The confusing AI signals continue

I’ve been living in Europe and doing business in the US for the last two decades. I was in SF the other week and, despite going there pretty frequently, was surprised by the energy I met. While ‘European goes to west coast and thinks there might be something to it after all’ is hardly original, it’s certainly nice to see and feel something a little less dystopian than what’s commonly reported on X (primarily about downtown SF). Entrepreneur First is now bringing its cohorts to SF and founders like Eoghan McCabe are sponsoring younger founders to spend time in the ecosystem over the Summer. 

Ultimately, the primary source of this energy is coming from VC investing into AI. While LLM breakthroughs haven’t yet led to a cohort of exits for pre-2021 companies, as an ecosystem QE strategy¹, it seems to be quite effective: FOMO and inference are powerful motivators for both founders and investors. This isn’t a snarky take, I plan to spend more time out there (rebalancing from time in LA). 

I’ve been mulling a few AI-related things lately. I’m not going to speculate on the technical roadmaps (I recommend Corinne Riley’s post instead) but other bookmarked thoughts include:

The speed of incumbent-to-participant in this cycle has been amazingly fast. It’s remarkable to see Microsoft, Meta, Google and Amazon controlling such huge swathes of territory at what still seems to be a very early stage. Obviously this is driven by capital requirements but perhaps a more subtle analysis is the age and specific experiences of leaders who are now running these companies. It’s an operator’s bias but I think investors significantly underestimate the impact of individual leaders versus the actual business.

Investing into this space is hard. ‘Don’t invest in the application layer’ says nearly everyone (while I see application layer companies raise money every week). With more focus on enterprise functionality, every foundational LLM developer day now comes with a bodycount of startups and I don’t see this slowing down any time soon. 

I get asked a lot by investors whether I think whether generative AI will replace game engines. It’s the wrong question. Yes absolutely, game engines (and development workflow) in their current paradigm will get replaced over time by things broadly labelled as AI. But that doesn’t mean the companies providing them will change. I think there has been a lot of incorrect analysis about the AI impact on existing industries. Distribution (in the enterprise sense) remains the most valuable asset. If you have partnered or outsourced to a particular company at some level of scale, it’s unlikely you are going to allocate new resources to AI experimentation with the aim of in-sourcing it back unless the savings are at least one (and maybe two) orders of magnitude. I just don’t think we’re anywhere near that. A case in point has been Keywords Studios. Over the last ten years they’ve acquired ~60 companies to become the go-to outsourcing partner for video games publishers, providing QA, localisation, content development and general live ops support. Their share price had been demolished over the last eighteen months as general speculation (and short-selling) mounted that AI development was going to destroy (or by share price interpretation, cut in half) their business. I have long thought this nonsense. The most rational future for many industries is that companies like Keywords will become the aggregating layer for AI tools, productising, professionalising and generally wrapping them in a useful workflow which can deliver additional value and functionality to their existing client base. EQT’s recent bid at 60% above the then share price suggests a similar view on the future. 

Speaking of M&A, who is going to buy the many venture-backed companies which are scaling in the space? It’s a fascinating question. As I mentioned, this cycle is unusual for the existing incumbent presence in the disruption wave. The churn rates for many gen AI tools look horrendous. Enterprise revenue durability looks questionable and if you accept that LLMs are fundamentally a deflationary force then it probably gets hard to argue increasing TAM dollar value. That’s a pessimistic list but really it’s just a forcing function to distinguish between value and market pricing. Gil Dibner nails this:

‘We are witnessing the "power of AI" - not just to drive huge economic outcomes for investors but to divorce investors from economic reality. Too many investors are still addicted to the sugar rush of irrational exuberance - and that high is only available today in AI.’

Re-reading this post, it sounds pretty bearish on AI investing. I think ‘cautious’ is more accurate.

“I'm in it for the long run like a marathon” Skrapz

---

Reading/listening/pondering

Listening to Razib Khan’s excellent (free) genetic history of the Indo-Europeans. DNA analysis is really the closest thing we have to ‘truth’ in history. It’s a bit technical in places but highly recommended 

I’ve been obsessing about the impact of changing demographics for a while so it’s gratifying to see the Economist make fertility rates their cover article this week. Similarly, this deep-dive by the FT into trying to convince Gen Z to watch sports is worth your time (and it’s not like Gen Alpha will be any easier). I fully believe that we’ll see highlight rights become the most expensive sports licensing component within the next decade. Bet against me?

Anime remains huge. A fascinating history piece into how Ghost in the Shell became a hit.

A vertical-ad-network incubator is a pretty interesting idea. 

Companies which are really banks in disguise.

1. Kind of the equivalent of Web3’s ‘Chris Dixon put’