizations leading the way on this. And just avoid this repetition.

17:16There is a perception about, especially in B2B technology, that categories as strong as ERP, CRM, SaaS, video games, even smartphones, a laptop. So there'll never be a new category again because AI is concentrating all of the resources and all of the AI elite specifically within US big tech. So it's certainly capturing well over 50 % of the available investment funds at the moment. Yeah, not the same thing. I mean, if you believe that, you're neglecting the lessons of the past. where massive categories were built on the most promising recent tech innovations. And we think AI will create new categories.

18:03And so rather than fewer new tech categories, there are going to be more. Oh, nice use of English there, I thought. Correct grammar. I do like that. So, I mean, if you look at the largest tech franchises of today, they're all built, as it says, on the English pound coin, if you read around the outside, on the shoulders of giants. Oh, also an Oasis album, Best Forgot, that one, I think. And an Isaac Newton quote originally, wasn't it? It was. A man from Northamptonshire with an affinity for apples. You know, without radio, no smartphones. They are just radios, right? Pretty much what they are. Yeah, without smartphones, no apps.

18:39Oh, God. You know, frankly, without radar, no microwaves. I think also, this is what I love actually, thinking about it, was without video games, no GPUs. GPUs. Wow. No Pong, no NVIDIA. No$5 trillion valuations because without GPUs, no AI. Without AI, quantum stays a science fiction idea. Right. And it ain't going to be. It will come. Great point because critically, I think AI and quantum's futures will be wedded together. Lots of folks, lots of folks though, particularly those ones who are pleased with their progress in the first century of IT, which we're coming up on, are pretty incapable of seeing past what a lot of people are calling the AI singularity in a very negative way, right?

19:27These are the doomsayers, and there's many of them, who think AI closes down the ability for new categories. It's everything's going to be AI. We disagree. New categories will be forged and fought over and ultimately dominated by players, some existing and many more yet to be minted. Yeah, I mean, the reality is we don't need to let the AI haters doom loop us into forgetting that different store matters in a world of AI slop and converging views. You know, the pirates, the punk rockers and the modern equivalent are working hard to stand out from the crowd. New category kings are coming. Yeah, and category theory, category design dictates humans are curious beasts.

20:15We mix thinking fast and slow with a heavy degree of human judgment. Judgment, judgment, that thing they never teach you at business school. That thing that AI does not have. And in the West, we all intrinsically know when a car is aimed, a new car is aimed at the utilitarian or the luxury end of a market. We just know. We've been conditioned to believe that some garments, which are retailed at many times, many multiple times the price of others, which are pretty much identical, we think that's worth more. We make a judgment. And a night's sleep in some hotels is worth many more than its rivals.

20:53We are constantly categorizing new experiences, and that ain't going to stop with AI. This categorization of goods and services happens. To put too far on a point on this, even when market forces change the objective cost of purchases. That's interesting. So the cost changes, but the judgment remains. Yeah. So a luxury EV now from China can challenge the dominance of a Tesla. The rise of fast fashion does not decimate, to use the proper sense of the word, at least one in 10, sales of luxury brands. Meal deals, of which, frankly, listeners, I'm very fond, still appeal as bargains regardless of the costs of the individual items, which could cost up to twice the amount you actually get charged for your fabulous meal deal.

21:46What matters in all of these examples is the categorization. A meal deal is a meal deal. Right. In America, a meal deal is a discounted bundle of three items that supermarkets charge here to great success. So how do all these obscure examples relate to our world of technology? um there's some clear parallels um think about how categories morph and merge i guess a good example is sas so sas products which would do one thing really well for instance hr payroll tax accounting are now consolidating into suites especially in the mid-market think think about sage or zero and what they've been up to recently.

22:36Yeah, incredible recategorization of what were separate categories merging. Messaging apps like X, formerly Twitter, Facebook, or sorry, Metas, WhatsApp, finally, finally becoming super apps. And we saw that recently with OpenAI allowing people to buy things straight off ChatGPT. Yeah, and of course, there's the classic example, which is Google's progress from a search engine to the Microsoft Challenger and more. Yeah, so those are three examples of merging. Now thinking about the opposite, where new categories emerge. You've got Dell's heavily advertised laptops. They're laptops, but they're called AI laptops.

23:20Yeah, I love phablets. They've got almighty phablets, looking for a product, looking for a solution and the latest rebrand is really foldable phones. I'd recommend that as a great secondhand buy. Yeah, I'm sort of cynical too, but having seen one in action, they're pretty useful tools, but a new category, fablets. And then Spotify is trying to move from just a place where you listen to pods like this into more of a creator platform. And if you recall, we have had the the founder of, sorry, the guy that took TikTok from nothing to a lot in EMEA on this very pod. So all of these are creating brand new categories.

24:03So why on earth would it be that many believe AI is going to crush creativity and force us all to be takers and not makers? I think the layers on top of baseline AI are where the future categories will emerge. Seems to be so, yeah. and you know and and frankly are emerging at the moment the the world will undoubtedly move ai at different paces although the laggards as is always the case will be severely punished um and this is a historical change because of the rate of development um much more so than those that held out against sass and cloud until they had no option other than to go there which is no bad thing.

24:49I think we're all pretty depressed with the sort of consolidators in the world of tech, just who bring up, frankly, revenue streams and milking them and not adding any innovation. But the big question we're all asking is, aside from the LLMs, who are these new likely category kings and queens and the largest? The LLMs are clearly going to be the largest companies the world has ever seen. They're already there. The winners are pretty hard to pick. But let's face some facts. you know that that game is at least halfway through some would say reaching endpoint and um you know the llms uh we've got enough of them um and with open source we've certainly got enough of them um there'll be some sub models but what's the potentially killer use cases that we're seeing right now how about um micro llms for phones yeah the right move to protect secrecy as edge AI computing becomes possible.

25:43Yeah, and we heard that on a recent pod here with Amber Vodigal talking about, in her case, the need for a micro-LLM on individuals' phones, which protects their database and keeps their privacy for sensitive things like female health. I'm possibly thinking here government-sanctioned agents, you know, RAG. The problem with that is, is that AI's future guardrails or simply just an enforceable pipe dream? Yeah, you can see this already. I mean, you know, here in the UK, we're well advanced on making tax digital. Not at the forefront. That would be probably somewhere like India. But yeah, retrieval augmented generative AI or RAG as they call it, where, you know, some of the data is held specifically for a certain purpose, perhaps to fulfill a government service.

26:32Definitely see that. I think if there's a third one, and we do like this, there's three here, it's probably going to be neuro-symbolic AI, which I mean you're going to hear a lot of that that particular word in future not the AI bit the neurosymbolic absolutely is that just going to be a some sort of vain dream to look into the black box of AI or is it going to be about humanizing AIs in inverted commas yeah I'm going with the I'm going with the former if you think about the success to date of LLMs it's all about words as symbols and largely and so you know what is the next best answer to a word in a sentence and you know we fall for the magic trick as humans and we say oh my god that thing can think it's just it's just plugging in language given all of this what would we say that the category designers of the future look like yeah well firstly don't you know don't be disheartened right there is plenty to play for there are going to be amazing businesses you know aside from the llms that AI will enable.

27:34And almost certainly, whatever those future category leaders are, they're going to be human. Because even in the world of AI to date, humans still make the buying decisions. So when we talked about what makes a category, why do you think of a car as a luxury car or a cheap runaround, humans categorize. So let's remember that humans will make those buying decisions and they will categorize, not AI. No matter how much AI leads them to any particular decision, that human will still make the decision. They will use judgment. The other characteristic, I would say, the second characteristic is that they're going to be AI augmented from the start, which probably means great things, right?

28:17It'll be a lot of deep research, breakthroughs on maybe healthcare and energy creation that we've never thought of before. So it just raises the bar. And I think the category is going to be really excited. And the other thing that you can do now with AI that I think is a real game changer is you can test ideas at massive scale before committing. So, you know, people talk about the one person startup. That's possible when you can take an idea and, for instance, trial advertisers, as WPP and others do, a thousand different executions of an ad to see if there's traction there from humans. so yeah and the other thing we're seeing is is that you know with google zero we're seeing the rise of ai optimization and you know being able to really think about what a customer needs in ways that they couldn't possibly express when they're just typing into a little google search box so really exciting times to build and trial new company new ideas there's so much power to be had from trialing ideas at scale and getting them into the hands of humans i think at present and I say at present because who knows what the future brings, humans make these buying decisions for your category.

29:29And especially when it's high ticket, high leverage items like technology, you know, these are not Temu or Shein impulsive purchases. These are thought through. That still happens. And while that happens and the process of buying relies to some extent on human intuition or human categorization, we're still going to need humans in the loop. There's everything to play for. Thank you for listening. If you want to learn more about category design, head to becategorical.com. If you need help designing and dominating your category, then get in touch. Contact details are in the show notes.