This doesn’t solve the problem because (tautologically) the more AI prices go down the less money the companies make. If right now today the companies are operating at a profit and a price war causes the API costs to sink 90% next year, and their capex amortization costs stay fixed.
AI prices going down means the models are improving, particularly from the efficiency angle (which is inevitable, given the nature of tech). That means all they have to do is maintain a large enough customer base at a rate high enough to ensure loss decreases continuously over time, until eventually the pass the point where they're just gaining. Healthy competition ensures that improvement savings are actually passed on to users in a measured manner, so they don't become too greedy in trying to get to and increase gains.
But now you’re describing a commodity, and the competition will erode profits, and their valuations are bananas, unless someone can find a business model that truly differentiate and creates a moat.
Is this really for comparable models though? Will folks at scale continue to choose Anthropic frontier AI model if OpenAI releases a similar generation at a 90% discount with comparable capabilities? It feels like the fungibility assumes delineation by capability _and_ cost. No one is choosing sonnet over opus at similar price points.
I could get into this line of debate that I find interesting but it doesn't contradict my point that the article itself is wrong and written on false assumptions.
I mean.. get into it! Isn't it much more interesting for you to defend the real world strength of your argument than it is to just gotcha the article? It's highly relevant to all of this either way. Don't leave us hanging!
Ok, so one last question - I think folks like Ed Zitron are too much I. The bear camp, he’s “AI is useless” and I’m not there, I do think AI changes the game. But everyone has been comparing OpenAI and Anthropic to the dotcom bubble, but isn’t -
The historical precedent on being able to capture value of a raw technology is not good. This is why the joke they promised us flying cars and we got ads.
So everyone says “Anthropic is like Uber”. But Uber is a service with people, not an underlying technology with commodity economics.
So in order for OpenAI and Anthropic to succeed (not google they have businesses to subsidize AI). LLMs would have to be the first technology where the businesses can capture value of the underlying technology itself.
So while I would concede that affordability is vulnerable to multiple angles of attack, I think that profitability is as well.
So what makes AI different than all previous technologies?
We're a few days out here, but I don't agree that a FAANG company will come out of AI, and I think this is a fundamental error to think there will be.
FAANG is rare/hard. It feels like, we've forgotten there's this whole giant middle of like normal companies. Oracle, Cisco, all the consulting companies. Boring companies with "only" a $500B market cap. I don't think any of the AI companies will synergize the like magic set of ingredients you need to be FAANG, and that's the point, they've grown so fast they "have" to to payback investors.
But it's weird that we call them a failure if they don't hit this utopian ideal that only a few companies in history have ever hit.
This doesn't really tell you anything useful. AI companies have both built huge datacenters and raised a colossal amount of money. Include caching, quantization and etc. All of those would allow them to undercut on price considerably, even more so if you count in all the users who don't actually cap out their plans. Prices going down doesn't really tell you anything about the production cost, especially in a market where every major participant is happy to burn money just for the marketshare.
Every 6-12 months or so we get an increase in one or more of things like: compute power, compute efficiency, GPU power, GPU efficiency, network bandwidth increase, memory speed increase, component density increase in the same form factor, etc.
For awhile it was every 2-3 years you'd start a hardware refresh. As companies moved into more and more training, this timeframe started to shrink. It went from 36 months to 24 months. From 24 months to around 16-18 months. Last I checked last year, it was at 12 months. I think things may have slowed because of component availability, but otherwise whole data centers would be 6-12 months into full operations before they would start a refresh cycle.
Not to mention the massive increase in power density demand and cooling demand per rack that entails.
So no, "AI costs" have not gone down, in fact they are more expensive on training AND inference than ever.
This is why many are concerned about the heroin drip of api costs into orgs. For the companies that are public, look into their financials. It's gonna hit companies and high volume users like a ton of bricks.
There are many research avenues which are open which reduces cost dramatically. Smaller task specific/ language specific/ domain specific models, in fact they could even be better. The earlier computers were the size of a building. So prediction based on current state into the unknown future possibilites is wrong. The hardware will be all the more valuable if cheaper ways to run become possible. The hardware gets cornered in a sense.
Because of it's unpredictability and massive dependence on the training data, when LLMs start hallucinating most of the time the only fix these "engineers" have is to feed it another LLM... The genius was the transformer architecture, and evidently none of us have a damn clue how it works
The quality of what you can get from DeepSeek V4 Pro for $10 is light years ahead of what you could get for $20 a year ago.
Likewise, the quality of what I can get from a local model like Qwen 3.6 on an RTX 5090 is light years ahead of what I could get a year ago on the same hardware.
This is just hand waving on the obvious consensus that cost per capability is going down. There’s no doubt about it. Hell you can run a Gemma 4 model on your laptop that mogs GPT 4. But yeah you can use fuzziness as an excuse and ignore the trend.
I'm no economist but if true don't you have the opposite problem? How do you get people to need X many tokens per day such that you can sell enough to make money? Wouldn't you need an absence of competition for that to be ok?
the demand for intelligence is infinite. you sound like someone in 1960 wondering what the hell we would even do with the functionally infinite cpu cycles we have available to us now.
I get your point but it still like begs the question right? If you are optimistic about it all, what is the good narrative? What does it all look like? Billions upon billions of prompts to the finest models every millisecond? And then we have to scale on top of that? To like what end? How many apps do we need to code? How many questions can a single person even ask on any given day? Do I lack some imagination here?
Like what if they don't necessarily have to be super duper money making machines to legitimate how useful and nice they are for you? Is that even conceivable? What if tomorrow we all decided they are more like utilities? Would that change anything intrinsic about them for you?
