But another risk is that if there are huge efficiency breakthroughs, either through hardware or software/model improvements. This is what killed the telecoms boom in the early 2000s - it wasn't so much that bandwidth growth stopped, but the efficiency of fibre optics far outpaced the demands.
I'll add that I think this scenario is unlikely. I think we've see that with more compute, we have more capable AI models. The more AI chips you have, the more training you can do, the smarter the AI, the more thinking tokens you can have per task, the faster the tokens/s, the more simulations you can run with CPUs to get the right answer, the bigger the model, etc.
This wasn't nearly as true with internet bandwidth during the dotcom.
I think how the boom can end is that models hit a wall in intelligence/compute but it seems like model capabilities are increasing faster and now slowing down - particularly self improvement is on the horizon.
I think the most likely way end it ends is politics and war. At some point, maybe China will stop all rare earth exports if their models are falling behind too far. All tech products depend on China's rare earth. The other is war. Sometimes the best way to stop your neighbor's progress is to just blow up their data center.
China is not falling further behind, rather is getting closer, and I expect them to eventually take the lead. MiMo recently RL'd for $3.5M, efficiency gains are significant, partly the effect of chip bans.
We aren't going to war with China over data centers and Ai. A new cold war is the best outcome compared to a hot war. Americans are not in the mood for more war.
Good post.
I'll add that I think this scenario is unlikely. I think we've see that with more compute, we have more capable AI models. The more AI chips you have, the more training you can do, the smarter the AI, the more thinking tokens you can have per task, the faster the tokens/s, the more simulations you can run with CPUs to get the right answer, the bigger the model, etc.This wasn't nearly as true with internet bandwidth during the dotcom.
I think how the boom can end is that models hit a wall in intelligence/compute but it seems like model capabilities are increasing faster and now slowing down - particularly self improvement is on the horizon.
I think the most likely way end it ends is politics and war. At some point, maybe China will stop all rare earth exports if their models are falling behind too far. All tech products depend on China's rare earth. The other is war. Sometimes the best way to stop your neighbor's progress is to just blow up their data center.
China is not falling further behind, rather is getting closer, and I expect them to eventually take the lead. MiMo recently RL'd for $3.5M, efficiency gains are significant, partly the effect of chip bans.
We aren't going to war with China over data centers and Ai. A new cold war is the best outcome compared to a hot war. Americans are not in the mood for more war.
I find all of these scenarios dubius or the outcome unlikely is some way.
Group 1 is manageable
Group 2 is Jevon's paradox and what I expect, but capital will still be running it all, they have the chips
I think that was the conclusion of the author as well.
I personally think it's politics and war.
Ai is here to stay, grow, and permeate.