I think it's unquestionably right that these companies can't all win, and those that don't win are going to burn a lot of money for nothing. However there's kind of two directions this can go: Compute gets cheaper, in which case there's no monopoly it'll be easy for many companies to make good models and there won't be pricing power on serving a good model. The other case is compute gets cheaper but we keep using more and more of it, so it does likely become winner take all. The first scenario is good for the economy but likely bad for the returns on these AI stocks. The second is maybe bad for the economy and maybe not even good for the winner.
Take Google or Meta: Today Google makes a shit-tonne of money and to make that money they need to run some servers. The servers are extremely cheap relatively to the revenue they make running the business. This makes them a very attractive stock - the core of why SAAS looks great. Now let's assume the monopoly path. Google can win. I think they likely will win. But now they're going to spending... how many hundreds of billions constantly training new models? The cost of providing the service suddenly isn't small relative revenue they're getting. So even for them it looks awful for their valuation.
> Fewer people applying for patents, because the minute you apply for the patent, it's available to everybody, which means every model can train on it
We know LLM companies have, for lack of a better word, "sidestepped" the copyright on millions of works with their "transformative fair use" arguments. Are LLMs also a way to sidestep patents?
I don’t see how? You can train on something pending patent, but what are the benefits? If it gets patented you open yourself to get sued, I don’t see how AI works around that, the idea itself is still patented? I think I’m missing something for the argument to make sense. Or is the idea that if too many people use your patented idea you won’t be able to enforce it? That sounds risky to me
LLMs are accelerants. They enable people to do patent and copyright infringement at a much larger scale. As we know from previous examples, if you break the law enough as a company eventually they have to let you keep doing it.
He's right, there is a race. It's going to be a natural monopoly or duopoly because the cost to train the next SOTA model is always increasing. I can see that there are only 3 companies competing for the duopoly or monopoly realistically: OpenAI, Anthropic, and Google. Everyone else has fallen behind. The flywheel of generate more revenue, get more data, get more compute train a better model might already be too great to overcome for anyone else.
I don't understand why he thinks OpenAI can't be one of the duopolies or become the monopoly. OpenAI's models are always the first or second best overall - usually the first. They are also leading in the consumer market by a wide margin. They also made a strategic decision that is paying off which was committing to more compute early on while Anthropic is hammered by the lack of compute.
> I can see that there are only 3 companies competing for the duopoly or monopoly realistically: OpenAI, Anthropic, and Google.
I could see people saying this in 2022, but now? No chance.
Chinese models keep demonstrating that SOTA can be approximated for a fraction of the cost. The innovation out of these companies keep showing diminishing returns, with a greater emphasis on the tooling and application layer. Having the right workflow with the right data is more important than having the right model. We could freeze AI now, and I'd bet good money that the current state of things is good enough to - not be first - but competitive for the next few years.
Even if we do end up with a oligopoly situaiton, it'll be less like Microsoft in the 90s and more like Microsoft now where they just give out windows for free, have support for WSL and the focus is on cloud services rather than their OS.
I think the performance of models is only one aspect. You have to take in account the cash flow, how much spending commitment the different actors have, debt, etc. OpenAI has taken some very risky commitments, of they don’t get the revenue to cover their expenses in the next few years their situation will be pretty bad
How can this become a monopoly/duopoly? There is no moat, the Chinese providers will continue to hunt the market leader at 10% of the price, there is no network effect (OpenAI's Sora was a play in that direction and failed).
I'm constantly amazed how this AGI/monopoly narrative can be kept up so long in the West, it just doesn't make sense (unless the state creates said monopoly by forbidding competition).
That's not what "moat" means. Claude Code has a castle. A "moat" is meant to protect the castle from invaders. It would be things like high switching costs, proprietary formats, network effects, etc. that aren't there.
In other comments people mention the "flywheel" of data and money feeding training, but there's a view that at some point the baseline open-weight models are "good enough" that the money will dry up.
> Open-weight models aren't going to be free forever.
The ones that are already released are, and they're already very good for most purposes and can be fine-tuned indefinitely, includin months or years down the line when processes have been optimized and things aren't as compute-heavy as they are now.
Yeah, no, i disagree. Frontier models were almost untouchable 6 month ago, but now i can get 90% of Opus 4.5 with any chineese model, or even with Mistral. The only thing i'm missing is the chain of thought that help me understand the "how" and "why" when AI fails at its task. For the "general purpose" AI, it's even worse, any free model i can run on my Intel Arc (yes, sorry, it was discounted an very cheap) i get like 80% of a frontier model, at virtually no cost, and i suppose Deepseek/Mistral are like 95% there.
Claude is kicking ass in coding but it seems like Codex is catching up fast. Claude Code's PR has taken a hit recently due to the lack of compute forcing Anthropic to dumb down the models. Codex has been gaining momentum.
Chip manufacturing aren't really differentiated either - it didn't stop TSMC from becoming the monopoly for high end chip nodes, capturing 90%+ of the advanced chip market. The reason they have is because Rock's Law makes it too expensive to build the next node unless you've generated enough revenue from the current node. I don't see why it isn't the same for SOTA models.
Chip manufacturing is insanely hard, it requires know-how, that's the moat. It's not money, otherwise the EU and China would have leading edge fabs.
Machine learning has no real moat. There's no network effect, it's not hard (you can just throw money at the problem). It's not data, because we have an existence proof that general intelligence can be trained by a few humans and a shelf full of books. The compute to do it is generally available. As soon as one organization releases open weights, everyone can use it immediately, even on modest local hardware.
I understand your argument, but I think you might be overestimating the intent of the users when they're using chatgpt.
The ones killing on ads are Google, Meta, and Amazon.
I just don't see how ChatGPT will gobble those market shares - ads are increasingly tied to sales attribution, and it would require a complete shift of the market for ChatGPT to take over the role of those 3 players.
People will still try to look for content around the products they buy, or will shop for prices, or will look for feedback from other users of the product.
IMHO, Google, Meta and Microsoft are best positioned to be the last ones standing because they have alternative cashflows.
The danger with OpenAI and Anthropic is that they might end up being the Sun Microsystems of the AI era. It will only takes them a couple of misteps along the wrong technology path for them to be out of the game.
Yeah, he falls in the categories "is able to underline the issues" + "Explain why you are beeing bullshitted", but also in the "Snakeoil salesman", and that made him hard to trust. Basically i listen to him to talk basketball and that's basically it.
I think it's unquestionably right that these companies can't all win, and those that don't win are going to burn a lot of money for nothing. However there's kind of two directions this can go: Compute gets cheaper, in which case there's no monopoly it'll be easy for many companies to make good models and there won't be pricing power on serving a good model. The other case is compute gets cheaper but we keep using more and more of it, so it does likely become winner take all. The first scenario is good for the economy but likely bad for the returns on these AI stocks. The second is maybe bad for the economy and maybe not even good for the winner.
Take Google or Meta: Today Google makes a shit-tonne of money and to make that money they need to run some servers. The servers are extremely cheap relatively to the revenue they make running the business. This makes them a very attractive stock - the core of why SAAS looks great. Now let's assume the monopoly path. Google can win. I think they likely will win. But now they're going to spending... how many hundreds of billions constantly training new models? The cost of providing the service suddenly isn't small relative revenue they're getting. So even for them it looks awful for their valuation.
> Fewer people applying for patents, because the minute you apply for the patent, it's available to everybody, which means every model can train on it
We know LLM companies have, for lack of a better word, "sidestepped" the copyright on millions of works with their "transformative fair use" arguments. Are LLMs also a way to sidestep patents?
I don’t see how? You can train on something pending patent, but what are the benefits? If it gets patented you open yourself to get sued, I don’t see how AI works around that, the idea itself is still patented? I think I’m missing something for the argument to make sense. Or is the idea that if too many people use your patented idea you won’t be able to enforce it? That sounds risky to me
LLMs are accelerants. They enable people to do patent and copyright infringement at a much larger scale. As we know from previous examples, if you break the law enough as a company eventually they have to let you keep doing it.
Not just as a company...
He's right, there is a race. It's going to be a natural monopoly or duopoly because the cost to train the next SOTA model is always increasing. I can see that there are only 3 companies competing for the duopoly or monopoly realistically: OpenAI, Anthropic, and Google. Everyone else has fallen behind. The flywheel of generate more revenue, get more data, get more compute train a better model might already be too great to overcome for anyone else.
I don't understand why he thinks OpenAI can't be one of the duopolies or become the monopoly. OpenAI's models are always the first or second best overall - usually the first. They are also leading in the consumer market by a wide margin. They also made a strategic decision that is paying off which was committing to more compute early on while Anthropic is hammered by the lack of compute.
PS. They've raised ~$200b total, not $1 trillion.
> I can see that there are only 3 companies competing for the duopoly or monopoly realistically: OpenAI, Anthropic, and Google.
I could see people saying this in 2022, but now? No chance.
Chinese models keep demonstrating that SOTA can be approximated for a fraction of the cost. The innovation out of these companies keep showing diminishing returns, with a greater emphasis on the tooling and application layer. Having the right workflow with the right data is more important than having the right model. We could freeze AI now, and I'd bet good money that the current state of things is good enough to - not be first - but competitive for the next few years.
Even if we do end up with a oligopoly situaiton, it'll be less like Microsoft in the 90s and more like Microsoft now where they just give out windows for free, have support for WSL and the focus is on cloud services rather than their OS.
I think the performance of models is only one aspect. You have to take in account the cash flow, how much spending commitment the different actors have, debt, etc. OpenAI has taken some very risky commitments, of they don’t get the revenue to cover their expenses in the next few years their situation will be pretty bad
How can this become a monopoly/duopoly? There is no moat, the Chinese providers will continue to hunt the market leader at 10% of the price, there is no network effect (OpenAI's Sora was a play in that direction and failed).
I'm constantly amazed how this AGI/monopoly narrative can be kept up so long in the West, it just doesn't make sense (unless the state creates said monopoly by forbidding competition).
There is clearly a moat - or Claude Code wouldn't be generating over $10b in ARR every single month.
That's not what "moat" means. Claude Code has a castle. A "moat" is meant to protect the castle from invaders. It would be things like high switching costs, proprietary formats, network effects, etc. that aren't there.
In other comments people mention the "flywheel" of data and money feeding training, but there's a view that at some point the baseline open-weight models are "good enough" that the money will dry up.
My guess is that the industry will consolidate. The winners will absorb the losers and focus on generating revenue.
Therefore, there will be a growing gap between open and free models and the proprietary SOTA models.
> Open-weight models aren't going to be free forever.
The ones that are already released are, and they're already very good for most purposes and can be fine-tuned indefinitely, includin months or years down the line when processes have been optimized and things aren't as compute-heavy as they are now.
What the open-weight labs have shown is that you can go from nothing to competing with SOTA models at a tiny fraction of the cost for the SOTA models.
If there is consolidation by absorption, that derisks attempting to challenge the SOTA providers, and so they will keep facing attempts.
That's definitely a moat. Being able to generate ARR every month.
Yeah, no, i disagree. Frontier models were almost untouchable 6 month ago, but now i can get 90% of Opus 4.5 with any chineese model, or even with Mistral. The only thing i'm missing is the chain of thought that help me understand the "how" and "why" when AI fails at its task. For the "general purpose" AI, it's even worse, any free model i can run on my Intel Arc (yes, sorry, it was discounted an very cheap) i get like 80% of a frontier model, at virtually no cost, and i suppose Deepseek/Mistral are like 95% there.
Out of those 3, only Google seems to be in the position to reach that kind of profit levels due to distribution and advertising.
Claude is kicking ass in the niche of coding and processes.
1 trillion is a lot of money for something that's not differentiated and protected in a massive market.
Does it look like OpenAI has that in place?
Cuban thinks they don't, and won't.
I wrote about how I think OpenAI is going to kill it in advertisements here: https://news.ycombinator.com/item?id=46087109
Claude is kicking ass in coding but it seems like Codex is catching up fast. Claude Code's PR has taken a hit recently due to the lack of compute forcing Anthropic to dumb down the models. Codex has been gaining momentum.
Chip manufacturing aren't really differentiated either - it didn't stop TSMC from becoming the monopoly for high end chip nodes, capturing 90%+ of the advanced chip market. The reason they have is because Rock's Law makes it too expensive to build the next node unless you've generated enough revenue from the current node. I don't see why it isn't the same for SOTA models.
Chip manufacturing is insanely hard, it requires know-how, that's the moat. It's not money, otherwise the EU and China would have leading edge fabs.
Machine learning has no real moat. There's no network effect, it's not hard (you can just throw money at the problem). It's not data, because we have an existence proof that general intelligence can be trained by a few humans and a shelf full of books. The compute to do it is generally available. As soon as one organization releases open weights, everyone can use it immediately, even on modest local hardware.
I understand your argument, but I think you might be overestimating the intent of the users when they're using chatgpt.
The ones killing on ads are Google, Meta, and Amazon.
I just don't see how ChatGPT will gobble those market shares - ads are increasingly tied to sales attribution, and it would require a complete shift of the market for ChatGPT to take over the role of those 3 players.
People will still try to look for content around the products they buy, or will shop for prices, or will look for feedback from other users of the product.
https://www.reuters.com/business/openai-makes-five-year-plan...
This is a 5 year pledge - likely based on hitting revenue goals and not just using investor money.
IMHO, Google, Meta and Microsoft are best positioned to be the last ones standing because they have alternative cashflows. The danger with OpenAI and Anthropic is that they might end up being the Sun Microsystems of the AI era. It will only takes them a couple of misteps along the wrong technology path for them to be out of the game.
In my experience, Cuban is generally pretty good at stripping away the stupidity and BS.
He's stating the obvious, but perhaps it needed to be said.
Sometimes he is the stupidity and BS.
Genuine question: could you please share some examples?
Yeah, he falls in the categories "is able to underline the issues" + "Explain why you are beeing bullshitted", but also in the "Snakeoil salesman", and that made him hard to trust. Basically i listen to him to talk basketball and that's basically it.
Why should they return your money if it's a Ponzi scheme?