Solid analysis on TSMC's monopoly problem. The comparison of capex grwoth (10%) vs revenue growth (50%) since 2022 really underlines how their lack of competition enabled cautious decisions. It's a textbook case of monoploy incentives being misaligned with industry needs. The 2-3 year fab timeline means this bottleneck probably shapes AI roadmaps until 2029 at least.
For sure, but in all fairness, they earned that monopoly. It was entirely self-inflicted by Intel, they should've split their foundry business 20 years ago.
Bypassing the grid makes sense from a speed to power perspective, but eventually these campuses will want to connect to the larger system for efficiency and financial arbitrage. And there are limits to how many gas plants can come online before the political pendulum catches up.
“OpenAI’s community energy commitments are smart politics, but behind-the-meter generation has become table stakes for anyone serious about speed.”
for sure. it’s really been interesting to watch this space. great post.
Thank you for the kind words, hope you've been well!
Solid analysis on TSMC's monopoly problem. The comparison of capex grwoth (10%) vs revenue growth (50%) since 2022 really underlines how their lack of competition enabled cautious decisions. It's a textbook case of monoploy incentives being misaligned with industry needs. The 2-3 year fab timeline means this bottleneck probably shapes AI roadmaps until 2029 at least.
For sure, but in all fairness, they earned that monopoly. It was entirely self-inflicted by Intel, they should've split their foundry business 20 years ago.
Bypassing the grid makes sense from a speed to power perspective, but eventually these campuses will want to connect to the larger system for efficiency and financial arbitrage. And there are limits to how many gas plants can come online before the political pendulum catches up.
If by eventually you mean 10+ years from now, yes I would agree.
I think gas plants are already sold out, jet turbines are the move now ✈️