There's a very important section in the podcast where the guests discuss capital availability. Dylan mentioned that we flew through $150B a year during the dot-com bubble.
This caught my attention, so I dug into it:
1. Dylan meant the total CapEx of telecom companies to build out internet infrastructure throughout the US. Telecoms spent ~$120B/year in CapEx (~$220B/year in 2024 dollars), basically building out fiber.
2. It's unclear to me exactly how much was raised to finance this (a lot of capital raised went into competitive M&A, so not strictly toward CapEx -- some figures say about $1.6T total raised), but it appears to have been at least $600B in bonds (~$1.1T in 2024).
3. These are all public markets raises, of course. And bonds are debt instruments, so these companies got overleveraged, and that debt eventually crushed those companies.
I think it's really important that these were raises for debt in public markets. Debt markets are much larger than equity markets, and public markets are much larger than private markets.
By comparison, private venture capital (for equity) only hit around $35B/year in 2000 dollars (~$64B in 2024 dollars). There were tons of IPOs, but those were also relatively small in terms of capital raised (~$72B market cap in the top 24 IPOs of 1999). Therefore, I think the guests are using an analogy that breaks in the details. Raising capital on the same scale as the Telecoms back in the day, which is clearly what's required for AI, requires a different capital structure:
1. Different from the capital structure they've used to-date, since raising $10B+ of equity in private markets is just very difficult, even for a god-tier fundraiser like Sama.
2. Different from the capital structure the Telecoms used (which eventually killed them): AI companies should not want to take on any debt, since profitability is so uncertain.
So, some financial/corporate engineering is probably a requirement for 2026 onwards, and that shouldn't be underrated as a logistical hurdle on the way to scaling. Public companies like Microsoft, Google, and Nvidia have a much easier answer to this than current-privates like OpenAI and Anthropic. There's some good discussion on this by Doug O'Laughlin here: https://www.fabricatedknowledge.com/p/lessons-from-history-the-rise-and. I also have some discussion of this topic in a recent substack essay of mine: https://loeber.substack.com/p/22-a-bubble-is-rarely-a-bubble
Some of this discussion reminded me of what I wrote here: https://open.substack.com/pub/semiconductor/p/the-relentless-pursuit-of-moores
There's a very important section in the podcast where the guests discuss capital availability. Dylan mentioned that we flew through $150B a year during the dot-com bubble.
This caught my attention, so I dug into it:
1. Dylan meant the total CapEx of telecom companies to build out internet infrastructure throughout the US. Telecoms spent ~$120B/year in CapEx (~$220B/year in 2024 dollars), basically building out fiber.
2. It's unclear to me exactly how much was raised to finance this (a lot of capital raised went into competitive M&A, so not strictly toward CapEx -- some figures say about $1.6T total raised), but it appears to have been at least $600B in bonds (~$1.1T in 2024).
3. These are all public markets raises, of course. And bonds are debt instruments, so these companies got overleveraged, and that debt eventually crushed those companies.
I think it's really important that these were raises for debt in public markets. Debt markets are much larger than equity markets, and public markets are much larger than private markets.
By comparison, private venture capital (for equity) only hit around $35B/year in 2000 dollars (~$64B in 2024 dollars). There were tons of IPOs, but those were also relatively small in terms of capital raised (~$72B market cap in the top 24 IPOs of 1999). Therefore, I think the guests are using an analogy that breaks in the details. Raising capital on the same scale as the Telecoms back in the day, which is clearly what's required for AI, requires a different capital structure:
1. Different from the capital structure they've used to-date, since raising $10B+ of equity in private markets is just very difficult, even for a god-tier fundraiser like Sama.
2. Different from the capital structure the Telecoms used (which eventually killed them): AI companies should not want to take on any debt, since profitability is so uncertain.
So, some financial/corporate engineering is probably a requirement for 2026 onwards, and that shouldn't be underrated as a logistical hurdle on the way to scaling. Public companies like Microsoft, Google, and Nvidia have a much easier answer to this than current-privates like OpenAI and Anthropic. There's some good discussion on this by Doug O'Laughlin here: https://www.fabricatedknowledge.com/p/lessons-from-history-the-rise-and. I also have some discussion of this topic in a recent substack essay of mine: https://loeber.substack.com/p/22-a-bubble-is-rarely-a-bubble
Two of my favorite content creators doing a collab 🙌
Fucking amazing! 💚 🥃
What a classic lmao, love it
Huawei is playing a somewhat analogous role to the centralized idea you discuss