That is the topic of my latest Bloomberg column, here is one excerpt:
Posting a query to ChatGPT consumes a lot of energy, by one estimate 10 times more than a Google search. Currently large language models are sufficiently limited that this is not a major factor in aggregate energy consumption. But as use of AI services increases, the energy burden will rise. Countries with expensive energy, or which will not allow energy consumption to rise much for climate or regulatory reasons, will look to import their AI services from energy-rich countries.
In the future, energy-rich regions may include Spain and Morocco with solar power, South Korea with affordable nuclear power, and whichever nations are pioneers in nuclear fusion. Those nations may end up as major exporters of AI-generated data. They might draw their AI inputs from the US, but specialize in cheap calculation and information transmission. And some regions of America may join this list as well, especially if they are well-suited for solar and hydroelectric power.
To be clear, the US will export a lot of AI services, through such companies as OpenAI, Google, Meta and Anthropic. But the US is not as good at building affordable infrastructure, and that will put it at a disadvantage in the AI revolution and distribute many of the gains abroad.
It remains to be seen whether there are higher profits in selling the original source code or the more derivative electricity-driven, infrastructure-based AI calculations. Nonetheless, this is a potential economic and national security risk for the US. It could end up with a strong lead in the source product, but fall badly behind in making (“manufacturing,” you could say) the final AI outputs.
Recommended.
The post How will AI remake the rules of international trade? appeared first on Marginal REVOLUTION.
Comments
Post a Comment