Monopoly or Rice Market - The existential question facing GenAI

Will it a be winner takes all market or like selling rice? Not a riddle but the core question raised by famous VC investor Marc Andreessen during last week’s Ray Summit. One potential future he outlined for Artificial Intelligence is that of a “monopoly and infinite profits because of scale” where the biggest company will have the best model that all wish to use and it can thus charge for at its own discretion. He compared this with what happened in search with Google. The other future is “in a race to the bottom, in which it actually turns out selling intelligence is selling rice“; high competition, little differentiation and slim margins.

The question thus is, which future is in store for the current leader in the field, Open AI? It’s new Chief Product Officer Kevin Weil outlined later at the same conference that he sees his company on a winning path. OpenAI effectively puts its hopes in geometric progression i.e. that its initial advantages will compound and keep it further and further ahead. Speaking in references to their new o1 model, the first to push past the reasoning frontier (Link), Kevin said “by the time people do catch up, we're going to try and be three more steps ahead.”

Given what is at stake, it’s worth looking deeper at the key factors to consider.

?

Resources: Is money all it takes?

It is unlikely that any model provider, even OpenAI can create a moat by denying other’s resources, notwithstanding the scarcity of cutting edge GPUs and top-notch GenAI research talents. Nevertheless, OpenAI has a clear financial advantage. Following its $157 billion valuation latest funding round, coupled with a credit line, it is now sitting on $10.6 billion in cash. An unheard of amount that no other GenAI startup can match. This will further fuel its already enormous burn/investment rate compounding its head-start (OpenAI is expected to see a loss of $5B in 2024 Link). The fact that more and more GenAI startups are quitting the race (character ai being the most recent example) is testament to the fact that keeping up is extremely costly.

Consolidation does however, not automatically mean we are headed to a monopoly stage. It is a common fact we seen in many startup domains. Moreover, while few startups can match OpenAI’s level of funding/investment, there are not only startups in this market. OpenAI’s competitors also include some of the biggest names in Big Tech, from Meta (whose recently announced video generator is now leading the field) to NVIDIA who just released a massive open-source model called NVLM 1.0 that rivals GPT-4o.

Flywheel effect: Is there one in GenAI and can it create a defensible moat?

In previous eras where new tech platforms came to be, their outsized success was in large part due to the first-mover gaining a consistently growing advantage by learning from user interactions. With every click by users, Google becomes better, thus we use it more, thus it becomes better, ad infinitum. However, given the pre-trained nature of Foundation Models user prompts can provide very limited value for further model training, if any.

Even if the learning flywheel is thus unlikely, we should discount network effects that can flywheel just as much. Unlike in the early days of, for example Facebook, it will not be the amount of users that improve the attractiveness of platforms like OpenAI though. Rather it will be the developers of GenAI solutions attracted to the scale of marketplace that OpenAI offers. If this AppStore like effect continues and compounds, it would point towards a more winner takes all future. Notwithstanding its enormous available funding, splitting focus between building a product/model and a platform/marketplace is a big challenge even for OpenAI. Thus it is not a certainty this effect will materialize.?

Application layer vs model layer: Where will the actual winners be minted?

This will be the most important determinant of whether we see a winner takes all or rice market: Will it will be the providers of models or creators of GenAI applications that generate the larger revenues in the long-run? Mark’s own VC a16z has been very vocal that it sees the application layer outgrowing the model one eventually (Link).

For now, GenAI startups outperform previous waves of tech transformation startups in terms of revenue growth (Link), mainly driven by model providers for now. However, seeing the consistently downwards price trajectory for Foundation Models, we are in fact heading towards a future where intelligence will be virtually free. LLMs have for that reason been often referred to as the fastest depreciating asset of all time. Even OpenAI acknowledged this fact when it launched its GPT-4o mini. If this trends continues, which is likely, we are more probably headed towards a rice market. Unless one or several players will see the abovementioned AppStore effect kick in.

But... Beware of Black Swans?

The debate is open and will be ongoing. Making it even more complex, we cannot discount potential impact of unknown unknowns. For instance, we cannot yet estimate the impact of regulatory efforts that are ongoing globally. Also there are some, most prominently Meta’s Yann LeCun, who argue that GenAI is not actually on the critical path towards Artificial General Intelligence aka AGI (Link). This suggests there might be another yet unknown pivotal tech moment ahead of us that would reset the race.

As complex as the discussion is, it is essential to have. Which vision of the AI future will materialize has tremendous implications for people, countries and companies.