David vs GolAIth
It's getting harder and harder for many founders and investors to see long-term opportunities around the edges of the major AI labs.
It’s a brutal reality driven by a few converging factors that all point to a centralisation and concentration in the AI race - at least for now.
The giants (Open AI, Anthropic, Google, Grok) have access to insane amounts of capital and other resources. Their cost of capital, for now, is structurally lower than a “normal” startup, meaning their ability to compete in multiple fields is relatively unconstrained. Even with Open AI reportedly planning to reign in some of its "side quests", ideas abound.
Add to that the media circus around these names and the free PR and distribution every new product release receives, and it becomes an ever tougher nut to crack to cut through the noise with anything else. Sure, there have been some specific breakouts in product-specific use cases of AI (e.g. Granola for notetaking, or Gamma for document creation), but they are rare.
On top of the unfair advantages from a capital and distribution point of view, the centralising forces of memory and context means that more and more of a consumer’s daily life gets sucked into the gravitational pull of these models. Putting aside the friction of switching between apps to complete tasks, the shared context and memory across all my needs is a powerful web of knowledge that the AI models hold on my that a startup can only dream about.
What to do, then, if you are in the seat of a founder building for the AI inevitability in consumer? In which areas is it possible to compete?
I think there are at least four areas to look at:
- Super-verticalised spaces: Use cases that rely on esoteric data and workflows that general purpose models won't bother replicating. In other words, the less 'horizontal' the better, at least for now. I've seen many vertical SaaS models pointing in this direction - build for a super specific customer you believe you can win globally.
- Regulated sectors: Areas where there are actual or perceived regulatory barriers will be far along the roadmap for the labs. In some markets (e.g. healthcare in the US) the size of the prize might be worth the embrace of the complexity, but for many markets particularly outside the US, my bet is they won't bother for a while, allowing a local market specialist to build scale.
- PR hornets' nests: The closer the use case sails to the winds of negative PR, the less attractive it feels for labs focused on dominating the chunkiest segments of the economy, typically with an enterprise-first approach. Adult content (note Open AI's recent pullback), under 18s, therapy, and so on are all contentious and fall into the "not worth the headache" bucket.
- Local spaghetti: Anything that requires integrations with multiple, local, fragmented end-points to deliver a brilliant end-to-end experience for the consumer. Building touchpoints with the 'real world' (think local service providers, small businesses, and so on) doesn't scale easily, and these labs are all about scale. Again here, a light-footed local startup should be able to out-compete a generalised player half-arseing it from abroad.
All of the above are hard to execute and scale, but that's exactly why they look more secure from the threat of becoming a feature on someone else's roadmap. They may also point to smaller markets and outcomes than the AI narrative implies, but a durable winner in a smaller market feels better to me than a temporary breakout that gets swallowed.
Software ain't what it used to be...