Arrived in Hangzhou for my last day and a half. Alibaba has installed what looks like a mini data centre in the Arrivals terminal just to be not at all subtle about where you are. There is an adopt-a-cow campaign running which I can’t tell is real or spoof.

Driving through the city on a dark, wet and misty evening with incredible buildings (and notably incredible lighting), it was a fairly intimidating welcome alright.

Went out for a walk to find some dinner. Total absence of westerners. Walked into the restaurant and the greeting girl actually did a full double-take when she saw me.

This is one of the local specialities: pork belly in a sweat treacle sauce. It’s unclear what flavour role the egg is meant to play but it seems to be a critical inclusion in every version of the dish I saw. Like many regions in China, the vegetable side-dishes are often the real stars. This was a mixed dish of chopped wild garlic, other greenery and tofu bound with some sesame oil. Magnificent.

I had ordered a couple of other dishes but someone came racing out of the kitchen to protest the sheer volume for one person. We compromised on some duck which had been cooked with an amazing smoked tea-like lacquer.

China feels decidedly less alien than the last time I was here. I think that’s down to a couple of things: how much more consumerist it’s become (so it looks more understandable to my eyes) and how comfortable everyone now is to communicate through translation apps (this is largely down to the new generation that are commonly running shops and restaurants). Ironically I feel it’s become 100x more accessible just as all the westerners seem to have stopped going there.

I am not a car guy. But I would fully buy a Huawei car if I could do it next time I upgraded my phone.

Tuesday

Took a walk around Hangzhou’s West Lake while on a call to someone in the US. I’ve been pleasantly surprised by how much my western working stack (gmail, slack, WhatsApp etc) basically works completely fine here with a VPN set-up, or often just directly on my cell data connection.

I am confused by chocolate here. On one hand I paid about £5 for a small selection of artisan truffles (about half a car journey to the airport) which makes sense as I understand current cocoa commodity pricing. But then on the other hand I stumbled upon an exhibition of dinosaur skeletons entirely sculpted from chocolate and I feel there is an arbitrage I’m missing.

I reset my economic and flavour palate by finding a tiny bakery which sells what I think are scones but which turns out to be corn bread. It costs about ten pence.

Retail is so interesting to study here. The lack of general consumption (in an economic sense) is a well documented (and high priority government) issue and the new malls seem deliberately designed to address that (speculation though, I must research more). Having been in a few across my three city tour, it’s always interesting to see the emerging brands section (often tucked into the lower floors somewhere). Curious whether this is a state policy or just normal commercial practice. This morning I was in an emerging brand section which had basically turned into an entire mini-mall focused on tweens/young teens.

Of course, maybe I’m totally wrong about this. From the FT:

“There is an ideological hardwiring at the top of the Chinese hierarchy to favour production over consumption,” says Daleep Singh, a former White House adviser under Joe Biden who is now chief global economist at PGIM, the asset management group. “China will continue to rely on the rest of the world to absorb their excess production because the domestic political cost of empowering their own consumers is too high.”

Perhaps, although I think the success of China’s domestic car industry might contradict this point somewhat.

I covered a lot of ground across the city today, especially in retail and some more tourist spots. I saw less than half a dozen westerners. Granted Hangzhou is probably not an obvious destination but still. I check my observation with the guy who runs the coffee shop I’m sitting in who says that most tourists he encounters are now from Japan, Singapore and sometimes France and Italy.

I never think of China as a food-first culture but it absolutely is. As much as (indeed moreso in many ways than) Italy or Spain. For my last evening, I did a little food crawl in a night market. It is of course magnificent. I leave this trip understanding Fuchsia Dunlop’s obsession a lot more.

On the way back to the hotel, I contemplate which specific demographic this startup was targeting with its range of electric mopeds.

And so ends a far too quick sprint around China. I’m going to write up my general reflections in a different post. I’m so glad I came -I think it’s so important for any founder or investor in the west to actually visit and see China in person rather than just reading about it (with apologies to Dan Wang).

Another flawless pickup (Trip really has this process nailed). Mid-range EV to my hotel is ~£15.

It’s been about twelve or maybe even fifteen years since I was in Shenzhen. I remember literally driving through construction sites. Today it’s hard to imagine any city in the world that feels more like an anime version of ‘the future’, far surpassing Singapore and Tokyo (probably several years ago). Although saying that, looking out of my hotel window, there is still construction going on everywhere.

Saturday

A better view from my elevated vantage point.

I know I said it earlier but for a city which has grown so quickly, it’s amazing how much construction is still going on here.

Grab the metro to visit Nantou Ancient City (SZ’s historical starting point). Surprisingly their ticket machines give me a billing error. First I’ve seen. By far the only westerner in the station and on train (which feels much less English-friendly than Beijing system-I guess Shenzhen isn’t a major destination for tourists).

Nantou Ancient City is fascinating for what it is and isn’t. It’s an attempt to maintain the city’s origins as a small trading town. But then they’ve turned the whole block into an integrated outdoor mall which combines boutique designer stores (albeit in a very Gen Z way), art galleries, lots of eating options and the usual mix of coffee and dessert spots. It walks the line between tacky and quality, often lurching wildly either side. Lots of younger teens, young families and a handful of somewhat confused tourists.

Definitely more Teslas in Shenzhen than Beijing. Fascinating how every major phone/telco company here also expanded into EVs.

It’s crazy to see the sheer number of Chinese EV brands. Quick research says 25! Down from 100 🤯

Curiously although there are rental bikes everywhere, it’s all regular pedal models, no e-bikes. Or scooters. Not sure if they skipped completely or just not in the cities I’m in.

I bounce across to Huaqianbei (HQB) to get a sense of the scale of electronics manufacturing in the region. Multiple buildings rammed full of electronics re-sellers (from components to full products) and many sales offices for manufacturing companies.

Interesting to see how quickly companies started shifting their marketing to Openclaw hardware.

Sunday

Not sure how localised this is to SZ but I’ve gone to another hybrid design/coffee park area (Oct-Loft). Think outdoor mall but with an emphasis on art galleries, design/brand consumer stuff, coffee spots (many) and a handful of restaurants. Partially an effort to see what Gen Z is doing here on a Sunday (would love to know how representative this spot is).

There’s a clear retail effort for highly aesthetically pleasing design to drive social action.

Also semi-anchored around various exhibitions:

It’s quite a good concept. Curious how long it’s been in existence. Vaguely similar to Meatpacking District or Soho vibes in NYC and Covent Garden in London although this is more a destination area than part of the core city.

So many photo shoots!

Dinner in Nanshan city. Probably the part of SZ which feels most like the future although absolutely helped by the brand new shopping mall complex which has been built out just six months ago (a lot of the surrounding buildings are owned by Pony Ma, Tencent founder).

Incredibly high end finish with Chinese brands (electronic, telco, fashion, drone) and major Western brands sitting side by side. And then some very boutique things like this specialist keyboard store which I need to come back to (I was on a pretty rushed tour).

Naturally you had drone food delivery happening, which almost felt predictable. This isn’t happening elsewhere in China, SZ is the testbed.

Dinner with a local investor who was educated in China:

• Geopolitics between US and China still makes a lot of business tricky but the jump in AI hardware investing in Silicon Valley has improved things as all that business is coming directly to SZ (via a lot of former Apple employees). US activity keeps pushing more countries to China

• Feels the US is now both too expensive and lacking innovation. I suggested China was 50% of London cost. He said probably lower and you could still live here for 10% of it if you really wanted to.

• Biggest difference he sees over the last decade: the Chinese have now found their identity and confidence. Still need to partner in areas internationally but 100x more sure of their capabilities than a decade prior.

• He counts 25-50 Chinese car brands as serious players but thinks Huawei are #1.

• Comparing SZ to Beijing, he said that it’s still true (as historically has been the case) that SZ is far away from the direct eyes of Beijing so they get more latitude on things

Monday

I read China’s latest 5yr plan this morning. It’s remarkable. Once you see a KPI for the number of patents to be registered, you begin to realise how seriously they take planning (and how engineering driven it is). Worth reading a summary. I suppose you can certainly eye-roll a bit at centrally planned capitalism etc. but on the other hand, I couldn’t imagine a single western government being capable of this.

It’s unclear quite what the vision is for SZ’s urban landscape but looking across the city from my hotel I see:

An AirPod monument (this is near the mall I was in last night):

An Eiffel Tower:

And, some pyramids:

Idk.

Later that morning I have a series of coffees in a square kilometre which houses some of the top engineering startups and AI researchers in China (and by extension, the planet). Apparently they all go to this one Starbucks (despite the abundant excellent coffee in China, somehow Starbucks retains a position of value). This is probably the first generation whose parents didn’t feel the need to send their kids to the US for university. A couple of examples:

• A 19yr old co-founder building a data capture hardware product for factory floors to train frontier models

  • Patiently educates me on the state of models for robots. Still in data constraint stage (versus software which is now compute constraint)

  • He’s focused on grabbing data from the tier 1 factories, selling it back plus process improvement to the factories and then selling data to frontier models

• A former Apple product person building assembly line robots

  • I pointed out the lack of westerners in SZ as well and it’s not my imagination. He thinks it’s crazy that more US and westerners aren’t coming here.

  • He talked about the difference in the factories now versus a decade ago. Today when he visits, their car parks are full. Combination of increasing salaries but also very cheap EVs (and govt subsidies for them).

I take a walk around another shopping mall (MixC, operated by the same developer as the one I was in last night).

Interesting to see a Korean brand called Gentle Monster which operates concepts around experiential retail and sunglasses. Similar to the style I saw in Oct-Loft park over the weekend.

It certainly generates a lot of earned media views.

Inside the mall is a similar high end mix of Chinese and western brands.

Head to the airport. £10 for a 30m luxury Didi ride. How can this be profitable? Maybe it’s just a giant data play?

SZ airport has a similar design to Beijing Daoxing. High end retail, vast amounts of space. I drop into China Southern’s business lounge but it’s surprisingly shit (I guess that particular budget sits with the airline).

China Southern itself is a delightfully old school airline. I’ve flown two legs in business (which is absurdly cheap, I’ve paid more for a one-way train ticket to Manchester) and it feels like the kind of service I’ve read about in pre-90s airlines. They insist on me wearing slippers, make a huge fuss about getting anything at all I want (I’m the only westerner on both legs but I genuinely don’t think that’s the reason), there’s a white cloth food service (even for this 90m leg!) and I think I’ve previously mentioned their champagne flutes.

Surprisingly they seem to squeak out a profit despite this kind of exuberance and I’m extremely pleased to discover this (curiously American Airlines is also a tiny shareholder). Somewhere here is a joke lurking about profitable airlines only existing in communist countries.

Next stop: Hangzhou.

If you are a frequent China traveller, you can probably skip these posts. On the other hand if you haven’t been at all or in a long time or specifically haven’t been to Shenzhen (which is note 2/3), then these might be interesting.

Wednesday (Beijing)

Beijing airport is HUGE. Probably the biggest I’ve ever been in. Didn’t take pics because immigration discouraged.

Had scheduled a driver to collect me and bring to hotel. All done flawlessly through Trip (formerly Ctrip). A ~40m journey in a mid-quality car with driver who met me at arrivals was the equivalent of about £18.50.

It’s surprisingly quiet here-major EVs and people not being rowdy. Maybe that’s Beijing? Or is it a lot of single people?

It’s so cheap. I mean maybe half or even a third of the west?

China has a food culture! I always forget this when thinking about it. Dinner is obviously roast duck because I’m in Beijing. I sneak in to a Siji Minfu about 10m before the main wave. On the way out, the queue looks like about an hour easily. The duck itself is only average but the side dishes are excellent.

I easily ordered enough for two people (half a duck and a bunch of other dishes) and it cost 374CNY or about ~£40 (this is a fairly high end place). In London it would comfortably have been about double that.

Thursday (Beijing)

It’s raining so I grab a DiDi to start a mini coffee crawl. It takes about 20m to cover ~2km at rush hour and costs about 20CNY (~£2). They have a turn-right-on-red concept similar to the US (quite a bit of US driving behavior was imported to China during WW2).

Surprising how much of a coffee culture exists here. A lot of excellent pour-over places, most of which are up side alleys but have seen a couple on larger streets (which obviously have the ubiquitous Starbucks and Luckin brands).

This place (Mer) also serves mini scones, which was quite lovely. Coffee prices here are vaguely similar to London but that reflects coffee bean prices for my rather elitist caffeine tastes (I order pour-overs of imported beans).

While I’m not in an especially tourist-heavy part of Beijing, I’m surprised just how few westerners there are around. Not so few that I’m getting stared at but definitely few enough that they’ll catch my eye at least.

I’m not sure how recent this is (although Reddit seems to suggest it is) but a lot of Chinese services (eg Meituan, Dianping) no longer work without a Chinese number. However you can absolutely survive on just WeChat, Amap and DiDi.

Got lunch in a random (although well reviewed) noodle spot vaguely near Forbidden Palace. Staff are consistently very patient with me pointing at pictures (and other creative ways of ordering) as long as I have Alipay.

This was £3 and excellent.

One of the most obviously divergent cultural trends I see is the prevalence of smoking here. Definitely less than the last time I was mainland but still pretty common among younger people than you would see on average in many western countries.

An interesting chat over another excellent coffee with a founder running an enterprise software business:

• Very bearish on Chinese economy. Thinks leadership is doing a terrible job and getting too socialist. Apparently local government just rolled our new law saying that firing anyone requires negotiating with their union

• I mentioned the lack of westerners around Beijing and this isn’t me. Post-Covid, so much has changes and China/Xi just not bothered about trying to bring people back

• Hugely bullish on AI. Has reduced headcount from 120 to 75 but increased revenue. Thinks he will reduce headcount again to 50 by end of year but continue to grow revenues. Using US models for development because they continue to be 6-7 months ahead but Chinese models for workhorse inference, so much cheaper. Thinks Anthropic is still undervalued at least relative to OpenAI.

• The Chinese VC ecosystem has shrunk massively. Most funds are now reliant on state LPs. Most US funds gone. Angel investing has shrunk by about 90% for him and his friends since 2021. Because companies basically have to be approved for IPO, unless you’re in a strategic sector like silicon or AI, a liquidity event will be difficult

• New generation of Chinese founders are more likely to have been educated in China than in the US so tend to think very differently, much more aligned with China superiority complex however also less ambitious because less capital available to them.

• Noticing far more Gen Z Chinese wanting to get government jobs because they’re safer, feels they generally lower work ethic

That night I go to dinner with some Australian investors who say that, contrary to my observations, China has actually opened a lot more in the last year or so to tourists. However the mix has changed radically. In 2008, US was ~90% of all tourism, today about 1%. They lament the state of Australian politics and are rather large fans of the competent government in China. They’ve been in Beijing for about three years and say ‘it’s the best place they’ve ever lived’.

Ten years ago, you’d see famous actors on the walls of the restaurants which were on the tourist circuit. Now it’s YouTube/Tiktok/Douyin influencers.

Beijing is full of these narrow alleyways (which have all sorts of interesting shops, restaurants and coffee spots on them). A lot have been redeveloped in previous years but now local government is coming under pressure to keep the remainder as proof of the city’s history.

Friday (Beijing/Shenzhen)

I’m flying out of Beijing Daxing airport (south of Beijing). Takes an hour to drive there but it’s a four lane freeway in excellent condition. Traffic is super-light.

Beijing Daxing is another huge airport. Security is quite thorough with proper pat-downs. They are actively reviewing every battery pack and are not at all happy with my Belkin, which gets confiscated (although made in China does not have their safety standard). My coffee grinder, which gets challenged in most western airports, is ignored for the second time. The interior is cavernous and new with high-end retail and what appears to be quite high quality restaurants (a food culture will do that).

Airlines seem to mostly ignore tickets here and anchor everything on ID (the opposite of western systems). This seems quite sensible and reduces document checking by 50%.

I’m flying on China Southern Airlines to Shenzhen and they are still serving champagne in crystal flutes. Magnificent.

Next stop: Shenzhen.

I am increasingly beginning to think about my own personal AI skills in this way too. I had a fairly exasperating experience with early OpenClaw which led me to opt out of trying to stay anywhere near the forefront of general prosumer AI usage. Jumping back in a couple of months later, the progress that Cowork and Rebel have delivered made me realise that this might be the only sane and practical approach to consistently incorporating AI into my life and work.

I suspect this may not just be me? Last week AI Enablement Insider released an intriguing survey from ~100 interviews with the people in companies who are buying AI professional services to help with the roll-out/adoption of the actual tools. It’s good, you should read it but that isn’t my point. One of the most common responses I hear from people who are being pushed to adopt more AI at work is some version of ‘I would love to find the time but I just have too many meetings’. Relatedly, a comment I hear quite frequently from CEO/founders is something like ‘I wish people would just SPEND THE TIME learning AI!’ and it’s always interesting to watch the reaction when I point out that they have the unique ability to not go to meetings whereas the rest of the company (which has been hired to literally be in those meetings) would probably get fired for that.

I have never trusted gut feelings. Now before you all lose your mind at me for saying something like that what I mean is that yes absolutely gut feelings and instinct are very much a (System 1) cognitive function. But by definition they are a whole collection of biases and I tend to think of them as a human sorting response, looking for an obvious ordering solution with a reward sensation that gives us closure that we interpret as a ‘correct’ assessment of a scenario. In his final book, focusing on noise in human judgement, Danny Kahneman calls this the illusion of validity. And like everything he’s ever written, it makes me wonder deeply about how much time is wasted on very bad decisioning processes. I highly recommend unless you haven’t read Thinking, Fast and Slow, in which case read that first.

Netflix recently announced a dedicated kids gaming app which is of no surprise whatsoever to folks in the business of kids content (as ever, Emily Horgan has some of the best thinking on this topic). I have a longstanding public bet (which so far nobody has taken the other side of) that AI frontier models (or whatever we end up calling them) will end up rolling out kids products/services to reduce churn as consumer AI behavior continues to converge towards commodity usage.

As some of you know, my quality standards for coffee are unapologetically high (and on that note I’m being pleasantly surprised by some Beijing coffee spots) but nonetheless the history of instant coffee makes for a fascinating read.

I’ve been reading a lot of 1920/30s history over the last few months (btw, I maintain a reading list for founders here) and I keep coming back to the idea that it’s entirely possible for seemingly likely things not to either not happen or take far longer to happen than anyone believed.

Dwarkesh and Dylan Patel had a very interesting conversation about an impending set of (capacity and materials-related) trade-offs which will have to be made between continued consumer electronics progress and improved AI capability. It’s really worth listening to. This is clearly going to play out in very interesting and not entirely predictable ways. You can have an abundance of intelligence but you might have to trade an abundance of device choices in order to access it. The war effort of the 1940s becomes the token effort of the 2030s? We shall see. AI as the next technology paradigm is a given but the timeline for its diffusion may be surprising.

This is not a surprising position to take for anyone who’s been around startups their whole life. Jerry Neumann just wrote a (predictably) good essay about how, for the literal industry of startup advice which exists, the success rate of startups hasn’t actually shifted in ten years. Although outlier returns have increased, venture backed companies now have a higher rate of failing than a decade ago. Building a vision is hard and nothing is a given, no matter how inevitable it might seem.

Coming back to trade-offs for a second. Asking about trade-offs is one of the more useful questions a board member can pose to a founder/management team. A lot of startups (and their boards) make decisions without realising they are in fact choosing trade-offs. A lot of unplanned scenarios happen because strategic decisions can have non-immediate impacts elsewhere in the company. It is like a slow-acting set of Newtonian business physics.

Other things

Aibl are building an AI Enablement directory for the UK and Ireland, which I think is much needed.

Max wrote a very good post on the new kind of holding company financing that’s needed for AI services. A few people have quoted this back to me.

History is not always written by the winners, it’s written by the historians, who can be a much more nefarious group of people. Ada Palmer’s book on the Renaissance is an excellent and dense combination of history and the history of the history.

There is a back story to this. Actually two.

The first is about pulling on a thread. For the second half of last year, Max, Daria and I made ourselves quite unpopular in certain circles by asking a lot of questions (both to companies and the AI teams selling to them) about just how easy or hard it was for businesses to meaningfully adopt AI tools. A lot of what we heard was stories about organisational and (mostly) human friction.

This in turn led us to almost a hundred conversations with founders of (mostly) bootstrapped AI training, consulting and implementation companies who are being hired (often behind the scenes) to help companies navigate this landscape. Your company has almost certainly hired one (or likely more) of them by now to help navigate its AI adoption. We’re actively tracking over 650, there are more started every day. Ben Evans has a delightfully timely quote in his latest newsletter:

OpenAI and Anthropic have…realised that grass-roots enterprise software adoption almost never works and giving people a tool and saying ‘it can do anything!’ isn’t useful.

Additionally, one of the sentiments we’ve heard quite frequently in our conversations is that the incumbent management consulting firms (Accenture, Bain, etc.) are simply not sufficiently AI-native to handle this challenge. Accenture’s ~$1b recent acquisition of Faculty (17x revenues!) suggests that’s a real issue. But it isn’t just the Fortune 500 level of companies who will need this support, every company is going to need some equivalent of a fractional Chief AI Officer or some kind of AI consultancy of record.

AI maximalists tend to dismiss the role of human-powered enablement and argue that agents will simply make these professional services (literally) redundant. But the evidence on the ground is pretty consistent: business adoption of AI (at every level from SMB to blue chip) requires leadership buy-in and understanding, planning, change management, learning and implementation. This led us to a high conviction thesis around AI Enablement professional services, and specifically around the newer, AI-native firms. In fact I am convinced that AI Enablement is now a far more important investment category than frontier models, data centres and power grids. Especially for Europe.

That covers the ‘what’. But the ‘how’ is also important.

Historically we have built companies which were, by design, singular bets on trends we saw emerging. 10xHumans is a holding company: a structure which allows us to place multiple bets on a long term trend which we have high conviction in. Many opportunities in this space do not fit neatly into VC (business model) or PE (size) paradigms. This is not unique to the AI Enablement space. I have talked about this evolution elsewhere: AI is decreasing the atomic unit of business and making an increasing percentage of new opportunities harder to reach for conventional fund structures. In some cases 10xHumans is building, in some cases we are investing and in other cases we will be acquiring.

Regular readers might have seen me talk about AI Enablement Insider, the industry newsletter that we’ve launched. But our platform strategy goes beyond that, we’ve invested in aibl (AI for Business Leaders), a new events and insights business for practical AI transformation in the UK. We met Richard, John, Terry and the team there quite early on and realised that as a group we were highly aligned on the challenges (and importance) around AI adoption for business. As well as an upcoming London event in October, aibl are also running a series of summits for senior leaders across people, marketing, technology and finance/ops teams. I get a lot of people asking me where they can get some function-specific AI support and this would be a good place to start.

There are (at least) three dynamics which new startups have to contend with. First and most obviously, AI has compressed the cost and team size of building software by an order of magnitude. But beyond this, it’s dropped the barrier to building almost entirely (I am long AI for a number of reasons but especially because it will increase entrepreneurial GDP). Second, the capital markets have bifurcated: an abundance of money at the very top (consensus bets at preposterous valuations) and at the conceptual (pre-seed, early traction) stages but a lot less in the middle (basically anything else). Third, a generation of operators who went through the 2021–2023 cycle have emerged with a bunch of interesting scar tissue and a much more nuanced view of what winning at startups actually means.

I don’t think you should ever build for a capital model but one of the highest leverage decisions you’ll make as a founder is being mindful of what kind of capital you don’t want and just avoiding those meetings and ecosystems. For anyone thinking about starting a new thing in 2026, here’s five models I think are viable.

1. Bootstrapped

Definitionally the most underrated category. Founders who never raise institutional capital, who grow on revenue, who hire slowly, and who own most of what they build. What’s changed in 2026 is that the tools available to a bootstrapped team are genuinely extraordinary. We are going to see so many more of these.

The fundamental logic of bootstrapping hasn’t changed: you need a customer who will pay you more than it costs to serve them, before you run out of money. What you cannot do is confuse “bootstrapped” with “figuring it out.” Bootstrapping requires more discipline than venture-backed building, not less. You don’t have a runway; you have margin. Protect it.

I have a suspicion that bootstrapping will remain more dominant in Europe than the US. But either way, on a ten-year horizon it’s hard to see how the number of deliberately bootstrapped (or seedstrapped, more on that below) and successful companies hasn’t exploded.

2. Seedstrapped

Flippantly I tend to think that seedstrapped is just the US version of bootstrapped, similar principles but with the caveat of ‘we’ll only ever take one investment’. Anecdotally I’ve seen a lot of seedstrapping by founders who have had less than stellar experiences with VC investments in a prior company.

I don’t think this is at all a terrible approach to building although I do worry about the dynamic with any investors. Look, maybe you took some money because you genuinely planned to build a sensible company without jumping on the VC-backed hampster wheel. Or maybe this was an experimentation phase to optimise for customer discovery. Either way, I think a lot of this tends to be experimentation to find out whether the thing is real. And if it’s real, you now have a choice: keep it close and build it profitably, or use the traction to raise on better terms than you would have gotten at zero. The reality for many founders who start with this approach is that things go well and the allure of a second cheque at a much higher valuation on highly favourable terms is just too good to turn down. But that’s the point where you absolutely need to have decided on what kind of company you’re actually building.

3. VC-backed (for speed)

Venture capital is still a very useful time-compression tool for the right category of companies: genuine power law markets, genuine defensibility, genuine founder-market fit.

I’ve watched the venture-backed playbook get applied to companies where the market is fundamentally too small, or the defensibility too thin, or the timing too wrong. And I’ve watched those founders spend years of their lives optimising for the wrong metrics because the model demanded it, then realising that giving away veto rights means ‘founder-run’ doesn’t always mean ‘founder-controlled’.

In 2026, the honest question to ask before raising venture capital is: “If I don’t raise this round, what’s the worst case?” If the answer is “I have to grow slower,” you may not need venture capital. If the answer is “a better-funded competitor takes the market before I can get there,” you probably do. Venture-backed building works when the race is real. When it’s not, you’re just paying for the privilege of answering to a cap table.

4. VC-backed (for capital scale)

The capital-intensive model has had a genuine renaissance in the last three years, driven partly by geopolitical urgency and partly by a growing recognition that some of the most important problems left to solve are ones that require atoms, not bits. [Sam Cash] In these sectors (Defence, energy, biotech, hardware, advanced manufacturing etc.), capital isn’t a strategic choice, it’s a fundamental requirement. You are unlikely to vibe-code a drone company, reversible computing chip business or modular nuclear reactor in a weekend.

What makes the capital-intensive game different is that your primary currency isn’t growth metrics; it’s technical de-risking. Every dollar you spend should be buying you a cleaner answer to the question: “Is this physically possible at the scale we need, and can we do it at a cost that makes commercial sense?” If you’re spending capital and the answer to that question isn’t getting cleaner, something is wrong.

The failure mode is timeline optimism. Capital-intensive companies die when founders underestimate how long the technical milestones take and over-promise to investors (or themselves) who don’t have the patience to wait. I know a few founders in this space and they have all fundamentally committed their lives to it. Not everyone can, or is willing, to do that.

5. Roll-Up

Maybe it’s the echo chambers I sit in but roll-ups seem to be getting more popular as a startup vector. I would distinguish between a new company you start deliberately to roll up a fragmented space (or similar strategy) versus an existing company with a consolidation strategy.

Roll-up strategies sound less risky than just building a zero-to-one company. The theory is straightforward: pick a fragmented space where the incumbent businesses are subscale and ideally underinvested. You buy them at modest multiples. You install better operations, better technology, shared back-office. The group now gets valued at a larger multiple because of scale and improved operations (also more potential buyers).

AI is a genuine operational lever for roll-ups which is one of the reasons we’re seeing so much conversation about (and capital available for) them. However it requires a different skill set from any of the other models on this list. You (plural) need to be a good operator before you are a good acquirer.

Choices

But founders and funders alike would do well to step back and ask themselves, “what game are we playing?“ Because the unfortunate reality is that most of them actually genuinely don’t know (Kyle Harrison)

I’m unclear whether it’s going to be the best time in history to build a company but I feel that 2026 is certainly the easiest time I’ve ever known to start one. About a million startup commentators have made the point that you should assume you will be building a startup for a long time. Despite that, it’s still worth saying it again. And it’s not just time, irrespective of model, startups require literal life force to will into being. At the very least be deliberate about what you don’t want to do.

I gave my whole life to the streets/Then the streets went and gave about ten to the Queen (Potter Payper)

A hugely underrated aspect of what’s happening with AI is the compression of generational gaps across software building (I don’t think ‘development’ is really the right term any more). We’re in a window where forty, fifty, sixty-year olds can be as current as a twenty-year old in producing software output.

Quite whether that phenomenon will translate as easily to other industries and whether software will actually eat the work is unclear but even if it stopped at software (a ~5% contributer to global GDP on a good day) it will be a fascinating experiment.

Anecdotally, it’s interesting to watch people (counter-intuitively) drop back a bit from some of the cutting edge AI chat groups, simply because they know that tuning back in in about six months will be far less painful and more productive than trying to stay up to the second on capability. I see that as a positive maturing signal.

Stephen Kinsella recently wrote a thought-provoking post about what kind of institutions a country needs to create more people who make things. I recently watched someone who would not describe themselves as remotely technical casually pick up Replit and vibe code a meal planner because they were frustrated by existing options. It makes me wonder more and more seriously whether countries like Ireland should simply focus on getting tokens into the hands of its population rather than administrating yet another government-funded startup program.

Let me mention something that I probably don’t want you to sign up to. At 10xHumans, Daria is writing a weekly newsletter called AI Enablement Insider, written specifically for people running AI professional services (e.g. training, consulting, implementation etc.). If this is you, please take a look!

Reading/listening

The 1929 stock market crash is one of those historical events which nearly everyone knows the wrong version of. Andrew Ross Sorkin’s 1929 is an excellent and highly readable account of this period, as interesting for correcting the facts as it is for describing what was happening in US society and politics at the time.

Speaking of general misconceptions about history…There are lots of cool moments in Dwarkesh’s interview with Ada Palmer on the Renaissance but my inner history nerd shrieked when she mentioned there was a document in the Vatican dating from ~1100 which recognises the existence of Vinland (or Canada as it’s known today) hundreds of years before it was re-discovered by Columbus. The things we forget. Enormously enjoyable and continues to confirm my theory that tech podcasters should do more history interviews.

Gen Z/Alpha marketplaces like Depop are fascinating ways to observe how new generations are running side hustles. Not sure what the backstory was after the Etsy acquisition but something clearly wasn’t working as they’ve just sold it to eBay for $400m less.

I posted a longish (at least for X) reply which I thought I’d expand on a bit here because I really worry that many serious people are getting confused by what is and isn’t happening.

Firstly for context, and especially for anyone who is new around here, we recently set up a holding company called 10xHumans which invests into AI professional services (and media). So I’ve spent time with ~100 founders of native AI consulting firms over the last 6 months across the US, UK and Europe. You’ve probably read (and been annoyed by) their posts on LinkedIn which make you feel inferior in your AI usage. However these folks are also the ones at the coalface of AI adoption in enterprise i.e. how it’s actually going.

Very simplistically, there are two stories here: one is the software/tech sector and one is basically everything else. For context, software/tech contributes about 5% to global GDP. Outside of software (which I’ll come back to in a bit), some observations from AI consultants we’ve spoken to:

1. In some cases ‘because AI’ is the cover for a bunch of layoffs which needed to happen because of strategic missteps. There is still a surprising amount of Covid decisions still lingering in companies. There are plenty of lower profile examples than Block’s recent AI-washing if you go looking.

2. Sustained enterprise AI adoption is much slower outside of software/tech than software/tech people realise. This is quite often down to mundane factors like a lack of bandwidth or general human factors (resistance to change, cold start, budget, etc).

3. It’s very common for AI projects in mid-market/SMB to have a cost-split of 10% licenses/tokens, 90% organizational readiness/change management.

4. The average paradigm for SMB/mid-market AI adoption remains some version of ‘I’m just waiting for a Google/Microsoft solution’.

Back to software/tech. There will definitely be job destruction/creation but overall the (early) data seems to show increased productivity (and anecdotally lower barrier to entrepreneurship). However outside of this sector, it is probably a cycle that plays out over 10yrs not 2yrs for some of the above reasons. I’ve had several conversations with consultants working on AI projects who feel that 15yrs might be a more realistic timelines for industry transformation. Either way I am not an AI bear and I don’t state these opinions with any kind of glee-I think a 1000x bigger risk than mass layoffs from AI is not enough investment going into *helping people/companies use AI meaningfully*. We need to massively accelerate AI Enablement in non-tech sectors to increase economic performance (especially in Europe where productivity is lagging).

To state the numbers, investors have put ~$800B into AI infra (models, data centres, GPU i.e. the capability) but less than ~$2B into AI Enablement (training, consulting i.e. the stuff which produces the economic gains).

Both governments and investors should be far less concerned about AI-related layoffs and far more concerned about the consequences of this imbalance. That’s worth a COBRA meeting or two.

By extension I often think about Roblox’s share price in the form of ‘how would YouTube have performed as a public company?’. An historical trawl of the ups and downs mapped to a notional share price impact suggests that the biggest video platform on the planet would have faced lots of moments when nobody believed that would be the outcome. Today YouTube is worth ~$550B.

Watching this week’s drawdown for Roblox, I feel like it’s another one of those moments. I’ve written about the Lindy Effect of major consumer platforms before and I think that AI puts that into relief for reasons of distribution. It’s really, really, REALLY hard to recreate the kind of consumer reach which spaces like Discord and Roblox have today.

Relatedly it feels like last week’s SaaS/Software sell-off also ignored this distribution reality in enterprise. Byrne Hobart nailed it:

This selloff seems overly fixated on enterprise software companies as software companies, rather than as companies whose core competency is to understand the True Org Chart, figure out a company’s needs, and sell the right products to them.

Can AI natives figure out true enterprise (and SMB) AI adoption muscle and reach before the incumbent software companies (of all sizes) figure out LLMs? Over the next decade (?) I think the outcome will be a mix. I fired up Claude’s Excel plugin over the weekend and immediately loved it (I have historically been a die-hard Gsheets user, resorting to Excel only in times of extreme modelling). Sample size of one but it instantly generated a full-year subscription to MS365 (on top of my Claude Max subscription and token spending). An interesting example of where native AI companies seep into the cracks and crevices of existing software distribution and output positive incremental productivity and revenue.

Other things happening

MTG (where I am on the board) is one of Europe’s least known unicorns. They announced a stellar set of results for Q4 and full year. And excitingly are moving forward with plans to IPO the PlaySimple division in India.

Excluding their IPO plans, there are some growing signals that Discord’s ecosystem is going to get much, much bigger. Levellr (I’m Chairman), which builds enterprise social tools for Discord, continues their mission to have every major video games legend on their cap table announced an investment round that included Mark Pincus (Zynga founder), Bing Gordon (EA), Phil Mansell (Jagex CEO), Frank Gibeau (Zynga CEO) and many others. Wildfire, which connects brands to Discord communities, also raised some investment (from us).

We have quietly started a very specific newsletter for AI Enablement operators (consulting, training, implementation) based on the many conversations that 10xHumans is having in that space. You can kind of think of it as the anti-SaaS newsletter but not in a mean way. Daria has written some very good posts on scaling founder-led sales, org structures and other operational challenges.

I re-read Richard Feynman’s report on the Challenger shuttle disaster. This is a piece of writing that accidentally finds me every few years. It’s short and really worth your time to think about how commercial pressures lead to inaccurate communication of probabilities.

And so I have been intrigued to watch the development of the AI personal assistant sub-space (e.g. Clawdbot, Mindstone’s Rebel, Anthropic’s Cowork), which you can think of as a wrapper for accessing all the cool features of LLMs that only the people who seem permanently engaged in AI groupchats have the ability to access.

The personal assistant UI will be an interesting test to see if AI acceleration in the workplace (meaning more usage, more frequently) is driven by tasks and workflows. Tasks are once-offs (build me a web app to analyse these expenses) and workflows are permanent (scan my inbox every week and grab my receipts) but the latter is a collection of the former. To that degree, I feel there is a Netflixian question lingering in the air about whether the AI personal assistants (e.g. Clawdbot, Mindstone, Cowork) can become app builders (e.g. Lovable, Replit etc.) faster than the app builders can become AI personal assistants.

The most interesting IPO this year

Assuming a fair wind, we should see some extremely interesting IPOs this year. Although I have some bias (we are an investor), I feel that Discord, which has filed confidentially, is by far the most intriguing of the 2026 IPO class. It remains one of the most underrated consumer platforms certainly for gaming but also vast parts of the Gen Z and Alpha audiences. Although the company has built up an impressive revenue base on their Nitro offering, I think a far bigger opportunity lies with community-driven marketing opportunities. Companies like Wildfire (where we are also an investor) are already enabling this for server owners (Logan spoke to Dean in Gamesbeat about everything he’s seeing on this front). Relative to the size of the Discord audience (~200m MAU), there is a multi-billion dollar set of brand flows waiting to come on-stream.

Other things I’ve enjoyed reading

Joe Ferencz (Gamefam founder) on Billion Dollar Memes in Roblox

A tracker of AI-roll ups by Sahil Patwa

Daria wrote an excellent piece about scaling founder-led sales in AI professional services (although applicable in any enterprise services offering) and how it goes against everything taught by SaaS sales.

The rise and reset of Sweden’s $19b gaming capital machine

A huge underwater archaeology find off the coast of France. So much of our ancient history is being rewritten using technology (scanning, DNA analysis etc.). Relatedly, Erich von Daniken passed away, who was one of the first to popularise the idea (albeit a rather aliens-infused version) that ancient human civilisation was a lot more sophisticated than we gave it credit for.

SuperAwesome, my previous company, acquired Starglow Media to expand into podcasts.

For context to all of this my investing activities are currently in three broad vectors. I’ll unpack those a bit more later but briefly they are:

• LFG Holdings (with Max): investing in opportunities across Discord and User-Generated Content (UGC) Gaming (e.g. Roblox, Fortnite).

• 10xHumans (under wraps but active with Max and Daria): investing in AI Enablement professional services (training/consulting/implementation).

• Personal/family office: PE, VC, private credit, special situations. Occasionally some direct company investments, usually very early, always founder driven.

I also sit on a couple of boards (MTG, a publicly listed mobile games publisher and Levellr, the Discord social listening platform) and advise a few others. So my observations come from a reasonably broad spread of data and behavior.

Broadly speaking, there are three themes I’m obsessing about.

The uninvestability of new digital spaces

One of the reasons that Max and I set up LFG Holdings in 2024 was in response to seeing new growth markets (especially new digital spaces) that were not conventionally shaped for either VC or PE. There are a variety of contributing factors to this phenomenon

Capital efficiency and business miniaturisation

Thanks to device saturation (distribution), attention saturation (content niches) and expanded monetization (particularly to younger audiences), the atomic unit of business has become smaller than it ever has been. Particularly true in digital media spaces, this has become more broadly true elsewhere and compounded by a general bootstrapping/seedstrapping mentality which rejects a lot of VC principles. AI moves this business miniaturization trend forward by an order of magnitude.

Business models

A bias towards bootstrapped building tends to go hand in hand with business models which many investors (in particular VC) are not comfortable with e.g. professional services, advertising, tools etc.

Demographics

A lot of new spaces we look at combine all the above factors with a generational gap (this isn’t just age, it’s also about definition of quality) that make it hard for investors to relate to the founders or vice versa.

Roblox (and to a lesser extent, Fortnite’s UEFN platform) is a good example of this. In the global gaming landscape (which has slowed down to a crawl), it is by far the fastest growing consumer platform. In fact, two of the biggest games in 2025 (Grow a Garden and Steal a Brainrot) were both built in Roblox. But it’s extremely difficult for funds to invest here: most game developers are under 25, entirely bootstrapped and aren’t building for an exit. The successful investors in this space have not been the VC funds but instead the (mostly) native developers who have built hybrid publisher/investor models focused on owning minority positions in a large portfolio of titles. Obviously you can buy the public stock for an aggregate position but that’s not giving you individual game exposure: If you had bought RBLX and a position in Steal a Brainrot six months ago your platform position would be down 25% but your game position would be up 300-400%.

It turns out that it is possible to know the future and yet be unable to find an easy way to invest in it. Business historians might argue that this was a similar landscape to that which Art Rock encountered in the 1970s. Perhaps. All we can say for certain is that, in several growth areas, there is a considerable amount of value out of reach for the conventionally-shaped vehicles for capital. As a result VC and PE are sitting on the sidelines waiting for scale players to emerge.

The importance of professional services in AI success

We haven’t spoken publicly about our 10xHumans strategy yet (I’d expect to be a bit more public about this in Q1) so this is something of a preview.

For the second half of last year, Max and I made ourselves quite unpopular in certain circles by asking a lot of questions (both to companies and AI sales teams) about just how easy or hard it was for business to meaningfully adopt AI tools. It has informed my opinion that a) AI will be long-term transformative but also b) that it will require a huge effort in training, consulting and implementation to help companies get there. Dwarkesh Patel made the point more elegantly in his December essay:

Models keep getting more impressive at the rate the short timelines people predict, but more useful at the rate the long timelines people predict.

AI maximalists tend to dismiss the role of human-powered enablement and argue that agents will simply make this aspect of professional services (literally) redundant. And yes, it may sound like a strange thing to suggest our AI strategy is to invest in everything except the technology. At this point we’ve built what we believe is the biggest market map of bootstrapped AI professional services firms across Europe and the US which has included at least a hundred founder conversations (more every day). The evidence on the ground is consistent: enterprise adoption of AI requires leadership buy-in, planning, change management, learning, implementation and that’s assuming their assumptions on underlying models remain consistent. Think of a time when you switched from Windows to Mac operating system (or maybe the opposite). Now imagine a company doing the equivalent of that and you’re beginning to grasp the challenge.

One of the sentiments we’ve heard quite frequently in our conversations is that the incumbent management consulting firms (Accenture, Bain etc.) are simply not sufficiently AI-native to handle this challenge. Accenture’s ~$1b acquisition of Faculty (25x revenues!) last week suggests that’s a real issue. But it isn’t just the Fortune 500 level of companies who will need this support, every company is going to need some equivalent of a fractional Chief AI Officer or some kind of AI consultancy of record.

This led us to a high conviction thesis around AI Enablement professional services, and specifically around the newer, AI-native firms. This is very much our kind of unconventional landscape, which is challenging for VC (bootstrapped founders, services business models) and PE (mostly small players) yet driven by extremely strong tailwinds.

Unconventional weapons for unconventional wars

There have always been investment opportunities in places where most people weren’t prepared to do the work. One way of describing what we invest in is PE types of risk/return operating in VC types of spaces i.e. too small for PE and too sensible for VC. This sounds like we’re driven simply by the opportunity space but it really comes down to the nature of founders we like working with. Ultimately it’s the intersection of the profitable and the interesting.

In 2025 it was surprising that we saw so many of these opportunities in spaces which have such strong fundamentals. A lot of that is AI-driven. Ultimately AI is lowering the barrier for the creation of sustainable entrepreneurship which is incredibly exciting. The counter-factual on what companies were not started because someone just couldn’t afford a developer is intriguing.

I am not here to declare the death of VC (or PE) or anything like it. But I am quite certain that we’re entering into a period where there are increasing opportunities for non-linear investors. This fracking-style model will continue to evolve. We set up 10xHumans to focus on the AI Enablement thesis and it builds on our learnings from LFG Holdings. Firstly we took some dedicated capital into the main vehicle to add to our capability. Secondly we have started to create a more platform approach to building (with Daria joining). I’ll write more about this (publicly) in a few weeks.

You can think of our approach to unconventional investing showing up in three ways for 2026.

Firstly, when we find a theme we like (rare), we look at it from an ecosystem perspective. Admittedly ‘ecosystem’ is now on par with ‘fly-wheel’ in terms of general abuse in corporate terminology but it’s still a useful model to think about the inter-connectivity of things. Within an ecosystem, we try to identify the most valuable parts and own as much of it as we can. By definition this will often mean we own interoperating parts of the same ecosystem. This is a feature, not a bug and it provides us with superb levels of data to base other investments on. For example, LFG Holdings is focused on behaviors emerging from Gen Alpha/Gen Z (theme) and we have filtered this to three ecosystems (Discord, Roblox and Fortnite). Our Discord ecosystem investments include Levellr (enterprise Discord community tools for major companies), Wildfire (Discord community activation for brands) and Discord itself (the platform). Triangulating signals from all three gives us a very well formed opinion on further investment opportunities we see in the Discord ecosystem.

Secondly, our vehicles for investing are deliberately not funds but instead holding companies. Even though it sounds very on trend for former founders to have a ‘holding company’, I think that one of the best edges to have in these new growth spaces is flexibility on how you capture value. Our north star is founder quality and that comes in a lot of different investment permutations.

Thirdly, I want to talk about creativity, which I really feel is underrated in business. Part of our edge comes from authenticity (read: scar tissue) as founders and experienced builders, the other part comes from the creative thinking which leverages that. Being unconstrained in deal structure and also growth strategy opens up lots of interesting ways to think about both investing models and scaling.

Thinking about 2026

There are lots of other people’s predictions which you can read so I won’t add to that bucket of noise here. A few themes I continue to ponder as 2026 starts to move:

• The unbundling of wealth management: I’ve written about this before but it feels like wealth management is bar-belled as either major bank or wealth management service (start-up or otherwise) and I would love a way to be able to think about this in a much more componentised basis.

• The upside in European economies that can be unlocked by AI: Although Europe’s economic situation remains pretty terrible relative to a lot of other areas, I increasingly wonder if this offers a huge upside in what can be unlocked by AI over time.

• The concept of special purpose collaboration vehicles: a concept I keep coming back to over time is the need for scale to unlock valuation step-change in markets where you have a collection of sub-scale VC-backed players. In the context of a lot of investments stuck in VC portfolios, I wonder if we’re going to see some industry-level M&A paradigm like this emerge.

It’s going to be an intriguing year.

Enablement

On Tuesday, Accenture announced they were acquiring Faculty (no official price but rumours of ~$1b which would be a remarkable multiple). Regular readers will know that I am very bullish on the AI Enablement space (which Max, Daria and I are focusing on with our 10xHumans strategy). I have very high conviction that a) AI will be long-term transformative but also b) that it will require a huge effort in training, consulting and implementation to help business get there. I was quite interested to see Dwarkesh publish an essay in December more or less saying the same thing:

Models keep getting more impressive at the rate the short timelines people predict, but more useful at the rate the long timelines people predict.

I’ve spoken to about a hundred founders of AI enablement consultancies over the last few months. There is a qualitative difference between native AI founder-led enablement firms and those which are pre-LLM. Speaking with enablement clients across Europe, it’s pretty common to hear them describe the main consulting firms (Bain, Accenture, McK) as too old-school: ‘it’s the same people who were giving data-lake strategy presentations now talking about LLMs’.

It was a good move by Accenture to acquire Faculty and potentially a genius one to make Marc their global CTO.

Reversibility

I’ve been a fan of what Rodolfo and Hannah are building at Vaire Computing before it even had a name. You have probably never heard of reversible computing but given the trajectory of the token economy, it’s likely you will be reading about it a lot more (Nvidia acquiring Groq is just the beginning of chip specialisation). Summarised to the point of inaccuracy, reversible computing is basically a very different (and historically mostly ignored) paradigm which Vaire are using to build low-power chips. The company has managed to not only gather most of the world’s reversible computing research talent but also deliver a viable tape-out.

A few weeks back Vaire quietly announced that both Stephen Wolfram (creator of Mathematica, Wolfram Alpha, and the Wolfram Physics Project) and Durgesh Srivastava (previously NVIDIA’s data center product architect) had joined their technology advisory board. I claim no semiconductor expertise outside of reading Dylan Patel’s newsletter but there is clearly something very interesting going on here.

My favourite reading/listening this year

Jessica Riedl, independent economist, deconstructing US tax myths. Have come back to this a few times. I’d love to see a version for the UK.

The FT’s write-up of Sam Altman’s kitchen reminds me of what we lost with Valleywag’s demise.

The story of Infinite Reality (a name which says so much about what’s going on here) has kept me endlessly entertained. How it started, how it was going, how it currently is. Maybe it concludes in 2026? I don’t know but surely either Matt Levine or Michael Lewis can’t be far from this.

Razib Khan wrote an excellent long essay about the overwriting of the Neolithic culture (ancient farmers) with the Indo-European (ancient pastoralists) that made me think deeply about where we are in the arc of history.

Gary Fox interviewed Sean Blanchfield (Jentic co-founder), completing a trilogy of interviews with Ronan Perceval and me that covers an arc of company building (both together and separately) across about twenty years and half a dozen companies. It’s a nice way to see how the Irish startup scene has grown over that period.

Christ Morton at Starboard released a very good set of board principles for startups (which I think scales beyond that stage). I suggest you make everyone on your board read them.

This interview with financial historian Russell Napier made me think about a lot of things.

My most enjoyed books this year

Making IT Work: a history of the computer service industry. This seems like a bizarre choice but it’s a surprisingly well written (the cover suggests otherwise) history of IT consulting since its effective origins after WW2. Behind every great technology bubble has been a lot of money made in helping companies actually adopt it.

Breakneck: China’s Quest to Engineer the Future. Dan Wang’s annual letters on China have always been excellent. His book is at least as good. The title is slightly misleading though, as this is very much a set of observations about both the US and China.

Apple in China: if you want to have a better understanding about the challenges of re-industrialising the west (especially for areas like semiconductors), read this book. Or just read it to understand a quite different perspective of Apple’s ‘outsourcing to China’ strategy over the last couple of decades. Super work by Patrick McGee

Proto: how one ancient language went global. Laura Spinney does a brilliant job of summarising the latest theories on where the Indo-European language originated. It’s also a gentle (I promise!) onboarding into the world of ancient DNA analysis too.

I can’t not mention The Power Broker, which I finally finished (and added to my personal essentials list). It’s intimidatingly enormous but start reading it anyway. You will understand New York (and other things) in an entirely different way. And maybe you’ll also understand why I disagree with Dan Wang’s take on Robert Moses.

The rest of my reading is on my Goodreads and additional listening recommendations scattered through my X feed. Possible 2026 sidequest: rebuilding Goodreads?

Oh, one last thing, my 2025 rap playlist is here. Mostly UK. All excellent.

For reasons of familiarity bias (I’ve personally built things there) and access (the main characters tend to operate quite publicly), I’ve always ended up reading a lot of media business history. And so a fairly recurring thought that pops into my brain is what would Steve Ross or John Malone be doing today if they started their business-building journey as AI natives?

A lot of Silicon Valley turns their noses up at the media sector because, as one investor memorably told me (referring to advertising but the point stands) ‘it’s just basic pricing arbitrage wrapped up in a dashboard, delivered by some humans, expensed by some lunch and then ultimately exited to whoever happens to be running Viacom at that point in time’. And yes, that is basically the business model. But the sector has been home to some of the most creative empire builders we’ve seen in the last few decades. Steve Ross built Warner Bros from funeral homes. Summer Redstone built Viacom from some drive-in movie theatres. Warren Buffett even took a close look at WPP once (allegedly, I don’t think he ever publicly confirmed it) which is a fascinating counter-factual.

One of the things I am enjoying immensely about the AI supercycle is just how like the media space it’s starting to become. I mean Sam Altman’s daisy-chaining of circular financing deals, Elon’s inevitable march towards some kind of AI/battery/solar/car/X/space conglomerate, Jensen’s growing realisation that capital scale is a strategy all seem to be making up for a couple of decades of more conventional business building. Now there are, quite reasonably, some questions about just how durable these strategies are. But as someone who has a niggling concern about the increased outsourcing of our reasoning to consumer LLMs, it’s quite nice to see this level of creative thinking.

Btw if you haven’t read about John Malone, Sumner Redstone or Steve Ross, I recommend Cable Cowboy, The King of Content and Master of the Game respectively. Media history is an excellent way of hammering home the importance of distribution.

Reading

I’ll publish my 2025 reading list in the next day or so but one last entry to sneak in. A few months back I recommended reading Apple in China and Breakneck as a pairing. Having just finished Adam Tooze’s Wages of Destruction (originally suggested by Joe Weisenthal), I would include this as added reading if you want to move from understanding the logistics of building products to culture to entire economies. I sit in a couple of investor group chats which currently have a deglobalisation theme floating around. It made me think about what deglobalisation truly means at a practical level and how so much, maybe all, of what we talk about today as deglobalisation, even tariffs, is really just at surface level.

On a more realpolitik level, it made me think about the industrial capabilities needed to wage a meaningful war. In WW2, it was really just the US and Russia which had the capability of doing this. I can’t speak very credibly on this but it feels like the 2025 version of that landscape is probably just China? It makes me extremely curious to understand how the Russian economy is actually being run at this point (suggestions welcome). It also makes me think about defense investing and the challenges it presents to governments (once you stop spending, you’re immediately creating an unemployment problem) and investors (timing cyclicality for entry and exit).

Yes, yes obviously power laws of investing are obviously not going away (as IPOs from Discord, SpaceX and Anthropic are about to make clear) but I remain intrigued by what’s happening elsewhere in the rest of the tail distribution. I tend to think of this as the space between VC and PE i.e. companies which are in a growth area but have the wrong business model for VC while also being too small for PE. The obvious reason to not go here as an investor is that you get all the downsides of both while having a somewhat capped upside. While we didn’t set up LFG Holdings (for Discord/UGC Gaming) and 10xHumans (AI Enablement) to ostensibly disprove that theory, it is precisely the space we play in. A surprising amount of alpha can come from creative thinking and collaborative building with founders.

Relatedly, as the guy who repeatedly got shushed at pre-2020 VC-backed board meetings for asking about exits and ‘who will buy this?’ type of questions, I am quite pleased to see the growing fascination by VCs with roll-ups (thank you to the 10+ people who sent me the latest Newcomer post). While kind of on the low-hanging-fruit side of creative business thinking, VC attention on a space remains a highly effective education method for newer founders. PE just doesn’t hype the way that VC can.

Speaking of the VCification of PE, a few weeks back I spent an afternoon hanging out with bootstrapped roll-up builders at the Rollup Europe conference in London. It was by far the most refreshing builder/investor gathering I’ve been to in years. The combination of energy and humility reminded me of pre-2010 SF meetups. As well as the classic search fund operators, It’s intriguing to see just how many former-VC-backed-founders are now in this space (either looking at vertical AI roll-ups or more traditional versions) and what drew them here. Talking to a few, some see it for the unit economics, others see it for the scaling leverage. Others describe it in terms of sanity.

It also made me feel quite bullish on Europe. Obviously on one hand, Dan Wang calls us a mausoleum for some reasonably solid economic reasons.

“Europeans have a sense of optimism only about the past, stuck in their mausoleum economy because they are too sniffy to embrace American or Chinese practice”

But on the other hand, seeing a new wave of founders going after Europe’s sleepy companies and transforming them makes me think about just how much upside potential there is. I feel like focusing on the place that everyone is running away from might be interesting.

Roblox

A lot of people pinged me about Roblox CEO Dave Baszucki’s interview on Hard Fork. To be clear, I don’t think all of Dave’s responses were great but as someone who has been deeper in both the technology and business of the kids space more than most, I have a few perspectives:

In an era where all the social platforms refused to acknowledge that kids really existed on the internet, Roblox was literally decades ahead by making young audiences the central consideration of what they were doing. That certainly doesn’t make them perfect but criticising the things they haven’t done (or done well enough) without calling out their investment into trust and safety is unfair. The internet would be a far less safe place for kids without a Roblox.

Critics of kids platforms frequently don’t understand that if you try and build the most restrictive environment possible, you will simply drive kids into adult platforms, which makes the situation worse. The genius of Roblox (and its contribution to society) is that they created an ecosystem which managed to appeal to younger audiences while having founder DNA which embraced the fact, as opposed to pretending it didn’t exist.

The internet was never built for kids. As time passes, this is gradually getting rectified but the solution is not as simple as bifurcating into ‘features off’ and ‘features on’. It has taken deep investment into technology by companies like SuperAwesome, Epic Games, Roblox, KID, Yoti and others to get there. We should absolutely criticise the deficiencies but it’s important to recognise the work that’s reducing the overall size of the dumpster fire.

I might write something longer over the holidays about this.

I tried to pay attention/But attention paid me (Lil’ Wayne)

‘Champagne pops/on Monday in case Friday flops’ (Kano)

I have some complicated views on bubbles and you probably do too. I’ve noticed that a lot of people confuse the Abundance agenda with being pro-bubble, suggesting either they haven’t read enough about the former or they haven’t experienced the downsides of the latter. A lot of the ‘bubbles-are-good’ conversation (even Byrne Hobart’s inflection-driven thesis) seems to come with a very hand-wavy treatment of who actually loses money when bubbles burst or deflate. It increasingly reminds me of that business write-off scene in Schitt’s Creek.

Look, like any good capitalist, I want my investments to go up. My mental model of capitalism is essentially a series of bubbles (or at least local pricing maximums), with varying degrees of volatility. And while a given investment may end up being long-term or short-term in duration, it still exists across this plane of (pricing) volatility. While you can make money on both types of investment, it does help to have a robust view about which category it falls into: NFTs or ETFs.

All that being said, when it comes to AI, I do have some sympathy with the cohort of AI leaders who are pre-calling ‘bubble’ in order to compress the Gartner hype cycle and get on with the business of building. On the other hand I am also generally of the view that anyone in an asset class who is decrying that same asset class for being in a bubble harbors some existential competition fears.

I have been spending a LOT of time diligencing bootstrapped training, consulting and implementation companies who are on the ground helping businesses adopt AI solutions. The last decade or so of enterprise SaaS has tended to focus founders on solving reasonably point-shaped problems with software but I’m pretty certain that AI adoption for companies of just about all sizes is going to be the opposite of that: a lot of Forward Deployed Empaths and customisation (on the path to agenticism).

In fact, my mental model of AI enablement/deployment is beginning to resemble ERP projects. Agentic solutions founders may scoff but I think it’s highly likely that for every $1 of enterprise AI revenue generated, about 50% of this will need to be spent on extremely unsexy categories like training, consulting and implementation (50% might seem high but a) ERP tends to have ~35% for the same non-software spend and b) we constantly ask founders this question and that’s the current consensus).

Everyone keeps talking about how the most important AI companies are the foundational models, data centres and energy providers. I’ve reached the point where I think that everyone is wrong about this: actually the most important AI companies in the world right now are the ones working on the entirely unglamourous job of enablement.

This has made me no less bullish on the long-term impact of AI but it has absolutely reinforced my view that a meaningful adoption cycle (by business) will look more like ten years rather than two. Plan your bubble investing accordingly.

Other reading

If you’re interested in some of this post, the always excellent Bryce Elder touched on a lot of these points ostensibly via the lens of Rightmove’s AI-investment-related profit warning recently, I highly recommend it. And you should also read this excellent piece which Jerry Neumann wrote about the challenges of investing in AI (he and I seem to share the distinction between investing versus investing-to-make-money).

Recently there has been some quite good writing (and speaking) on fixing government entropy. Although I do wonder whether energy is better expended towards private cities rather than improved governments. Time will tell. In the meantime I recommend Matt Clifford on making the UK rich again and John Collison on how to make Ireland stop going backwards

Sort of related to the above, fellow Power Broker nerds will be extremely interested in this excellent long piece by Jim Waterson on the Mosesification of London which nearly happened.

Some of you will probably be mean and say something like ‘America focuses on capital and growth, China focuses on growth and rare earth minerals and Europe rearranges its boards’. This kind of criticism probably doesn’t understand the friction to growth that bad boards can produce. Bad boards are very, very bad for economies. I have been on them, I know. Europe needs better boards, there is a fairly direct line of sight to improved economic growth.

It’s also worth pointing out that at some level of early stage growth, the best board is almost certainly no board. Don’t waste time with corporate governance until you have a defined corporate thing to govern. I’m always surprised by the number of founders who ping me to ask about being chairman (not even NED!) and they haven’t even figured out who their customer is. Stop wasting your own time. Just build. Then add some (useful) structure.

The point of this post is not just to improve boards (everywhere), it’s because I’m genuinely concerned about some of the things I see slipping through. Very few people reading this will have paid attention to the First Brands bankruptcy but you should-I’m willing to bet several high quality lunches that we will see some kind of similar off-balance sheet debacle across the startup space (most likely data centre) over the next few years.

A role which the Starboard principles doesn’t suggest boards should do but which I am absolutely adamant about is that directors should understand, and indeed interrogate, the revenue recognition policies of their companies. I have written [in other posts] about the widespread ARR fraud which is becoming endemic in founding and investing circles. Jason Lemkin won yet another place in my heart for making the (historically extremely heretical) statement that founders should go to jail for misrepresenting revenue:

This may sound a bit crazy at first — but I wish a few more founders would go to jail. What do I mean? Look when you invest, you take risk. That’s part of the game. Some work out, some don’t. It’s OK in and of itself. You have a loss ratio to accomodate that. But what’s picked up the past few years is fraud from founders. That, combined with highly accelerated investing paces — hot deals are often now done just on a Saturday — gives investors almost no time to do proper diligence. And that’s OK, as long as everyone is honest. But making up ARR numbers? Claiming pilots are annual contracts? Accelerating contracts and ‘recognizing’ all the revenue in month one to raise capital? Claiming customers that disappear weeks after a round closed? If founders though they might go to jail if they did this — everyone would slow down with the B.S. numbers and metrics just a little. The best ones would all get funded, it doesn’t matter. But maybe, folks would slow down before they just basically make things up. That would benefit the entire ecosystem. Every VC I know has a $20m-$100m+ loss coming in their portfolio from one of these deals. Often, several.

There is a long and interesting discussion to be had about the spectrum of optimistic thinking which goes from leadership (‘we landed another customer-amazing!’) to selling to investors (‘we landed another customer-they might buy more!’) to outright lying (‘we landed a customer and they are now in a multi-year contract thus this is ARR!’). At this stage of the movie, I would expect any startup with external investors to have diligenced a sample of the company’s contracts and any startup with external directors to have interrogated the company’s revenue recognition policy. I suspect I am being optimistic.

“Main ingredients to this dish/Goes like this/Here’s the list/Materialistic, greed and lust, jealousy, envious” -D’Angelo

Other reading

Modern Times Group (where I am on the board) ran an excellent (imho) Capital Markets Day last week to lay out the updated strategy for the company. Although this is a public markets event, I actually think it’s quite a good exercise for many private companies of a certain scale to go through for their shareholders.

I’m very bullish on Discord as a platform and our holding company has been investing here for a while now. Unsurprisingly, as the ecosystem grows in importance for brands, developers and everyone else, Discord architects are becoming a thing. I kind of told you so.

For Secret New Thing reasons, I am deeply immersed in Jeffrey Yost’s comprehensive history of IT services from the 1950s until reasonably recently. It’s remarkable how similar the patterns are with different waves of computing.

This is a fairly normal post-exit motion for me as I search for an interesting problem to take a run at. The criteria for ‘interesting’ has evolved over time but broadly speaking it needs to be a) much larger in the future than it is today while b) also holding the probability of credible unit economics. It’s surprising how much that last filter reduces list sizes.

We’ve found something which we feel quite strongly corresponds to this. I promise I’m writing a whole other post to talk about what exactly that is (and indeed who ‘we’ are) but suffice it to say the challenge of last-mile AI adoption creates the scope to build some extremely interesting companies. Both the product-building and capital-delivering functions of AI have largely been crafted with a technology-first orientation yet it seems increasingly clear that AI adoption needs a much more human-centric approach. Given the complexity of how even small companies work, maybe the world also needs Forward-Deployed Empaths.

In many respects this is an extension of our learnings from LFG Holdings (which will continue to exist and invest) about leveraging our weirdness of being a little bit PE and a little bit VC. The new vehicle is called 10xHumans and we’re hiring for a couple of very specific roles, which you are absolutely welcome to speculate on.

Btw, I suspect the frequency of these posts will drop a little but they certainly won’t stop, the feedback and replies I get from writing is informative, occasionally funny and always deeply appreciated.

Reading

Speaking of AI, I was reading OpenAI’s recently released paper on how people use ChatGPT and I was stunned to discover that there are twice the number of people using it for recipes than games and role-playing. Where are my DMs at?

Also speaking of AI, something that resonated deeply was Jerry Neuman (renowned and recently retired investor) writing on how most investors won’t make money from AI. Things can be both extremely successful and also uninvestable.

Ahead of a rumoured IPO, Discord is quietly starting to open to more brand and ad engagement. I think there is a massive upside case here.

A very good interview with Russell Napier about the slow unwinding of the EU

I’ve just finished reading Dan Wang’s excellent Breakneck which came a couple of books after I finished Patrick McGee’s equally excellent Apple in China. I recommend both to massively improve the context of understanding China’s relationship to the world right now. For maximum enjoyment, read in the opposite order to me.

I’d always thought of Roblox as copying YouTube’s ad model (from hands-on to format) but maybe that’s too simplistic. YouTube is now sort of getting into brand engagement (like Roblox).

Factoring in Capex, investment and M&A since 2017 (but not including the equity upside from the public stocks), you can get to a number of ~$2 trillion which has been spent on AI. If you squeezed that into a single year, it would be roughly 0.5% of world GDP. Maybe just about enough to call it the ‘Altman Put’ effect?

I occasionally catch myself wondering how the world would feel if this AI wave hadn’t happened. I suspect that without the Mag-7 growth (outside of which the S&P 500 is basically flat over the last three years), we’d be making a lot more comparisons with the 1970s. Although to be clear there would still be plenty of structural change happening around you:

• Attention saturation is resetting media models

• Tariffs are upturning international trade models

• Population demographics are rewriting government spending models

• Drones are reshaping military models

• Climate change is rebooting our definitions of long-term

• Etc.

One of the things I love about changing environments is that it gives you a rare opportunity to evaluate management in sectors which rarely change this much, this quickly. VC is a great example. Normally you need about twenty years to know whether VC partners are any good (my simplistic distinction between investing and investing-to-make-money).

I’ve had a lot of quiet conversations with VC partners to kick around ideas of what their firms need to look like in the future. Change is just beginning here. For example, I’ve been fascinated to see how some firms are reconfiguring more like PE firms. General Catalyst, Thrive Ventures, Asymmetric and Elad Gil are now actively investing into vertical AI roll-ups (think Constellation Software but with AI). Unthinkable a few years ago.

In fact, AI is an interesting case in point where one of the biggest opportunities will be in the least technical (and thus least VC-suited) area: consulting and training. There is a broad sense that some combination of agents, co-pilots and humans will allow businesses to ‘adopt AI’. But I think almost every technology company is radically underestimating just how human companies can be. That’s also a challenge for VC models.

Other things

I would like to speak to some high energy BD/sales pros who have come from the consulting and/or corporate training world for a secret AI project. Please DM me if that’s you or if you’ve had a great experience with someone.

I’m becoming increasingly obsessed with ancient tattoo analysis and feel there might be a Scroll Prize type of opportunity here to unlock some very cool things. I would love to talk to more people about this.

Kate O’Loughlin asks who will write the AI playbook for kids?