I wrote [a post](https://eastwind.substack.com/p/fable-in-shackles) about something I think investors are underappreciating in the AI trade.
Most of the AI discussion the past few years has been about investing in bottlenecks (compute, memory, optics, power, etc.). But the recent Fable / Mythos restrictions made me think the next bottleneck could be something different. I call it intelligence sovereignty (and Jensen calls it sovereign AI).
The basic idea is that hosted frontier intelligence can suddenly be restricted by either the vendor or by the government.
That has a few downstream implications:
* Non-US governments and strategic industries will push harder for local AI stacks, even if the models are somewhat worse
* Open-source models become more strategically important, but may also stop keeping pace with the closed frontier
* More companies will look at self-hosted / private AI deployments, especially in sensitive sectors like cybersecurity, biology, defense, etc.
* This is a boon for inference providers, neoclouds, hyperscalers, and OEMs that can help enterprises run models privately
* It may also weaken the assumption that US frontier labs automatically become the default intelligence layer for the rest of the world
* Countries that don't have frontier intelligence will need to make up for it in inference time compute & manpower (to steer weaker models)
Now, some of the names that benefit from this trade might be the same as the current AI bottleneck trade.
Full piece here: [https://eastwind.substack.com/p/fable-in-shackles](https://eastwind.substack.com/p/fable-in-shackles)
