DragonGod's Shortform

DragonGod

This is a special post for quick takes by DragonGod. Only they can create top-level comments. Comments here also appear on the Quick Takes page and All Posts page.

Obviously AI is a real technology and my default expectation is that scaling up the existing paradigms as at October 2026 would lead to transformatively capable systems within the decade.

Is it a financial bubble though?

Should I model the AI industry like the internet in the dotcom era where the underlying technology was real and important but the financials were out of whack.

Or is it more or less smooth sailing from here? I guess how contingent on transformative AI in the near term are the valuations for Anthropic, OpenAI, Nvidia, SpaceXAI, Coreweave, etc.?

Or given my expectations for AI progress on the default path, should I expect a massive market correction in the near future?

What is currently being priced in to existing valuations and what is not? What is required for the AI companies to justify their valuations?

What does the world in which the bubble pops look like?

The biggest proponents of "it's all a bubble" (e.g. Ed Zitron) also seem to think the technology is fake, so I've been unable to extract useful signal from their content since the overall epistemics are just whack all around?

And the vibe I have gotten from several others is that the technology is very real and the market is nonetheless (overheated/due for a correction): "AI is the Internet".

And there's like a third position where the market isn't accurately pricing in AI yet. Where Nvidia's market cap grows to 10T+ in 2027, Anthropic's annual revenue approaches 1T before 2030, etc.

I guess what I'm trying to figure out is what epistemic position I should be in wrt AI industry financials? I am not at all financially savvy and don't have a grounded independent position on how healthy the industry is. The cyclic financing feels like it could be a structural risk, but I don't know how much of a concern it should be.

I'm much more bullish on AI capabilities than on AI economic transformation because US regulation sucks arse, they can't really build quickly anymore and can barely build at all, institutions are slow to adapt and in the adoption of new technologies, etc. So I expect GDP to be a significantly lagging indicator.

A major market correction does feel like it should lengthen my timelines because my expectation of reaching transformatively capable AI systems on the default path is just ~scaling from here onwards. And if the available financing for scaling dries up, it does feel like my timelines should lengthen as a result?

I'd say it's likely that at least some part of the AI stack is in a bubble right now. When you look at the valuations right now, the market is basically forecasting that margins stay high and the rents from AI are distributed across the stack. There are a lot of ways that could break.

Start with chips. Compare how the market values the RAM stocks (SK Hynix, Micron) versus the way it values the fabless chipmakers (Nvidia, AMD, Broadcom, etc.). Everyone has spectacular margins right now because chips are scarce but the market currently Micron at something... (read more)

6Brendan McGarry3h

Yes, I completely avoid anything with people like Ed Zitron now because they're clearly neither informed nor interested in being precise, which is a shame because the field could benefit from some thoughtful criticisms and suggested changes. As far as I can tell, most informed people don't expect the technology to stall. OpenAI and Anthropic have both been clear that shocks could cause serious problems for them, including but not limited to investors not seeing any returns at all. A variety of causes such as interest rate hikes, a semiconductor shortage (such as if Taiwan got invaded), etc., could cause the frontier labs to face serious headwinds. A framing I haven't seen much is that it seems like the platform companies and the labs are fighting to see who can get the bigger share of the profits. Meta is widely believed to have released the first large open-weights model in order to "commoditize their complement" (have cheap bots widely available to not cut into their profits as an ad company); it was Amazon who produced the report that Claude Fable "could be jailbroken" (overblown claim with potential motivated reasoning) despite being a significant investor in and partner with Anthropic; and other platform companies and other parts of the tech stack/supply chain have seemed to want to maintain access to open-weights models and not have a clear frontrunner (like when Anthropic had Mythos before anyone else had comparable models). My view is that the platform companies stand to benefit if there is a "controlled explosion", or at least a reining in, of frontier labs. If one or both of the frontier labs were to go out of business, there'd very plausibly be a bidding war for the parts between Google, Microsoft, Amazon, etc. A hyperscaler would vacuum up that bankrupt lab's weights, employees, research, and other assets and keep scaling, so I'd expect things to largely continue uninterrupted. Even setting the labs aside, I don't imagine a correction would slow progr

DragonGod's Shortform

7

Ω 3

DragonGod's Shortform

7

Ω 3