Yeah - people are talking about replacing jobs with AI. As if it is not totally obvious that people like Sam Altman will totally bleed you dry after you fired all your workers. You will not save on your wage bill, you will simply give the money to Sam Altman
I’ve been preaching this for the past couple of years. Everything up until now has been entirely about gaining market share, and AI will never be cheaper than it is right now, and it’s not cheap.
Just look at the “earnings” for companies like openAI. They are 1000+% in the red. It’s impossible for them to change their sales model enough to make that profitable. As more data centers go up, the operating costs are also going to go up.
I’ve been telling people that now is the best time in the past decade or more to learn how to code. There will be positions available in the coming years when the only junior devs available are vibe coders.
The crazy thing is, this isn’t really a “squeeze” in the traditional sense. The problem was that every single mainstream AI product has been heavily subsidized…because it’s wildly expensive and not even close to being profitable.
That sort of subsidization was only going to last for so long. The dam is starting to crack. People aren’t ready to pay what AI truly costs.
Pretty sure they picked the wrong tech to try and lock people into. It isn’t hardware and doesn’t have some kind of proprietary interface that takes time to get used to when switching. Some models might be better than others at specific things, but not enough to justify the prices they are going to charge for output you have to review and fix.
This is literally the easiest thing to jump ship from.
It would be if so many companies hadn’t bought into the microsoft ecosystem so hard.
My company did the usual BS of “well we have an e5 license so it’s free”. Now we are married to all their stupid shit and have no relationships with other providers.
End of the month gonna be expensive. Now expect a scramble to cut master service agreements and contracts with others in a panic.
It isn’t hardware and doesn’t have some kind of proprietary interface that takes time to get used to when switching.
The proprietary interface is kinda the largest selling point, besides the once cheap prices. But the $10 plan going from 300 multi-hour prompts per month to 20 quick prompts per month effectively makes it worthless. Though, you could just pay a few hundred/thousand per month and continue, or bring your own API key from elsewhere, but at that point just use someone else’s interface. It’s not that much better than competitors and copilot isn’t offering anything unique.
That’s the stupidest thing about these AI companies’ valuation.
They don’t even really own anything!
Their models – their main proprietary IP – are not copyrightable or patentable, and not legally protected in any way. Any competitor can copy them at any time and then offer the same service for cheaper, without the overhead costs for training. The giants of the AI industry could easily be undercut and replaced at any time.
This is literally the easiest thing to jump ship from.
I’m not sure about that. We see professional developers complaining all the time when their AWS or GitHub account is banned. But this time we’re talking about vibe coders who have less skills than the average developer.
I believe that is largely their employers’ decision. It is easy to switch, but it is the boss who makes that decision, not the tech-literate team. Bosses like big well-known companies that can absorb blame so they can continue to cash their big pay checks even after failures that were clearly the fault of the 3rd party that no other bosses could blame them for choosing.
This is literally the easiest thing to jump ship from.
It depends how heavily you are leaning on ML tools to do business processes honestly.
It’s easy to implement something that mostly works and doesn’t need a ton of baby sitting, but moving from one solution to another is like rebuilding an ERP if you have gotten deep enough into the weeds.
This bubble is super scary though. The only things I can see propping it up would be world governments once the tech companies and other large enterprises halt spending. I don’t think the US can shoulder the costs and nobody else is gonna lol
Have you seen the IPOs and the rule changes that the stock exchanges and index funds made to please the AI overlords? It’ll be US pension funds left holding the bag when the bubble goes pop
The unique thing about GitHub Copilot (and all the other vibe-coding tools) is that they’re speed-running the playbook because this shit is not profitable. It can’t be. Their costs scale up with usage, unlike every other business that can take advantage of economies of scale, so they’ve skipped the slow, steady enshittification phase and jumped directly into the “squeeze blood from this stone to keep the scam going a little longer” phase.
Plain inference is profitable actually, that’s why there are a hundred inference providers on OpenRouter who compete by undercutting each other. The labs however aren’t profitable because training the models is a huge drain.
As others have pointed out, the compute cost of inference is only one small part of the puzzle. All the frontier model providers - which OpenRouter gives you access to - are massively raising their prices in desperate bids to recoup the cost of model generation in the first place.
There’s really no hiding from the token apocalypse unless you’re running a model on your own hardware.
But they can only do that because others are doing traning, no? There’s no point at which you can go “okay it’s all inference from here”, the model needs to be updated with new information/guardrails/context to continue being useful for most use cases
In a vacuum maybe but are they profitable if you add the infrastructure investments to the mix? What about model development? There was a shit ton of money that was spent. Covering the running costs is not enough. At some point someone has to pay for the investments.
I remember having to sit down my boss and explain how it can only become more expensive over time. It’s the big tech playbook after all. Didn‘t matter. I‘m told again and again how AI is only becoming stronger and cheaper. Especially during salary negotiations. Nasty stuff. They know I know it‘s BS and they still cling to this nonsensical narrative because it would be very beneficial to them and very bad for me.
Open weight LLMs are actually pretty cheap because there are competing providers. But something tells me your boss isn’t using openrouter to find the best price per million tokens lol
The strategy is always to gain a monopoly or near monopoly on a market before pushing for the enshittification of the product to reduce costs and maximize profits, once customers have become dependent on said product, then pray that most choose the path of least resistance which is staying and dealing with the worse and more expensive version of what they’re used to rather than retraining or restarting from zero elsewhere.
This is going much faster than most private equity enshittification. That usually takes about 5 years to start. And has a system or hardware that makes switching painful. That doesn’t apply here, it’s easy to switch and almost a drop in change.
They are burning money so quickly that they don’t have the luxury of using the “slowly boil the frog” standard VC tactic, so they have to start the boiling before the frog is comfortable.
Cars are addictive because if you don’t have a car, and they are not common, you don’t need one. But they are confortable, and this is what hooks people - our Naenderthal brain is wired to save energy and effort. Because cars are, at first, faster than, say, streercars, they are used more and more, and especially people use them for commuting. Because of Marchetti’s constant, people use their daily time budget of about one hour for commuting in a car, and travel longer and longer distances. What results collectively is subururban sprawl, small shops in residential areas, as well as jobs nearby disappear and so on.
Once you are there, it is very difficult to go back, because daily distances are too large for streetcars, bicycles, and buses.
And to add insult to injury, cars stop to be fast once many people use them, since they need enormous amount of longitudinal space to be operated safely. This huge space requirement further accelerates sprawl.
Cars are an collective addiction, highly damaging. This kind of AI might become another.
Human beings are terrible at balancing short-term gains for long-term consequences. It’s mixed into our DNA. Our ancient ancestors, securing immediate calories or escaping a threat was a matter of life and death. Long-term planning wasn’t as critical as immediate survival. Now do note, that’s not an excuse for the people who foolish went head long into this.
This is why this struggle with the rich and powerful is eternal. It fundamentally taps on an ingrained flaw we collective fall for every single time. There is no one solution, there can never be one solution. People must forever fight themselves and the powerful from the exploitation of this fundamental flaw of humanity.
I’m sorry, did everybody else not see this coming from miles away? This is the private equity playbook.
When something is too good to be true, you ALWAYS have to be ready to either jump ship, massively change how you do things, or pay through the nose.
Yeah - people are talking about replacing jobs with AI. As if it is not totally obvious that people like Sam Altman will totally bleed you dry after you fired all your workers. You will not save on your wage bill, you will simply give the money to Sam Altman
I’ve been preaching this for the past couple of years. Everything up until now has been entirely about gaining market share, and AI will never be cheaper than it is right now, and it’s not cheap.
Just look at the “earnings” for companies like openAI. They are 1000+% in the red. It’s impossible for them to change their sales model enough to make that profitable. As more data centers go up, the operating costs are also going to go up.
I’ve been telling people that now is the best time in the past decade or more to learn how to code. There will be positions available in the coming years when the only junior devs available are vibe coders.
Private equity?
Admittedly, it’s capitalism. But I just chose a negative entity.
It’s not even just private equity, this is more or less the exact strategy Microsoft uses.
Also reminds of Android
It reminds of any software made by any company that is at all well known. They are all operating from the same playbook.
The crazy thing is, this isn’t really a “squeeze” in the traditional sense. The problem was that every single mainstream AI product has been heavily subsidized…because it’s wildly expensive and not even close to being profitable.
That sort of subsidization was only going to last for so long. The dam is starting to crack. People aren’t ready to pay what AI truly costs.
And it doesn’t seem like they ever will be. The LLM value proposition is already dubious at today’s subsidized rates.
Pretty sure they picked the wrong tech to try and lock people into. It isn’t hardware and doesn’t have some kind of proprietary interface that takes time to get used to when switching. Some models might be better than others at specific things, but not enough to justify the prices they are going to charge for output you have to review and fix.
This is literally the easiest thing to jump ship from.
It would be if so many companies hadn’t bought into the microsoft ecosystem so hard. My company did the usual BS of “well we have an e5 license so it’s free”. Now we are married to all their stupid shit and have no relationships with other providers. End of the month gonna be expensive. Now expect a scramble to cut master service agreements and contracts with others in a panic.
If you’re a pure vibe coder, this is literally 9/11.
The proprietary interface is kinda the largest selling point, besides the once cheap prices. But the $10 plan going from 300 multi-hour prompts per month to 20 quick prompts per month effectively makes it worthless. Though, you could just pay a few hundred/thousand per month and continue, or bring your own API key from elsewhere, but at that point just use someone else’s interface. It’s not that much better than competitors and copilot isn’t offering anything unique.
That’s the stupidest thing about these AI companies’ valuation.
They don’t even really own anything!
Their models – their main proprietary IP – are not copyrightable or patentable, and not legally protected in any way. Any competitor can copy them at any time and then offer the same service for cheaper, without the overhead costs for training. The giants of the AI industry could easily be undercut and replaced at any time.
The hardest part about the copying is the actual copying without having access to the weights or even just a ready to run file for the model.
IIRC Deepseek kinda did something like that by asking ChatGPT tons of questions to train their own model or something
Yeah, you can do that … or some good old fashioned corporate espionage.
Or, hell, just ask ChatGPT for its weights model. With how shitty these AI companies are at security and guardrails, that might just work.
It can still be pretty difficult to jump ship for any large corporation, but yeah there’s certainly harder things.
I’m not sure about that. We see professional developers complaining all the time when their AWS or GitHub account is banned. But this time we’re talking about vibe coders who have less skills than the average developer.
I believe that is largely their employers’ decision. It is easy to switch, but it is the boss who makes that decision, not the tech-literate team. Bosses like big well-known companies that can absorb blame so they can continue to cash their big pay checks even after failures that were clearly the fault of the 3rd party that no other bosses could blame them for choosing.
And they need to subscribe to access and execute their troubleshooting options.
All of them also bring their own comfortable export feature.
“I want to share all of this with my team. Create the prompt that is necessary to do this”
It depends how heavily you are leaning on ML tools to do business processes honestly.
It’s easy to implement something that mostly works and doesn’t need a ton of baby sitting, but moving from one solution to another is like rebuilding an ERP if you have gotten deep enough into the weeds.
This bubble is super scary though. The only things I can see propping it up would be world governments once the tech companies and other large enterprises halt spending. I don’t think the US can shoulder the costs and nobody else is gonna lol
Have you seen the IPOs and the rule changes that the stock exchanges and index funds made to please the AI overlords? It’ll be US pension funds left holding the bag when the bubble goes pop
astronaut meme.
The unique thing about GitHub Copilot (and all the other vibe-coding tools) is that they’re speed-running the playbook because this shit is not profitable. It can’t be. Their costs scale up with usage, unlike every other business that can take advantage of economies of scale, so they’ve skipped the slow, steady enshittification phase and jumped directly into the “squeeze blood from this stone to keep the scam going a little longer” phase.
Wait, what? You say it is a scam? A kind of technical Ponzi scheme? A Madoff syndicate on steroids?
Good sign that it may be over soon.
Absolutely 100% do not count on it
Plain inference is profitable actually, that’s why there are a hundred inference providers on OpenRouter who compete by undercutting each other. The labs however aren’t profitable because training the models is a huge drain.
As others have pointed out, the compute cost of inference is only one small part of the puzzle. All the frontier model providers - which OpenRouter gives you access to - are massively raising their prices in desperate bids to recoup the cost of model generation in the first place.
There’s really no hiding from the token apocalypse unless you’re running a model on your own hardware.
But they can only do that because others are doing traning, no? There’s no point at which you can go “okay it’s all inference from here”, the model needs to be updated with new information/guardrails/context to continue being useful for most use cases
In a vacuum maybe but are they profitable if you add the infrastructure investments to the mix? What about model development? There was a shit ton of money that was spent. Covering the running costs is not enough. At some point someone has to pay for the investments.
Didn’t deepseek train their model for $6m?
I remember having to sit down my boss and explain how it can only become more expensive over time. It’s the big tech playbook after all. Didn‘t matter. I‘m told again and again how AI is only becoming stronger and cheaper. Especially during salary negotiations. Nasty stuff. They know I know it‘s BS and they still cling to this nonsensical narrative because it would be very beneficial to them and very bad for me.
Open weight LLMs are actually pretty cheap because there are competing providers. But something tells me your boss isn’t using openrouter to find the best price per million tokens lol
The strategy is always to gain a monopoly or near monopoly on a market before pushing for the enshittification of the product to reduce costs and maximize profits, once customers have become dependent on said product, then pray that most choose the path of least resistance which is staying and dealing with the worse and more expensive version of what they’re used to rather than retraining or restarting from zero elsewhere.
Capitalism 101.
Which is why deepseek releasing it’s model open source was such a big deal. The monopoly isn’t feasible.
I didn’t see it coming as I am on Codeberg 😎
I mean you could equally be on github and just not use copilot
This is going much faster than most private equity enshittification. That usually takes about 5 years to start. And has a system or hardware that makes switching painful. That doesn’t apply here, it’s easy to switch and almost a drop in change.
They are burning money so quickly that they don’t have the luxury of using the “slowly boil the frog” standard VC tactic, so they have to start the boiling before the frog is comfortable.
The tactic has worked for drug dealers for decades
Also for example for the car industry.
Cars are addictive because if you don’t have a car, and they are not common, you don’t need one. But they are confortable, and this is what hooks people - our Naenderthal brain is wired to save energy and effort. Because cars are, at first, faster than, say, streercars, they are used more and more, and especially people use them for commuting. Because of Marchetti’s constant, people use their daily time budget of about one hour for commuting in a car, and travel longer and longer distances. What results collectively is subururban sprawl, small shops in residential areas, as well as jobs nearby disappear and so on.
Once you are there, it is very difficult to go back, because daily distances are too large for streetcars, bicycles, and buses.
And to add insult to injury, cars stop to be fast once many people use them, since they need enormous amount of longitudinal space to be operated safely. This huge space requirement further accelerates sprawl.
Cars are an collective addiction, highly damaging. This kind of AI might become another.
Human beings are terrible at balancing short-term gains for long-term consequences. It’s mixed into our DNA. Our ancient ancestors, securing immediate calories or escaping a threat was a matter of life and death. Long-term planning wasn’t as critical as immediate survival. Now do note, that’s not an excuse for the people who foolish went head long into this.
This is why this struggle with the rich and powerful is eternal. It fundamentally taps on an ingrained flaw we collective fall for every single time. There is no one solution, there can never be one solution. People must forever fight themselves and the powerful from the exploitation of this fundamental flaw of humanity.