Just look at the dot com bubble,

That example supports my argument, not yours. During the dotcom boom hundreds and thousands of companies started up hoping to gain market share and become leaders in the "Internet E-commerce" space as we called it during that time. Lots and LOTS of these companies failed, but not all of them.

The dotcom boom gave us giant behemoths that exist today like Amazon and Google. Amazon is the default online store. Google is the default search engine and online ad networks. This is exactly my point. Most of the AI LLM model companies will fail and cease to exist, but the few that do survive will be massively successful in their space.

I don’t really have anything else I want to dive into beyond this.

Uh, okay. I'm confused about the purpose of your initial post. Did you just want to make statements and not be challenged on them or have conversation on it?

You can get a dot com domain name for about $11 (depending on availability),

And during the dotcom boom it was $35/year (with a minimum 2 year sub), but that was about regulatory scarcity by InterNIC (or later Network Solutions), not a technological limitation.

my internet bill in Chicago is around $20 monthly for a GIGABIT of speed. My 5G phone plan is about $60 a month.

Again, this supports my argument, not yours. During the dotcom boom days there were thousands of little ISPs each selling dial up. Most of those are gone, but a handful that developed broadband are the companies that are mega players with large revenue today.
I don't know who your ISP is, but for conversation lets say its Comcast. Comcast Corporation's annual net revenue for fiscal year 2025 was $123.71 billion.

So my point being, these things that people used to believe briefly would be insanely valuable are kind of just common and understood at this point. Technologies that were once inaccessible or very expensive are now available to many very cheap, it’s how it’s happened every time before.

I think you're seeing commoditization of technology and assuming that it means less money is spent. Its the opposite. In 2001 how many internet subscriptions did you have? Probably one, your ISP and it probably cost you $15-$25 for dial up internet or maybe $60 for broadband (Cable Modem, ADSL, or possibly ISDN).

How many internet subscriptions do you have today? You've already listed two, your residential ISP, and your mobile phone internet connection. How about Spotify? Netflix? HBOmax? Some fitness app perhaps? A cloud based digital picture frame service? How about SaaS solutions? Adobe Creative Cloud maybe? How about any online gaming? Gamepass? An MMORPG? Something with real money to in-game currency for digital items?

I am pretty confident you're probably spending way more on internet services today than you were during the dotcom bubble bursting.

You won’t need a data center to run LLMs that are powerful enough to do certain real world job tasks.

To do some tasks sure. We're starting to see that today with small tasks and embedded running on mobile phones, but I think its naive to assume that thats going to be all the AI each of us is going to use and need in the years ahead. Even then, those local models running on phones I mentioned are tied to other monetized services where you either pay for it or your data is sold in exchange for their use.

If local computing was enough to do computing needs we wouldn't have cloud computing. We have cloud computing. We will likely have some form of cloud AI far into the future too that eclipses the ability of our local compute to handle.

And like you mentioned, companies like Amazon and Google don’t even need to make money on their LLMs, the competition is massive.

Like all for-profit businesses, they don't need to make more money, but they want to. If AI is a business offering people will pay for (and it certainly is in at least some places/industries) companies will try to corner that market and the profits that come with it. Lots will fail as has happened in history with other bubbles. A handful will succeed and reap the profits.

You can get a dot com domain name for about $11 (depending on availability), my internet bill in Chicago is around $20 monthly for a GIGABIT of speed. My 5G phone plan is about $60 a month.

So my point being, these things that people used to believe briefly would be insanely valuable are kind of just common and understood at this point. Technologies that were once inaccessible or very expensive are now available to many very cheap, it's how it's happened every time before. You won't need a data center to run LLMs that are powerful enough to do certain real world job tasks.

And like you mentioned, companies like Amazon and Google don't even need to make money on their LLMs, the competition is massive.

When everyone can run local models that can do most tasks for less than 1000th the price, no one will pay for AI services.

I don't agree with that argument. If that logic held then no one would be using cloud services because they'd just buy enough compute to run all their workloads locally, yet cloud servers and hyperscalers are only growing, not shrinking. While some of that growth is from customers buying AI (using either GPUs or dedicated silicon), even without the AI growth cloud services continue to grow.

There’s a reason they’re trying to keep RAM prices so high.

Who is "they" here? Do you think there is a conspiracy to inflate RAM prices and that the high prices are not because of RAM demand in the free market?

You can use a local LLMs 9b to do tasks that a frontier model couldn’t do 2 years ago.

Sure, but I think we can all agree that even today's most modern models are shit at lots of jobs and tasks. I believe there will be a practical limit to our technology that is not AGI, but I think we're still a long way from it.

Will there be profitable companies in the future that heavily rely on AI? Yes, but it’s not going to be anything related to software. Any AI focused company will never be profitable.

I'm not sure I'm following you distinction here. The profitable companies in your statement, where are they going to get and run their AI models?

There are certainly some AI-only companies like OpenAI and Anthropic, and I agree both of them have a tough road ahead to profitability because they run on someone else's compute/silicon. However there are companies that are highly vertically integrated that have their own compute like Google and (barf!) Grok.

I see a future where the non-vertically-integrated AI companies get bought up by someone with their own large scale compute real estate. Amazon struck out building their own AI (Nova), so I'm guessing Bezos and the board would love to buy up Anthropic or OpenAI if they could afford either.

When everyone can run local models that can do most tasks for less than 1000th the price, no one will pay for AI services.

There will be no "winner". There's a reason they're trying to keep RAM prices so high. I just watched some dude on YouTube create his own AI integrated IDE. Something like Cursor is already being directly out compete by OpenCode and CommandCode.

You can use a local LLMs 9b to do tasks that a frontier model couldn't do 2 years ago. These companies won't be profitable.

Will there be profitable companies in the future that heavily rely on AI? Yes, but it's not going to be anything related to software. Any AI focused company will never be profitable.

• "AI companies [together] are never going to earn the money back that was invested [every single one of them] in them"

OR

• "AI companies are [No single AI company is] never going to earn the money back that was invested in [them] that company

I agree with the first, I disagree with the second. Further, the AI companies (and even investors) know this too. What is happening right now is every AI company is pooring massive resources into becoming the default AI solution. This is the current generation's VHS vs Betamax war. Prior wars were Blu-ray vs HD-DVD, Coke vs Pepsi, etc.

The single winning AI company (and maybe one more) will be able to pay back this initial investment, and much more.