Lots of stuff to just completely love here, but maybe my favorite part is the emphasis on the very simple solution of raising corporate income taxes as a broad-based non-winner-picking method of raising the necessary revenue to fund <strikethrough> deficit reduction </strikethrough> whatever sets of programs we need to manage the AI transition well. I am sort of blown away by how tempted many people are to fundamentally reinvent the wheel here. But regular old wheels work very well for the problem we're confronting — specifically the problem where companies rightly have incentives to find creative ways to lower labor requirements. We should capture, as a society, some of the performance gains they get from doing that!
Geopolitical supply chain tensions in semiconductor manufacturing reinforce the need for diversified chip sourcing and energy-efficient model architectures across enterprise software sectors.
The data center hosting question is essentially chip export controls in reverse, and it runs into the exact same structural problem — leverage built on an ally's trust doesn't survive the moment that trust turns out to be conditional. Anton's point that allies could weaponize hosting access against eleventh-hour export controls is the same lesson COCOM learned in the 1950s: collective economic coercion only holds as long as every partner believes the terms are being applied in good faith, not as a one-way ratchet. Diversifying compute across allied territory doesn't just diversify risk, it diversifies who gets a say — which might be the more interesting geopolitical story here than the raw infrastructure numbers.
Nice discussion. On job displacement, there is also the possibility of dividing up the work that remains: https://www.amazon.com/dp/B00U0C9HKW
The angle of the last question is quite unique. Great job including it!
Lots of stuff to just completely love here, but maybe my favorite part is the emphasis on the very simple solution of raising corporate income taxes as a broad-based non-winner-picking method of raising the necessary revenue to fund <strikethrough> deficit reduction </strikethrough> whatever sets of programs we need to manage the AI transition well. I am sort of blown away by how tempted many people are to fundamentally reinvent the wheel here. But regular old wheels work very well for the problem we're confronting — specifically the problem where companies rightly have incentives to find creative ways to lower labor requirements. We should capture, as a society, some of the performance gains they get from doing that!
AI as a political crisis is a regime path. The investment question is still which policy and capex constraints reprice first.
Geopolitical supply chain tensions in semiconductor manufacturing reinforce the need for diversified chip sourcing and energy-efficient model architectures across enterprise software sectors.
The data center hosting question is essentially chip export controls in reverse, and it runs into the exact same structural problem — leverage built on an ally's trust doesn't survive the moment that trust turns out to be conditional. Anton's point that allies could weaponize hosting access against eleventh-hour export controls is the same lesson COCOM learned in the 1950s: collective economic coercion only holds as long as every partner believes the terms are being applied in good faith, not as a one-way ratchet. Diversifying compute across allied territory doesn't just diversify risk, it diversifies who gets a say — which might be the more interesting geopolitical story here than the raw infrastructure numbers.