Disclaimer 1: I am not an economist, this is my best-effort attempt to work some of this stuff out. Hopefully it at least serves as a starting point for further conversation.
Disclaimer 2: I would expect that factoring in ethical considerations when investing will lead to lower returns than is otherwise possible.
Disclaimer 3: This is not financial advice.
The dilemma
Over the past few years I’ve found myself in the following rather frustrating predicament:
I believe the hype around AI and expect significant growth for frontier labs and the hardware companies that supply them. At the same time I believe that a slower pace of AI development is more likely to lead to a beneficial outcome for humanity compared to rushing full speed ahead.
As such, despite having strong conviction that the stock price of say NVIDIA was going to rocket upwards in value, I’ve sat out on the sidelines.
In case it’s not immediately clear how an individual’s investment in say NVIDIA can help to marginally speed up AI development, let me briefly outline the mechanisms involved.
By purchasing the stock of a company, you apply upwards pressure to its stock price. A higher stock price allows the company to do the following things:
Raise money more easily, at more favourable rates or with less stock dilution
Acquire other companies more cheaply (using their own stock as currency)
Compensate employees and attract talent without draining cash reserves
I appreciate that any one person’s stock purchase is a drop in the ocean (excluding certain wealthy individuals), but nonetheless I like knowing that my own personal investments marginally move the needle in the right direction. Who knows, maybe if enough people did the same thing the impact would be perceptible.
So what investment options do I have?
So am I doomed to sit on the sidelines whilst everyone else has fun?
Or are there investment options that both:
1) Perform well in a timeline where AI development continues to progress and deliver value
2) Don’t increase the speed of said AI development (or even better reduce it)
I’m not going to go into much detail on point 1) I’ll leave that up to the reader to decide whether assets are likely to appreciate or depreciate.
However, what I will try to do is cover point 2) i.e. consider a bunch of common asset classes and give my thoughts around whether an investment there would marginally speed up/slow down/have no impact on the pace of AI development.
How about stocks?
Looking first at ETFs.
If you take the Vanguard S&P 500 ETF (VOO) as an example, nearly 30% of it is made up of frontier labs and adjacent hardware/infra companies.
For my liking that feels like too much money being allocated to companies investing in the AI infrastructure build out.
How about an ETF focused on small caps? For example Vanguard Small-Cap ETF (VB)
There is stuff in there that is pushing AI forwards, e.g. hardware, energy, utilities, resource extraction etc. But at least there isn’t as much going directly to frontier labs. So this at least seems like an improvement on a large cap focused ETF.
How about non-US ETFs? e.g. FTSE Developed Europe UCITS ETF (VEUR)
This European example has a smaller percentage of companies with an AI, hardware and infrastructure focus, compared to US broadmarket funds. Even if ASML does make up nearly 4% of the allocation.
I won’t go evaluating funds ad infinitum, however I think the general observation that can be made is that non-US ETFs are probably preferable from an AI risk perspective. Also, even within the US, small cap ETFs may be preferable to large cap ETFs, considering the concentration of AI risk within the Mag 7.
Of course by picking individual stocks or even sectors, I could more consciously avoid contributing to the AI infra build out. i.e. favouring industries such as:
Pharmaceuticals
Agriculture
Consumer Tech
Media & Entertainment
Over industries such as:
Frontier AI
Computing hardware
Resource extraction
Energy
How about bonds?
Public sector investment in AI makes up a relatively small amount of total government expenditure, so the percentage of money from a government bond purchase that goes on to fuel AI development is relatively low. US government bonds are probably more AI fuelling than others though given the US in particular is on the AI frontier.
Some corporate bonds are very much fuelling the AI infrastructure build out. Like stocks, the extent of this is going to vary across industry verticals and across individual companies (with some hyperscalers issuing bonds solely for the purpose of funding this build out).
How about commodities—e.g. precious metals, rare earths?
Purchase of certain commodities with inelastic supply could actually increase costs for companies building out AI infrastructure, and thus could marginally slow AI development.
For maximum impact in terms of raising costs of the AI infra build out, the purchase would need to be of the physical commodity (or at least of a physically backed ETF). It’s unclear to me what the effect of purchasing via a synthetic ETF would be.
Note, purchasing stock of companies involved in the resource extraction itself (e.g. mining and refining) would have the opposite effect and reduce the costs of the AI infra build out.
Are there any particularly supply side inelastic commodities required for the AI infra build out?
Firstly there are various ‘by-product’ metals that are not the main base metal target of the mines that produce them. As such the supply of these by-product metals is more dependent on the price of the base metal being mined.
Some example by-product metals required for the AI infra build out include (non-exhaustive): silver, germanium, gallium, indium and ruthenium. Purchasing and hoarding these metals would likely marginally increase the cost of the AI infra build out.
Then there are base metals like gold, copper and tin. As they are primary targets of mining operations, their supply is more elastic than by-product metals but they are still considered relatively inelastic (given it takes time to explore and mine additional reserves).
Whether buying and hoarding would be net negative or net positive probably varies. I hypothesise that gold purchases might be net negative as it probably isn’t a primary AI bottleneck, but increasing its price could increase the supply of byproduct metals that might be more constrained. Of the three, copper feels like the bigger AI infra buildout bottleneck, so it feels like a purchase might still increase the cost of AI infra build out even if it increases the supply of by-product metals.
If anyone has a firmer idea of how to model the various interactions here and their relative importance I would love to hear it!
How about real estate and land?
I would expect residential real estate purchases to have no impact on AI development timelines.
That’s not always the case for agricultural or commercial real estate though (if the land is suitable for data centre development). For example purchasing farmland that otherwise could become a data centre would marginally raise the cost of the data centre build out.
How about crypto?
Bitcoin with its proof of work consensus mechanism directly competes with the AI build out in terms of various resources:
Data centre real estate (particularly in low energy cost locations)
Energy
Hardware
The greater the price of Bitcoin, the more hardware and energy will be dedicated to performing hash calculations aimed at satisfying the proof of work for the next Bitcoin block. At least in the short term, a higher Bitcoin price should raise costs for the AI infra build out.
In the longer term it’s possible that the additional demand for hardware from the crypto industry could lead to greater profits for hardware companies and therefore greater investment in further production capacity. Whether net negative or net positive therefore probably depends on takeoff timelines.
For other non-proof of work crypto (e.g. Ethereum) which are less energy and compute intensive, I can’t see as strong a link between the price of the cryptocurrency and the cost of AI infra buildout.
Assumptions
All of the above assumes that a slower pace of AI development is likely safer. It also assumes that AI progress is likely to continue to be heavily influenced by compute and infrastructure scaling. I’m aware there are alternative scenarios that could play out, but I’m reasoning based on what I believe to be the most likely scenario.
Investing in light of AI risk
Disclaimer 1: I am not an economist, this is my best-effort attempt to work some of this stuff out. Hopefully it at least serves as a starting point for further conversation.
Disclaimer 2: I would expect that factoring in ethical considerations when investing will lead to lower returns than is otherwise possible.
Disclaimer 3: This is not financial advice.
The dilemma
Over the past few years I’ve found myself in the following rather frustrating predicament:
I believe the hype around AI and expect significant growth for frontier labs and the hardware companies that supply them. At the same time I believe that a slower pace of AI development is more likely to lead to a beneficial outcome for humanity compared to rushing full speed ahead.
As such, despite having strong conviction that the stock price of say NVIDIA was going to rocket upwards in value, I’ve sat out on the sidelines.
In case it’s not immediately clear how an individual’s investment in say NVIDIA can help to marginally speed up AI development, let me briefly outline the mechanisms involved.
By purchasing the stock of a company, you apply upwards pressure to its stock price. A higher stock price allows the company to do the following things:
Raise money more easily, at more favourable rates or with less stock dilution
Acquire other companies more cheaply (using their own stock as currency)
Compensate employees and attract talent without draining cash reserves
I appreciate that any one person’s stock purchase is a drop in the ocean (excluding certain wealthy individuals), but nonetheless I like knowing that my own personal investments marginally move the needle in the right direction. Who knows, maybe if enough people did the same thing the impact would be perceptible.
So what investment options do I have?
So am I doomed to sit on the sidelines whilst everyone else has fun?
Or are there investment options that both:
1) Perform well in a timeline where AI development continues to progress and deliver value
2) Don’t increase the speed of said AI development (or even better reduce it)
I’m not going to go into much detail on point 1) I’ll leave that up to the reader to decide whether assets are likely to appreciate or depreciate.
However, what I will try to do is cover point 2) i.e. consider a bunch of common asset classes and give my thoughts around whether an investment there would marginally speed up/slow down/have no impact on the pace of AI development.
How about stocks?
Looking first at ETFs.
If you take the Vanguard S&P 500 ETF (VOO) as an example, nearly 30% of it is made up of frontier labs and adjacent hardware/infra companies.
For my liking that feels like too much money being allocated to companies investing in the AI infrastructure build out.
How about an ETF focused on small caps? For example Vanguard Small-Cap ETF (VB)
There is stuff in there that is pushing AI forwards, e.g. hardware, energy, utilities, resource extraction etc. But at least there isn’t as much going directly to frontier labs. So this at least seems like an improvement on a large cap focused ETF.
How about non-US ETFs? e.g. FTSE Developed Europe UCITS ETF (VEUR)
This European example has a smaller percentage of companies with an AI, hardware and infrastructure focus, compared to US broadmarket funds. Even if ASML does make up nearly 4% of the allocation.
I won’t go evaluating funds ad infinitum, however I think the general observation that can be made is that non-US ETFs are probably preferable from an AI risk perspective. Also, even within the US, small cap ETFs may be preferable to large cap ETFs, considering the concentration of AI risk within the Mag 7.
Of course by picking individual stocks or even sectors, I could more consciously avoid contributing to the AI infra build out. i.e. favouring industries such as:
Pharmaceuticals
Agriculture
Consumer Tech
Media & Entertainment
Over industries such as:
Frontier AI
Computing hardware
Resource extraction
Energy
How about bonds?
Public sector investment in AI makes up a relatively small amount of total government expenditure, so the percentage of money from a government bond purchase that goes on to fuel AI development is relatively low. US government bonds are probably more AI fuelling than others though given the US in particular is on the AI frontier.
Some corporate bonds are very much fuelling the AI infrastructure build out. Like stocks, the extent of this is going to vary across industry verticals and across individual companies (with some hyperscalers issuing bonds solely for the purpose of funding this build out).
How about commodities—e.g. precious metals, rare earths?
Purchase of certain commodities with inelastic supply could actually increase costs for companies building out AI infrastructure, and thus could marginally slow AI development.
For maximum impact in terms of raising costs of the AI infra build out, the purchase would need to be of the physical commodity (or at least of a physically backed ETF). It’s unclear to me what the effect of purchasing via a synthetic ETF would be.
Note, purchasing stock of companies involved in the resource extraction itself (e.g. mining and refining) would have the opposite effect and reduce the costs of the AI infra build out.
Are there any particularly supply side inelastic commodities required for the AI infra build out?
Firstly there are various ‘by-product’ metals that are not the main base metal target of the mines that produce them. As such the supply of these by-product metals is more dependent on the price of the base metal being mined.
Some example by-product metals required for the AI infra build out include (non-exhaustive): silver, germanium, gallium, indium and ruthenium. Purchasing and hoarding these metals would likely marginally increase the cost of the AI infra build out.
Then there are base metals like gold, copper and tin. As they are primary targets of mining operations, their supply is more elastic than by-product metals but they are still considered relatively inelastic (given it takes time to explore and mine additional reserves).
Whether buying and hoarding would be net negative or net positive probably varies. I hypothesise that gold purchases might be net negative as it probably isn’t a primary AI bottleneck, but increasing its price could increase the supply of byproduct metals that might be more constrained. Of the three, copper feels like the bigger AI infra buildout bottleneck, so it feels like a purchase might still increase the cost of AI infra build out even if it increases the supply of by-product metals.
If anyone has a firmer idea of how to model the various interactions here and their relative importance I would love to hear it!
How about real estate and land?
I would expect residential real estate purchases to have no impact on AI development timelines.
That’s not always the case for agricultural or commercial real estate though (if the land is suitable for data centre development). For example purchasing farmland that otherwise could become a data centre would marginally raise the cost of the data centre build out.
How about crypto?
Bitcoin with its proof of work consensus mechanism directly competes with the AI build out in terms of various resources:
Data centre real estate (particularly in low energy cost locations)
Energy
Hardware
The greater the price of Bitcoin, the more hardware and energy will be dedicated to performing hash calculations aimed at satisfying the proof of work for the next Bitcoin block. At least in the short term, a higher Bitcoin price should raise costs for the AI infra build out.
In the longer term it’s possible that the additional demand for hardware from the crypto industry could lead to greater profits for hardware companies and therefore greater investment in further production capacity. Whether net negative or net positive therefore probably depends on takeoff timelines.
For other non-proof of work crypto (e.g. Ethereum) which are less energy and compute intensive, I can’t see as strong a link between the price of the cryptocurrency and the cost of AI infra buildout.
Assumptions
All of the above assumes that a slower pace of AI development is likely safer. It also assumes that AI progress is likely to continue to be heavily influenced by compute and infrastructure scaling. I’m aware there are alternative scenarios that could play out, but I’m reasoning based on what I believe to be the most likely scenario.