Comment by jsnell

Information retrieval for consumers is an existing high margin $250B/year business at a minimum (search ads revenue). That market is being disrupted before our eyes. That's 20x right there. There's going to be fierce competition and OpenAI is by no means guaranteed it, but surely they have more than just a puncher's chance.

The ad-monetized consumer market is funny in that they tend to be winner take all. Nobody can compete on price, because the ads go where the users are, and the users don't pay. And preferences are sticky, after making a choice, the users don't switch just for incremental improvements. So

The software development industry is likewise well in the process of being disrupted. LLMs for programming market seems to have grown from nothing to >$10B in a year. And the software development market is hundreds of billions / year, if you just consider the employment costs. We don't know exactly how this will play out, but again there's at least an order of magnitude more growth available there.

The above two are just the places where the impact is already obvious and it's clear that we don't need to assume any additional increase in capabilities. But an increase in capabilities seems really likely. Even if it turns out that we're right now on the cross-over of the sigmoid, and the asymptote won't be ASI or even AGI, a large proportion of knowledge work is also at risk. And then the addressable market is trillions if not tens of trillions.

I know this is hand-wavy, apologies for that. Doing this kind of analysis properly is both hard work and would require specialized data sources. But I think that's the general intuition for why high valuations for frontier labs are justifiable (and the same justification for bigtech capex). It's a lottery ticket with good odds for redistributing existing markets, as well as another set of lottery tickets for some set of probable markets, though we don't know exactly which.

With AGI or ASI, the addressable market is all economic activity, and at that point basically anything is justifiable.