I find it odd that investors are still so motivated to drop money into AI, specifically LLMs, not because I don't see value in the technology, I do, but because it feels to me like it only takes a few months (if that) for your company which might currently dominate with the leading product, to be left behind by someone else who just came out with a better, cheaper, and faster product.
I guess this is true for most tech, but with LLMs it seems to be at a pace that just doesn't feel worth it to invest knowing someone else will just leap over your investment soon, and that in a year or two, when you want some return (maybe longer), that your tech might have completely stagnated and completely left behind by everyone else's. I dunno, just a weird thought.
I used to not understand as well, but then a friend asked me a question: "When Claude recommends a book, why doesn't it just link directly to Amazon and get a referral fee?"
I believe what investor's are chasing in LLMs is the same thing that Zuckerberg was chasing in the metaverse: not a new "technology" a new medium. The key insight in the development of the web was that a few companies can effectively control the entire thing (at it's peak both FB and Google basically had their tendrils on every page out there).
Apple's mobile play showed that their seeing the value in the fact that mobile was a brand new medium early on. This lead them to having control over a market that Google had some play in, and Facebook really couldn't get a hold on.
LLMs represent a new opportunity. The answer to my friend's question is very likely: "Because they eventually want to sell you the book through them". Sure there's not much of a moat, there wasn't much of a social media moat back in the day either, but eventually there will be a few winners and those winners will divide what may very well be the next "mobile" or even "web".
I already spend more and more of my online time chatting with Claude about various thing, most of the books I've ordered in the last few months came from conversations with Claude. I can absolutely imagine a world where AI is how I accomplish most of my purchasing, precisely because right now AI is not rotten with SEO spam. That will of course change, but likely not in exactly the same way the web changed.
There's a very likely future where Claude/ChatGPT/DeepSeek become iPhone/Android/Huawei, what investors are hoping is that they can make it Claude/XAI/DeepSeek (or some variant), because owning a piece of that can effectively be even bigger than owning a large amount of stock in early Amazon.
> I used to not understand as well, but then a friend asked me a question: "When Claude recommends a book, why doesn't it just link directly to Amazon and get a referral fee?"
How much do I pay to make it recommend my book, rather than your book?
Interesting enough my friend's background was in SEO so this topic also came up, and currently one of my books does show up immediately when asked about the subject!
The early days of Google felt very similar to the current state of LLMs. The web at the time was filled with the laziest keyword spam, search results where awful. Then Google changed all that (hard to even imagine today): you nearly always got the results you wanted without garbage. Early Google fought spam pretty aggressively, old time HNers might recall when the Genius lyric site was caught gaming the system (more aggressively than standard SEO) and was permanently punished in Google results. Similarly early Facebook was actually pretty pleasant, just a way to catch up with old high school classmates, I don't even remember ads in the early versions.
My guess is LLMs will follow a very similar path, but because it is a new medium it won't evolve in exactly the same way. I suspect once the major players are established there will be no more SEO, but rather major advertising agreements behind the scenes. One nice thing about SEO is that anyone could do it, sadly I think we'll soon be back in a world where only big players really matter.
To be clear, I don't have any fantasies that "this time it will be better!" I'm pretty certain we'll see the same enshitification cycle play out, but I'm trying to enjoy at least being on the less-shitty part of the cycle for a moment.
Yeah but that last paragraph is the crux of it: No one is using Grok. They have no path to profitability. They have very low traction among consumers, and even lower traction among enterprises. Their CEO seems to think that copy-pasting code into a web browser is a good strategy for attracting programmers. Its most common usage is on X where people say "@grok is this true", which is fine, except its unclear if anyone is actually paying for that. Over the past week, OpenRouter has been processing ~125M x-ai/grok-3 tokens per day; compare that to anthropic/claude-sonnet-4 @ ~45B/day and openai/gpt-4.1-mini @ ~5B/day.
Yeah its only one somewhat specialized source, but its a big one. The only time anyone talks about Grok or xAI is their now-annual accomplishment of taking the intelligence crown, for about a month, before OpenAI comes out and beats them, then Anthropic comes out and beats them, then Gemini comes out and beats them, then DeepSeek crashes the stock market because china scary, then we repeat again. These things are so damn fungible now that it startles me that any investors would touch any of these companies at their current valuations.
The future where Claude/ChatGPT/DeepSeek become the next iPhone/Android/Huawei is only going to happen if these companies make hardware and traditional software operating systems. OpenAI is trying. But, by the way, one of those companies on the right side is already doing that and has a world-leading AI strategy (Google), so I'm not convinced the status quo is as disruptable as some people think.
I wonder to what extent it matters… but, Musk has mostly lead companies where he comes with a new competitor against incredibly stagnant competition, right? (banking, cars, rockets). He seems to have a strategy of convincing some young engineers to work really hard on a neat new idea. That, combined with the stagnant competition, produces a sudden dramatic leap past the competition that gets people interested.
LLMs were already the hot-shit new tech that was attracting a ton of excited talent. What’s xAI got to offer that OpenAI doesn’t? Training off a mediocre social media network I guess.
Yea, I don’t think investors really internalize that LLM companies have effectively zero moat.
Customers can easily jump ship. Everyone is using effectively the same compute, nearly identical training data, very similar algorithms, etc. Worse past investments in a model become outdated quickly.
I think Hacker News readers underestimate how sticky the typical customer is.
The average subscriber to these services isn’t following a constant stream of LLM news and jumping from platform to platform every time the leaderboard changes slightly. The average customer signs up and then inertia and familiarity take over. They stay signed and comfortable.
Yeah, I think grandparent has a point, but at the same time, Google Search also didn't have a strong moat, or at least not one that couldn't be replicated in a few years, and it still positively dominated the market for decades.
When the price is $0, the competitor has to be substantially better for consumers to switch. Bing reached quality comparable to Google, but never noticeably exceeded it. So people didn't switch. OTOH, the switch from Netscape -> IE6 -> Firefox -> Chrome did happen.
But when the price is non-zero, that gives another dimension to compete on, and monopolies without significant barriers are much less common. It looks like good LLM's are going to have a price tag, so a monopoly is going to need a moat.
Dominated market sure, able to charge any kind of subscription fee no.
If winning the LLM’s game long term means the product needs to compete on cost and be the best to win, that’s not going to be nearly as profitable in the near term.
In terms of AI coding assistant, almost every product out there allows you switch to a different model with a single click. If Gemini is not giving you the answer you want, what do you do? Switch to Claude and bang you have it.
The only thing that has some stickiness is products like ChatGPT. But that's so much more than just a model -- it's a product with many features smoothly integrated.
xAI does not currently have any product that is sticky for the average user.
Is the average consumer stuck to an LLM provider at this point, though? I guess the first one that normal people get attached to will be the one that gets into the Windows Start Menu, Siri, or whatever the Android equivalent is.
Machines with LLM acceleration hardware are barely coming out at this point. I think it is early to expect people to really be stuck to anything.
I'm astonished at the stickiness of chatgpt though. It's clearly not the best platform/model but all my none tech friends just equate LLMs = ChatGPT (4o as well).
o3/o3-pro is probably the best model, or very close to being the best model, overall. It beats Grok 4 in writing/composition and analysis tasks. [1] Performance is a toss-up between o3 and Claude 4 Opus, but I find o3 easier to interact with and more trustworthy. (Less likely to push back against requests and more likely to attempt to fulfill them in good faith.)
4.5 is also great for certain things. Of all models, it's the second best writer. (DeepSeek R1 is the best prose stylist, surprisingly!)
It just benefitted from the Kleenex/Google effect due to being the first real mainstream breakthrough of GPT tech among average users ("normies").
I expect, whether OpenAI is even still around in 15+ years or not, that my grandparents will continue telling me about "that ChatGPT" well into the future.
Doesn't that suggest a 200B valuation in xAI is stupid, though? Nobody is using Grok except Elon's fanboys, where do they imagine these sticky customers are going to come from?
Please check google trends in regards to Grok (on a worldwide scale, and compare to openai, claude and deepseek). The winner is not clear at all. And grok is surprisingly popular.
>No one is using Grok. They have no path to profitability. They have very low traction among consumers, and even lower traction among enterprises. Their CEO seems to think that copy-pasting code into a web browser is a good strategy for attracting programmers.
Leaving aside the controversies, Elon is one of the best players at capitalism there has ever been. As per his wealth. And to think you and I know more than someone with a deep network of Investors and fellow CEO's would be a bold claim.
The other thing that would scare me: What percentage of Anthropic's revenue comes from Anysphere/Cursor? Think about that for a second: Given that we know Cursor is on a nine-figure ARR now, the intrinsic value in being the "default" model in Cursor, similar to Google being the default search on iOS, is now worth billions of dollars. Does Anysphere have significant exclusivity arrangements with any of these frontier model providers? They took funding from OpenAI. The ludicrous clip at which Anthropic's revenue could drop by whole percentages just because Anysphere pushes a change to prefer competitors models, has got to be the scariest business arrangement in the history of computing.
None of these companies will survive if they don't own multiple parts of the stack. As deep as possible, to the hardware if they can. That's why Anthropic is investing heavily in Claude Code, and why OpenAI is looking at hardware. You can't just build models and sell tokens, yet that seems to be Grok's entire strategy right now [1]
B2C might not be sticky, but enterprises can't easily migrate and it takes a quite long cycle of decision making to replace one product with another, in this case LLM.
I think everyone is betting on making a good product and convince enterprises to buy from them, this is where you can get sticky customers
Microsoft really put this notion under a spotlight with how they deal with LLMs in VSCode Copilot. The LLMs are just a drop-down menu that you can switch between at any moment, on a whim. They all plug into the same toolchains and are essentially generic commodities. They have slightly different prices for usage but that's about it, their outputs aren't wildly different. On topic for this post: it's very telling that Grok models aren't available in Copilot and nobody cares.
I think this arrangement is probably the future of LLM usage, and it does not bode well for everyone betting the farm that their model will be special.
Search engine are as well. At least in the eu they literally are the first thing a browser asks you when you launch it for the first time.
On a dropdown you have 2 clicks to do to change. For search engine you are prompted to using Google is one click; using something else costs the same click.
Google is still managing an outrageous domination.
Being percieved as the best is still a huge headstart. 99% of the population is not using “last weeks llm that topped AIME and ARC-AGI“. They are using “ChatGPT” with the default model selected.
People are going to switch when “their tech friend tells them to switch”. The same way they switched from internet explorer to chrome.
Once you reach that position you can afford being “not the best but good enough” for a long time.
xAI needs to convince investors that: OpenAI is struggling so there is an opening to take the crown and be popular enough to get people to switch. They have twitter to help make that happen.
And they need to convince that no one else is going so much better than they are anytime soon; they just need to be good enough
> They are using “ChatGPT” with the default model selected.
Funny, because the OpenAI models are not the first or even second choice for anyone I know who uses Copilot for coding agents. Anthropic and Google are absolutely stomping them in this space.
Search engines are pretty different from LLMs, in any case: they all have different UX right in your face, different functionality, etc. The LLMs simply generate text, that's all they do, and the differences are far more subtle.
Your sample are tech people using Copilot, which is very small population sample. Hundreds of millions of casual users around the world default to ChatGPT, and for example in my country, it's basically household name at this point. They haven't even heard about Claude or Gemini. For Google, it looks like like Google+ vs Facebook situation all over again.
It's the same with DuckDuckGo's search and Cursor, the models are interchangeable and there's no "walled garden", none of them can assert any control or restrictions on others, it's a strict meritocracy.
> I find it odd that investors are still so motivated to drop money into AI, specifically LLMs...
Me too, but what shocks me the most are the insane amounts of money that are being moved. Does the tech really cost or has that value? Why always the ridiculous amounts?
Say you're trying to value a 1% stake in a leading AI company. You use a standard "expected value" formula to calculate the current value of the investment opportunity. For each possible outcome, you take the dollar value of that outcome, multiply it times the likelihood of the outcome, then sum the values of all possible outcomes. With this type of calculation, possible outcomes which are very unlikely can still be extremely valuable if the dollar value of that outcome is high enough to compensate for the low probability of success.
Now imagine you believe that this AI company might eventually deliver AGI, or even a Kurzweilian superintelligence. It would be easy to justify a near-infinite dollar value on that best-case outcome. And, of course, a possible outcome with infinite value confers infinite value on the investment as a whole, as long as its value is non-zero.
What is their product though? It will turn into surveillance or military applications, which would explain the interest of the Saudi sovereign wealth fund.
When you spend a lot of time working with a model, you probably start developing a "relationship" with it. You know how to describe things, what to expect. The model has learned your preferences. I don't think I will be jumping from model to another just like that. Not at least to save a few bucks on monthly fees.
For more generic AI it won't be just about the models. It's also about access to data. Everybody is closing the gates to content. xAI is in interesting position because they have all the data in X. X is great source when you want to know what is happening in the world. And they have also a deal with Telegram (although I'm not sure to what data they can access).
And even if the models become commodity, I don't think they will be worthless. Plenty of companies make big money on selling commodities. If AI delivers, selling access to the models should be a business opportunity of the century. Even if margins would be low, the business will be huge and infinitely scalable.
XAI is valued not for its intelligence, but for its practical potential—especially in robotics and autonomous vehicles. In robotics, XAI helps ensure machines behave predictably in complex environments. In cars, especially self-driving systems, explainability is key to trust, regulatory approval, and user adoption. Intelligence without interpretability is a liability in these fields.
If xAI can pull this off, it holds massive potential for passive income and a large user base—even if it's not the best LLM. Success in integrating with robotics and vehicles could secure long-term contracts, licensing deals, and infrastructure partnerships. The market rewards utility, not just raw intelligence.
Shaming those who profit off Nazi ideology seems wise for USA.
For operating in Germany I'd imagine this would be a more difficult sell, a German branch of it, I wonder if it would be restricted based upon the Nazi behavior of Musk.
(Child of Musk commented how it was definitely a Nazi salute)
It's so hard to decide who to believe in this dispute. FT with multiple sources; or, a guy who's committed securities fraud in the past lying about this exact subject matter.
I guess this is true for most tech, but with LLMs it seems to be at a pace that just doesn't feel worth it to invest knowing someone else will just leap over your investment soon, and that in a year or two, when you want some return (maybe longer), that your tech might have completely stagnated and completely left behind by everyone else's. I dunno, just a weird thought.
I believe what investor's are chasing in LLMs is the same thing that Zuckerberg was chasing in the metaverse: not a new "technology" a new medium. The key insight in the development of the web was that a few companies can effectively control the entire thing (at it's peak both FB and Google basically had their tendrils on every page out there).
Apple's mobile play showed that their seeing the value in the fact that mobile was a brand new medium early on. This lead them to having control over a market that Google had some play in, and Facebook really couldn't get a hold on.
LLMs represent a new opportunity. The answer to my friend's question is very likely: "Because they eventually want to sell you the book through them". Sure there's not much of a moat, there wasn't much of a social media moat back in the day either, but eventually there will be a few winners and those winners will divide what may very well be the next "mobile" or even "web".
I already spend more and more of my online time chatting with Claude about various thing, most of the books I've ordered in the last few months came from conversations with Claude. I can absolutely imagine a world where AI is how I accomplish most of my purchasing, precisely because right now AI is not rotten with SEO spam. That will of course change, but likely not in exactly the same way the web changed.
There's a very likely future where Claude/ChatGPT/DeepSeek become iPhone/Android/Huawei, what investors are hoping is that they can make it Claude/XAI/DeepSeek (or some variant), because owning a piece of that can effectively be even bigger than owning a large amount of stock in early Amazon.
How much do I pay to make it recommend my book, rather than your book?
The early days of Google felt very similar to the current state of LLMs. The web at the time was filled with the laziest keyword spam, search results where awful. Then Google changed all that (hard to even imagine today): you nearly always got the results you wanted without garbage. Early Google fought spam pretty aggressively, old time HNers might recall when the Genius lyric site was caught gaming the system (more aggressively than standard SEO) and was permanently punished in Google results. Similarly early Facebook was actually pretty pleasant, just a way to catch up with old high school classmates, I don't even remember ads in the early versions.
My guess is LLMs will follow a very similar path, but because it is a new medium it won't evolve in exactly the same way. I suspect once the major players are established there will be no more SEO, but rather major advertising agreements behind the scenes. One nice thing about SEO is that anyone could do it, sadly I think we'll soon be back in a world where only big players really matter.
To be clear, I don't have any fantasies that "this time it will be better!" I'm pretty certain we'll see the same enshitification cycle play out, but I'm trying to enjoy at least being on the less-shitty part of the cycle for a moment.
Kagi gives that feeling too and is much faster than any chat bot there is…
Yeah its only one somewhat specialized source, but its a big one. The only time anyone talks about Grok or xAI is their now-annual accomplishment of taking the intelligence crown, for about a month, before OpenAI comes out and beats them, then Anthropic comes out and beats them, then Gemini comes out and beats them, then DeepSeek crashes the stock market because china scary, then we repeat again. These things are so damn fungible now that it startles me that any investors would touch any of these companies at their current valuations.
The future where Claude/ChatGPT/DeepSeek become the next iPhone/Android/Huawei is only going to happen if these companies make hardware and traditional software operating systems. OpenAI is trying. But, by the way, one of those companies on the right side is already doing that and has a world-leading AI strategy (Google), so I'm not convinced the status quo is as disruptable as some people think.
LLMs were already the hot-shit new tech that was attracting a ton of excited talent. What’s xAI got to offer that OpenAI doesn’t? Training off a mediocre social media network I guess.
Customers can easily jump ship. Everyone is using effectively the same compute, nearly identical training data, very similar algorithms, etc. Worse past investments in a model become outdated quickly.
The average subscriber to these services isn’t following a constant stream of LLM news and jumping from platform to platform every time the leaderboard changes slightly. The average customer signs up and then inertia and familiarity take over. They stay signed and comfortable.
But when the price is non-zero, that gives another dimension to compete on, and monopolies without significant barriers are much less common. It looks like good LLM's are going to have a price tag, so a monopoly is going to need a moat.
If winning the LLM’s game long term means the product needs to compete on cost and be the best to win, that’s not going to be nearly as profitable in the near term.
In terms of AI coding assistant, almost every product out there allows you switch to a different model with a single click. If Gemini is not giving you the answer you want, what do you do? Switch to Claude and bang you have it.
The only thing that has some stickiness is products like ChatGPT. But that's so much more than just a model -- it's a product with many features smoothly integrated.
xAI does not currently have any product that is sticky for the average user.
Err... What about twitter? Where it is firmly embedded. And Telegram recently has been paid to do the same.
Machines with LLM acceleration hardware are barely coming out at this point. I think it is early to expect people to really be stuck to anything.
It explains OpenAI’s valuation but no one else
4.5 is also great for certain things. Of all models, it's the second best writer. (DeepSeek R1 is the best prose stylist, surprisingly!)
[1] - This is Grok 4: https://x.com/i/grok/share/e51O9rK0W7UaIN81nFBQoJDSs
This is o3-pro, same question: https://chatgpt.com/s/t_68710185cf04819185dc25233280e46b
o3 made fewer mistakes and drafted a more neatly structured and better written output.
I expect, whether OpenAI is even still around in 15+ years or not, that my grandparents will continue telling me about "that ChatGPT" well into the future.
https://imgur.com/a/a5oUldD
https://arstechnica.com/tech-policy/2025/07/xai-gets-an-air-...
I wonder how investors have priced in their regulatory situation.
They bought a cargo ship worth of GPUs though
OpenAI also announced their own web browser too.
Leaving aside the controversies, Elon is one of the best players at capitalism there has ever been. As per his wealth. And to think you and I know more than someone with a deep network of Investors and fellow CEO's would be a bold claim.
None of these companies will survive if they don't own multiple parts of the stack. As deep as possible, to the hardware if they can. That's why Anthropic is investing heavily in Claude Code, and why OpenAI is looking at hardware. You can't just build models and sell tokens, yet that seems to be Grok's entire strategy right now [1]
[1] https://x.com/elonmusk/status/1943178423947661609
I think everyone is betting on making a good product and convince enterprises to buy from them, this is where you can get sticky customers
I think this arrangement is probably the future of LLM usage, and it does not bode well for everyone betting the farm that their model will be special.
Google is still managing an outrageous domination.
Being percieved as the best is still a huge headstart. 99% of the population is not using “last weeks llm that topped AIME and ARC-AGI“. They are using “ChatGPT” with the default model selected.
People are going to switch when “their tech friend tells them to switch”. The same way they switched from internet explorer to chrome. Once you reach that position you can afford being “not the best but good enough” for a long time.
xAI needs to convince investors that: OpenAI is struggling so there is an opening to take the crown and be popular enough to get people to switch. They have twitter to help make that happen.
And they need to convince that no one else is going so much better than they are anytime soon; they just need to be good enough
Funny, because the OpenAI models are not the first or even second choice for anyone I know who uses Copilot for coding agents. Anthropic and Google are absolutely stomping them in this space.
Search engines are pretty different from LLMs, in any case: they all have different UX right in your face, different functionality, etc. The LLMs simply generate text, that's all they do, and the differences are far more subtle.
Me too, but what shocks me the most are the insane amounts of money that are being moved. Does the tech really cost or has that value? Why always the ridiculous amounts?
Too much money with no place to go?
With Elon companies investors hope that the market will behave just as irrationally as with Tesla.
With other companies they hope that they will be bought up by bigtech.
Now imagine you believe that this AI company might eventually deliver AGI, or even a Kurzweilian superintelligence. It would be easy to justify a near-infinite dollar value on that best-case outcome. And, of course, a possible outcome with infinite value confers infinite value on the investment as a whole, as long as its value is non-zero.
xAI > LLM
mission > product
For more generic AI it won't be just about the models. It's also about access to data. Everybody is closing the gates to content. xAI is in interesting position because they have all the data in X. X is great source when you want to know what is happening in the world. And they have also a deal with Telegram (although I'm not sure to what data they can access).
And even if the models become commodity, I don't think they will be worthless. Plenty of companies make big money on selling commodities. If AI delivers, selling access to the models should be a business opportunity of the century. Even if margins would be low, the business will be huge and infinitely scalable.
If xAI can pull this off, it holds massive potential for passive income and a large user base—even if it's not the best LLM. Success in integrating with robotics and vehicles could secure long-term contracts, licensing deals, and infrastructure partnerships. The market rewards utility, not just raw intelligence.
Nazi in charge is a huge detriment IMO.
Shaming those who profit off Nazi ideology seems wise for USA.
For operating in Germany I'd imagine this would be a more difficult sell, a German branch of it, I wonder if it would be restricted based upon the Nazi behavior of Musk.
(Child of Musk commented how it was definitely a Nazi salute)
> These rumors are false. xAI is not seeking funding right now. We have plenty of capital.