The previous post looked at how you might invest for a scenario where AI can do most white-collar work (AGI, roughly speaking), without being broadly superhuman (ASI). However, a short1 transition from AGI to ASI seems plausible, even likely under certain conditions. My focus on the simpler AGI scenario is partly a case of looking for the keys under the streetlight. So in this post I’d like to think through how to invest for a full ASI scenario.
Firstly, I will assume the ASI is aligned. If an unaligned AI causes human extinction, investment performance is irrelevant. Similarly, an omnibenevolent AI that distributes its surplus to all humanity would also make this analysis moot. So let’s focus on a middle scenario where property rights and market mechanisms remain in effect across the transition.
ASI would be equivalent to the sudden arrival of tens of millions of above-genius-level scientists, coders, administrators, dealmakers, etc, working day and night, with a raw thinking speed that could be 10x-100x faster than human. They would make very rapid technical progress, including improving themselves. Decades or centuries of progress could happen in years, perhaps even in days in some fields. Many people have a hard time taking this seriously, in the same way Covid was hard to take seriously in January 2020, but it seems quite plausible to those paying close attention, myself included. It is not like the world has never changed like this before. Precedents include the evolution of photosynthesis, the rise of homo sapiens, and AlphaZero rediscovering all human chess knowledge in an afternoon.
ASI would be king of information space, but might initially require human assistance to get things done in the physical world. There would be an interesting transition period, until it can build and control a fleet of robots to act for it. At that point, it could also make rapid physical progress (assuming it’s allowed to).
Does it even make sense to talk about an investment strategy for a scenario like this? Carpe diem in the meantime? I think it’s possible to say something.
An ASI would probably build a parallel economy, largely from scratch. The new economy would very rapidly outgrow our legacy one, and prices would rapidly shift to reflect things of value to the ASI or its controllers. Human productive capacity, and the fruits of human productive capacity, would see huge (relative) devaluation, in much the same way that humans have little use for the economic output of chimpanzees. For example, you could expect existing software to become nearly worthless.
What current assets might rise in price with this new, rather extreme market?
For a brief transition period there might be very high prices for whatever the AI needs to bootstrap itself. Human physical labour would be more valuable to the AI than human intellectual labour, so you might see wage spikes for physical work. There might be extreme demand for key physical assets such as chip fabs, power plants, factories suitable for robot manufacture, associated supply chains, and maybe private stores of unique data. It’s hard to predict exactly what’s on the list – e.g. what about launch capacity to orbit, or DNA sequencing machines, or pipettes? Any of these could become briefly extremely valuable. In the medium term, they would then sharply devalue as they get superseded by better AI-created replacements.
There have been suggestions that interest rates would spike, on the basis that AI would need to pull in a lot of capital very quickly in order to grow. I don’t see that happening – if AI can self-generate resources of very high value, why would it need to borrow from people? Interest rates also didn’t rise during the Industrial Revolution, which is the closest historical parallel.
What about after the transition period, when the economy is AI dominated? Are there any existing assets that would retain value? Land seems like the obvious possibility. Any kind of land might do reasonably well, and land that’s desirable to AI or to newly mega-rich humans would do particularly well. Other goods that are inherently supply constrained, such as historical artefacts, could also do well.
Finally, what about ownership of the ASI itself? To the extent that companies retain ownership and control of the ASI they create, they could become unimaginably valuable. A winner-take-all dynamic is possible: in the absence of some legal restraint, a lead of even a few months in reaching ASI could rapidly snowball due to the internal growth, and the ability to acquire key resources for future growth ahead of anyone else.
This much power in the hands of a private company seems too much, so perhaps the ASI would be nationalized. In this scenario, citizens of the leading nation may capture the bulk of the value.
Overall, I think the safest positioning is to be diversified2, but liquid enough to react as things become clearer. In a world that is changing very fast, the premium investors demand to hold illiquid assets should be much higher. Existing PE funds would do badly, and perhaps counter-intuitively maybe VC as well. Real estate might be a counter-example.
A diversified portfolio is normally a safe set-and-forget bet, but is it possible to be diversified enough? What if a private company is the first to develop ASI3? Even a very broad portfolio is not going to capture that gain, and to a first approximation it might be the only thing that matters. I think this is a real problem, though the capital requirements of ASI seem large enough that a winner with no connection to public markets is relatively unlikely. Even if that happens, a broad index would probably have enough value in “transition period resources” that it should still do OK-ish.
Boiling this down, here are some specific ideas:
- Diversify broadly, stay liquid and pay attention
- It seems like a particulary bad time to do PE.
- VC, which usually does well in tech transitions, may actually be a bad bet too. The exception would be funds that own direct stakes in potential ASI creators.
- Own a basket of the foundation model companies, including the private ones if you can.
- Most of them have a connection to public markets, i.e. OpenAI-Microsoft, Anthropic-Amazon, etc.
- The value of all other software goes to zero.
- In particular Nvidia, as a fabless semiconductor company, is “software”. It goes to zero.
- The value of some key enabling physical assets could spike, at least temporarily.
- Chip fabs are the obvious bet here4.
- But to do well you may need to sell them at just the right time.
- Buy yourself some property, it will never be so cheap again
- Own real estate generally
- Buy a private museum collection?!
- Own some sources of private valuable data?
- Not sure what fits the bill here.
- If existing money becomes relatively worthless, then being leveraged seems more attractive than usual? I think this is probably correct, but leverage is always dangerous.
- Consider becoming a US citizen
- Be the Irish state and tax the winners :-)
Footnotes
- By short, I mean a 1-5 years. A few people worry about “foom” scenarios, where the transition happens on the order of days, but this seems very unlikely to me. ↩︎
- Though being overweight in companies with prominent AI efforts would be hard to argue against. ↩︎
- The major labs like OpenAI and Anthropic don’t really count here, because public companies own sufficiently large stakes that a substantial part of the value will flow to public markets (at a guess, about 50% in the case of OpenAI, maybe 30% in the case of Anthropic at time of writing). However, this isn’t currently the case for xAI, SSI, Deepseek, etc. For xAI and SSI, there is some connection to public markets via listed companies that own stakes in the VC funds who have backed them, though only a small percentage of the value would reach public markets via this route. Deepseek has no connection to the public markets at all, to my knowledge. ↩︎
- Intel could be the most under-priced current asset in this scenario. ↩︎