I actually think it is going to change in the next 3-6 months. Anthropic just filed for IPO, and I believe openAI is gonna do the same soon as well. Being publicly traded comes with mandatory reporting on a lot of stuff they haven’t really shared fully yet, and the profit margins are gonna go under a microscope… except that you can see what’s wrong with the profit margins with the naked eye.
They’re gonna jack up the prices for LLM usage like crazy in the coming months, and it’s gonna start being a huge paywall to people using it. For students, it’s gonna go from convenient to prohibitively expensive virtually overnight. I think the problem in the educational domain may solve itself.
Amazon ran for many years post-IPO with no profit. Now they’re profitable and far into rent-seeking mode. It’s a well-understood strategy when the goal is to grab market share.
What about cheaper llms from other locales? It’s hard to imagine the government blocking access writ large, though I’m sure many in this administration salivate at the thought.
This isn’t about anyone blocking access like censorship. This is an economic issue. The LLM companies are all hemorrhaging cash, and none of them have a clear or realistic path to profitability.
It’s not a question of how good or bad the LLM is, it’s a question of watt hours and bandwidth. It takes a certain amount of electricity to run so your prices and profit margins are directly correlated with the price of electricity.
LLMs run out of data centers with cheap electricity and cheap bandwidth are going to be the cheapest ones on the market. For electricity this would typically be places with cheap renewables nearby like large hydroelectric plants. Bandwidth is a little trickier as there’s not as obvious an indicator of where bandwidth is cheap and plentiful but typically it’s going to be near major population centers. Putting those together there’s probably only a small handful of locations in the world where it’s economically viable to run these data centers.
I’ve seen a lot of colleges buying their students subs to one of the major LLM’s. So maybe come new semester that will change but it depends on the timing.
The only company making money on this shit at the moment is Nvidia. That’s it. None of the companies that are actually doing the model development and deployment are making money - they’re hemorrhaging cash, in fact. They have no clear or realistic path to profitability, either.
I actually think it is going to change in the next 3-6 months. Anthropic just filed for IPO, and I believe openAI is gonna do the same soon as well. Being publicly traded comes with mandatory reporting on a lot of stuff they haven’t really shared fully yet, and the profit margins are gonna go under a microscope… except that you can see what’s wrong with the profit margins with the naked eye.
They’re gonna jack up the prices for LLM usage like crazy in the coming months, and it’s gonna start being a huge paywall to people using it. For students, it’s gonna go from convenient to prohibitively expensive virtually overnight. I think the problem in the educational domain may solve itself.
Amazon ran for many years post-IPO with no profit. Now they’re profitable and far into rent-seeking mode. It’s a well-understood strategy when the goal is to grab market share.
What about cheaper llms from other locales? It’s hard to imagine the government blocking access writ large, though I’m sure many in this administration salivate at the thought.
This isn’t about anyone blocking access like censorship. This is an economic issue. The LLM companies are all hemorrhaging cash, and none of them have a clear or realistic path to profitability.
Agreed.
What happens when we can buy access to a Singaporean or French LLM on the cheap as the US monoliths raise this paywall?
They’ll lobby to prevent you from doing that, and will try strangling the Singaporeans and Frenchies with intellectual-property lawsuits.
Then everyone uses that provider for a few month and they have to raise prices as well or go bankrupt.
I’m pretty sure French AI companies (and maybe the others) will have to raise their prices. There is no magical alternative.
That’s… not how LLM infrastructure works.
Yep, this person did not take DeepSeek into account.
I wouldn’t put it past them banning it once they see how much it’s kicking their asses though
It’s not a question of how good or bad the LLM is, it’s a question of watt hours and bandwidth. It takes a certain amount of electricity to run so your prices and profit margins are directly correlated with the price of electricity.
LLMs run out of data centers with cheap electricity and cheap bandwidth are going to be the cheapest ones on the market. For electricity this would typically be places with cheap renewables nearby like large hydroelectric plants. Bandwidth is a little trickier as there’s not as obvious an indicator of where bandwidth is cheap and plentiful but typically it’s going to be near major population centers. Putting those together there’s probably only a small handful of locations in the world where it’s economically viable to run these data centers.
That’s not a constant across all LLMs. There are big differences between LLMs in how much inference you get per watt-hour.
I’ve seen a lot of colleges buying their students subs to one of the major LLM’s. So maybe come new semester that will change but it depends on the timing.
I hope you’re right. I have half expected AI to be dealt to students in the way every student gets Office 365.
I am just nervous is all. Maybe each “Student” profile gets a set number of tokens but you can upgrade to “Student+” for more. I joke but I’m scared.
The only company making money on this shit at the moment is Nvidia. That’s it. None of the companies that are actually doing the model development and deployment are making money - they’re hemorrhaging cash, in fact. They have no clear or realistic path to profitability, either.
I’ll bet the people running the data centers are coining it too.