The WH has said it hasn't approved any sales, but it's not clear China is buying, and it seem they are making good progress on their huawei ascend chips. If China is basiclly at parity on the full stack (silicon, framework, training, model), and it starts open weighting frontier models at $0.xx/M tokens, then yeah, moat issues all around one would imagine? Not surprised to see Anthropic complaining like this: https://www.anthropic.com/news/detecting-and-preventing-dist... - but I don't know how you go back from it at this point?
This article is significantly better written than most anti-OpenAI/AI articles, and for that I am really grateful. I am generally an AI booster (lol), so I am happy to read well-considered thought pieces from people who disagree with me.
That being said...
> The one place where OpenAI does have a clear lead today is in the user base: it has 8-900m users. The trouble is, there’re only ‘weekly active’ users: the vast majority even of people who already know what this is and know how to use it have not made it a daily habit. Only 5% of ChatGPT users are paying, and even US teens are much more likely to use this a few times a week or less than they are to use it multiple time a day.
This really props up the whole argument, because the author goes on to say that OpenAI's users are not really engaged. But is "only" 5% of users paying of a 8-900M user base really so inconsequential? What percentage of Meta's users are paying? Google's? I would be curious to see the author dig deeper here, because I am skeptical that this is really as bad as the author suggests.
Moving on to another section:
> If the next step is those new experiences, who does that, and why would it be OpenAI? The entire tech industry is trying to invent the second step of generative AI experiences - how can you plan for it to be you? How do you compete with this chart - with every entrepreneur in Silicon Valley?
Er, are any of these startups training foundation models? No? Then maybe that is how you compete? I suppose the author would say that the foundation model isn't doing much for OpenAI's engagement metrics (and therefore revenue), but I am not sure I agree there.
Still, really good article. I think it really crystalizes the anti-OpenAI argument and it gives me a lot of interesting things to think about.
The main problem with OpenAI/Anthropic is that their only moat is their models, and it has been proven that you can clone a model through distillation. Although the performance is not exactly the same, it gets very close to the original.
People underestimate the lead OAI has with their post-5.2 models. The author does not strike me as someone who closely follows the progress frontier labs make in US and around the world.
These very valid points apply to all companies trying to make money off of proprietary models, which means margins are going to collapse in a vicious price war that will make Uber vs Lyft seem tame.
As margins collapse capex will collapse. Unfortunately valuations have become so tied to AI hype any reduction in capex will signal maybe the hype has gotten ahead of itself, meaning valuations have gotten ahead of themselves. So capex keeps escalating.
None of this takes into account the hoarding effects at play with regards to GPU acquisition. It's really a dangerous situation the industry is caught in.
Sometimes I like to imagine what this would be like if the technology had appeared 25 years ago.
First off, nonetheless open publishing stuff. Everything would have been trade secrets.
Next off no interoperable json apis instead binary APIs that are hard to integrate with and therefore sticky. Once you spent 3 or 4 months getting your MCP server setup, no way would you ever try to change to a different vendor!
The number of investors was much smaller so odds are you wouldn't have seen these crazy high salaries and you wouldn't have people running off to different companies left and right. (I know, .com boom, but the .com boom never saw 500k cash salaries...)
Imagine if Google hadn't published any papers about transformers or the attention paper had been an internal memo or heck just word2vec was only an internal library.
It has all been a net good for technological progress but not that good for the companies involved.
I keep hearing about how the app integrations will be where the AI value is and then I see the actual app integrations and they are between useless and mildly helpful.
From what I can see Anthropic's big bet is that they will solve computer use and be able to act as an autonomous agent. Not so sure how fast they will progress on that. OpenAI on the other hand - I have no idea what they are planning - all I'm reading is AI porn and ads.
Google seems to be lackluster at executing with Gemini but they are in the best position to win this whole thing - they have so much data (index of the web, youtube, maps) and so many ways to capitalize on the models - it's honestly shocking how bad they are at creating/monetizing AI products.
1) the opportunities for vertical integration are huge. Anthropic originally said they didn’t want to build IDEs, then realized the pivot to Claude Code was available to them. Likewise when one of these companies can gobble up Legal, Medical, etc why would they let companies like Harvey capture the margins?
2) oss models are 6-12 months behind the frontier because of distillation. If labs close their models the gap will widen. Once vertical integration kicks off, the distillation cost becomes higher, and the benefit of opening up generic APIs becomes lower.
I can imagine worlds where things don’t turn out this way, but I think folks are generally underrating the possibilities here.
Worth noting that it’s not a winner-takes all situation. There’s definitely space for differentiation.
Anthropic is in favor with developers and generally tech people, while OpenAi / Gemini are more commonly used by regular folks. And Grok, well, you know…
We have yet to see who’s winning in the “creative space”, probably OpenAI.
As these positionings cristallize, each company is likely going to double down on their user’s communities, like Apple did when specifically targeting creative/artsy people, instead of cranking general models that aren’t significantly better at anything.
Everyone is actually underestimating stickiness. The near billion users OpenAI has is actually a real moat and might translate into decent chunk of revenue.
My wife, for example, uses ChatGPT on a daily basis, but has found no reason to try anything else. There are no network effects for sure, but people have hundreds and thousands on conversation on these apps that can't be easily moved elsewhere. Understandable that it would be hard to get majority of these free users to pay for anything, and hence, advertising seems a good bet. You couldn't have thought of a more contextual way of plugging in a paid product.
I think OpenAI has better chance to winning on the consumer side than everyone else. Of course, would that much up against hundreds of billions of dollars in capex remains to be seen.
My nontechnical friends only know about ChatGPT, all other LLMs are a complete and total mystery to them outside of what is built into Google's search engine and Copilot. I imagine they represent the majority of consumers. It'd require significant marketing campaign for most of them to switch or for OpenAI to make a substantial mistake.
I disagree. Are people really that attached to their conversations though?
Anecdotally, the vast majority of my own conversations and coding interactions are transient in nature, to the point where I prefer to use the ‘temporary’ mode in whatever tool I’m using.
For coding, every project needs a plan and readme to get whatever agent back up to speed with what the task is. Anyone with a paid-for GH Copilot license knows that you can just switch between whatever provider at a whim, depending on the needs of your task or financial requirements.
I think people will find it easier to revert back to Siri 2.0 if that ever materialises, in which case the stickiness moat is bridged by a more familiar and widely integrated abstraction layer.
Just as people underestimate bundling and multiple-product companies. As soon as LLM corpos will start increasing prices to actually match expenses and recouping their immense debts customers will very quickly catch up how OAI product is x5 times more expensive than Google's and the only moat is is to open pre-installed Gemini :) .
Competing in freeware products is impossible as soon as monopoly emerges. Competing in paid products is way easier, especially after free money age has ended.
I never considered that. When I change LLM models its usually due to two reasons.
1. the current AI model is producing answers that do not met my needs so I try multiple others at the same time and the one that produces the best answer I stick with until I have this problem again.
2. there is a new model released and advertises a new capability that I want to try out.
I can imagine that for many people the answer that ChatGPT generates is adequate enough that they never need to try another model even if better answers exists from another model. For people with less complex needs this is a very real stickiness. Why make the effort to try something new if the answer is adequate.
In this case, OpenAI would only f*k up if they change the pricing significantly, add intrusive ads or their answers become significantly worse.
But also a billion users is ChatGPT's biggest weakness. So many free users burning compute up. So many incentives to nerf the intelligence to affordable levels. Sounds like a nightmare.
There's no stickiness per se. The thing is, all modern LLMs are good enough to answer everyday questions most people ask. Moreover, I really think most people using LLMs for more specific tasks (e.g., coding) will not be able to distinguish top LLMs in a blind test.
Tech companies are one of the jewels in America's (USA's) crown. If we build a bunch of huge AI companies, rivals will probably continue to release open AI models which undermine the US's influence in the world.
This is confirmation bias. HN and other tech people are focusing on the programming aspect of AI more than anything else. The average user does not use it for that, and they don't care. ChatGPT became something like Kleenex.
> The models have a very large user base, but very narrow engagement and stickiness, and no network effect or any other winner-takes-all effect so far that provides a clear path to turning that user base into something broader and durable.
I think this is clearly wrong. Users provide lots of data useful for making the models better and that is already being leveraged today. It seems like network effects are likely in the future too. And they have several ways to get stickiness including memory.
These sorts of doom articles are interesting in that they are from the perspective of tech company valuations. Why is this the important perspective?
For the humanity perspective, this doom is very optimistic. It says that these LLMs currently disrupting the platforms cannot themselves be the next platforms.
Maybe no one will have 'the ability to make people do something that they don't want to do' sort of power with this next stage in computing.
I speak native English and barebones high school Spanish. I recently visited Costa Rica and almost every time there was a language barrier issue (unknown word or phrase), the local folks opened ChatGPT, said what they were trying to say in Spanish and then had ChatGPT convert it to English. It was everywhere.
If you were forced to choose just one of all the competing players, which is "the one" you will use?
For me, the choice is ChatGPT, not for its Codex or other fancy tooling - just the chat. Not that Claude Code or Cowork is less important. Not that I like Codex over Claude Code.
OpenAI lost the race to nerds' hearts. In the latest benchmarks, OpenAI is simultaneously cheaper (like 50% less?) and scores hire in coding and tool use benchmarks (GPT-5.3-Codex trounces Opus 4.6), yet all the coders want to marry Anthropic. I don't think OpenAI understands how to sell, if they even had a product to sell.
I would love to dunk on this or something, but the lesson is that it's all about distribution.
Sama is really good at that, and also.. gotta give props for a lot of forward thinking like the orb, which now makes a lot of sense to me, as non-Apple/Google proof of personhood.
I think this is the best article on open AI that I've ever read. A lot of content these days will try to paint OpenAI in sensational ways that really doesn't get to the bottom of whether open AI has an economic mode, and this article does a very thorough job of explaining why OpenAI doesn't have power like the other platforms.
And so this goes back to my theory that open AI's execution is basically to get it itself in a position where the market cannot afford to have it implode. Basically, it wants to or it needs to be too big to fail. And I think we're already kind of seeing the politicization, if you will, sort of the rocket race between two superpowers or large powers on the AI front, and I think that Might be a viable strategy.
Yeah, I was a bit surprised that the author didn't mention that facet of Open AI. It did mention infrastructure goals, but the reality is that Open AI's infrastructure spending commitments have inflated the stock prices of quite a few hardware companies (like Micron and WD) and caused a strain on the market.
The real danger here is how over-leveraged Open AI is. No other AI player is as exposed. Their massive spending commitments are all precariously balanced on the other end by their user base, and if that evaporates, the whole thing will fall apart and that could crash the stocks of other players ...and by crash, I mean bring them down to a realistic value. But the economy is counting on this to work, which is why I believe that Open AI's strategy here really is to make the market exposed to Open AI's risks.
80 comments
[ 3.1 ms ] story [ 68.0 ms ] threadThe WH has said it hasn't approved any sales, but it's not clear China is buying, and it seem they are making good progress on their huawei ascend chips. If China is basiclly at parity on the full stack (silicon, framework, training, model), and it starts open weighting frontier models at $0.xx/M tokens, then yeah, moat issues all around one would imagine? Not surprised to see Anthropic complaining like this: https://www.anthropic.com/news/detecting-and-preventing-dist... - but I don't know how you go back from it at this point?
That being said...
> The one place where OpenAI does have a clear lead today is in the user base: it has 8-900m users. The trouble is, there’re only ‘weekly active’ users: the vast majority even of people who already know what this is and know how to use it have not made it a daily habit. Only 5% of ChatGPT users are paying, and even US teens are much more likely to use this a few times a week or less than they are to use it multiple time a day.
This really props up the whole argument, because the author goes on to say that OpenAI's users are not really engaged. But is "only" 5% of users paying of a 8-900M user base really so inconsequential? What percentage of Meta's users are paying? Google's? I would be curious to see the author dig deeper here, because I am skeptical that this is really as bad as the author suggests.
Moving on to another section:
> If the next step is those new experiences, who does that, and why would it be OpenAI? The entire tech industry is trying to invent the second step of generative AI experiences - how can you plan for it to be you? How do you compete with this chart - with every entrepreneur in Silicon Valley?
Er, are any of these startups training foundation models? No? Then maybe that is how you compete? I suppose the author would say that the foundation model isn't doing much for OpenAI's engagement metrics (and therefore revenue), but I am not sure I agree there.
Still, really good article. I think it really crystalizes the anti-OpenAI argument and it gives me a lot of interesting things to think about.
As margins collapse capex will collapse. Unfortunately valuations have become so tied to AI hype any reduction in capex will signal maybe the hype has gotten ahead of itself, meaning valuations have gotten ahead of themselves. So capex keeps escalating.
None of this takes into account the hoarding effects at play with regards to GPU acquisition. It's really a dangerous situation the industry is caught in.
First off, nonetheless open publishing stuff. Everything would have been trade secrets.
Next off no interoperable json apis instead binary APIs that are hard to integrate with and therefore sticky. Once you spent 3 or 4 months getting your MCP server setup, no way would you ever try to change to a different vendor!
The number of investors was much smaller so odds are you wouldn't have seen these crazy high salaries and you wouldn't have people running off to different companies left and right. (I know, .com boom, but the .com boom never saw 500k cash salaries...)
Imagine if Google hadn't published any papers about transformers or the attention paper had been an internal memo or heck just word2vec was only an internal library.
It has all been a net good for technological progress but not that good for the companies involved.
From what I can see Anthropic's big bet is that they will solve computer use and be able to act as an autonomous agent. Not so sure how fast they will progress on that. OpenAI on the other hand - I have no idea what they are planning - all I'm reading is AI porn and ads.
Google seems to be lackluster at executing with Gemini but they are in the best position to win this whole thing - they have so much data (index of the web, youtube, maps) and so many ways to capitalize on the models - it's honestly shocking how bad they are at creating/monetizing AI products.
1) the opportunities for vertical integration are huge. Anthropic originally said they didn’t want to build IDEs, then realized the pivot to Claude Code was available to them. Likewise when one of these companies can gobble up Legal, Medical, etc why would they let companies like Harvey capture the margins?
2) oss models are 6-12 months behind the frontier because of distillation. If labs close their models the gap will widen. Once vertical integration kicks off, the distillation cost becomes higher, and the benefit of opening up generic APIs becomes lower.
I can imagine worlds where things don’t turn out this way, but I think folks are generally underrating the possibilities here.
Anthropic is in favor with developers and generally tech people, while OpenAi / Gemini are more commonly used by regular folks. And Grok, well, you know…
We have yet to see who’s winning in the “creative space”, probably OpenAI.
As these positionings cristallize, each company is likely going to double down on their user’s communities, like Apple did when specifically targeting creative/artsy people, instead of cranking general models that aren’t significantly better at anything.
My wife, for example, uses ChatGPT on a daily basis, but has found no reason to try anything else. There are no network effects for sure, but people have hundreds and thousands on conversation on these apps that can't be easily moved elsewhere. Understandable that it would be hard to get majority of these free users to pay for anything, and hence, advertising seems a good bet. You couldn't have thought of a more contextual way of plugging in a paid product.
I think OpenAI has better chance to winning on the consumer side than everyone else. Of course, would that much up against hundreds of billions of dollars in capex remains to be seen.
Anecdotally, the vast majority of my own conversations and coding interactions are transient in nature, to the point where I prefer to use the ‘temporary’ mode in whatever tool I’m using.
For coding, every project needs a plan and readme to get whatever agent back up to speed with what the task is. Anyone with a paid-for GH Copilot license knows that you can just switch between whatever provider at a whim, depending on the needs of your task or financial requirements.
I think people will find it easier to revert back to Siri 2.0 if that ever materialises, in which case the stickiness moat is bridged by a more familiar and widely integrated abstraction layer.
Competing in freeware products is impossible as soon as monopoly emerges. Competing in paid products is way easier, especially after free money age has ended.
1. the current AI model is producing answers that do not met my needs so I try multiple others at the same time and the one that produces the best answer I stick with until I have this problem again.
2. there is a new model released and advertises a new capability that I want to try out.
I can imagine that for many people the answer that ChatGPT generates is adequate enough that they never need to try another model even if better answers exists from another model. For people with less complex needs this is a very real stickiness. Why make the effort to try something new if the answer is adequate.
In this case, OpenAI would only f*k up if they change the pricing significantly, add intrusive ads or their answers become significantly worse.
I think this is clearly wrong. Users provide lots of data useful for making the models better and that is already being leveraged today. It seems like network effects are likely in the future too. And they have several ways to get stickiness including memory.
For the humanity perspective, this doom is very optimistic. It says that these LLMs currently disrupting the platforms cannot themselves be the next platforms.
Maybe no one will have 'the ability to make people do something that they don't want to do' sort of power with this next stage in computing.
Sounds good to me.
Personally I only see Google (Gemini), X (Grok) and the Chinese models having a chances to still be alive in 1-2 years.
For me, the choice is ChatGPT, not for its Codex or other fancy tooling - just the chat. Not that Claude Code or Cowork is less important. Not that I like Codex over Claude Code.
I would love to dunk on this or something, but the lesson is that it's all about distribution.
Sama is really good at that, and also.. gotta give props for a lot of forward thinking like the orb, which now makes a lot of sense to me, as non-Apple/Google proof of personhood.
And so this goes back to my theory that open AI's execution is basically to get it itself in a position where the market cannot afford to have it implode. Basically, it wants to or it needs to be too big to fail. And I think we're already kind of seeing the politicization, if you will, sort of the rocket race between two superpowers or large powers on the AI front, and I think that Might be a viable strategy.
The real danger here is how over-leveraged Open AI is. No other AI player is as exposed. Their massive spending commitments are all precariously balanced on the other end by their user base, and if that evaporates, the whole thing will fall apart and that could crash the stocks of other players ...and by crash, I mean bring them down to a realistic value. But the economy is counting on this to work, which is why I believe that Open AI's strategy here really is to make the market exposed to Open AI's risks.