Elon Musk has sold social media site X to his own xAI artificial intelligence company in a $33 billion all-stock deal, the billionaire announced on Friday.
Who do you mean with that? Companies like OpenAI or Anthropic, or do you also include the likes of Google/Amazon/Microsoft?
With the former I can see it, but the later also profit from providing the infrastructure (and have other profitable business), so imo those will be just fine.
I definitely see Google/Amazon/Microsoft shedding a huge amount of market cap when the time comes to write-off the 100s of billions they invested the past two years.
They just don’t have any feasible path to recouping those investments.
Sure, they’ll never go fully broke, that’s just a nice word for emphasis.
I actually think Google is going to win this one, they’re the only ones making their own hardware to run their own models. Open ai are starting down that road but they’re years behind.
When you look at the pricing of all the AI companies, Google’s is so much cheaper (orders of magnitude) and they’re not having to pay Nvidia billions to do it.
As i understand it most of the money they are investing goes into new datacenters. So when a model gets outdone by a new one they still have those, unlike e.g. OpenAI that use other companies resources (i think microsoft and oracle mostly?). In a way companies that use those external clouds to train their own models are financing the investments needed for the big players.
AWS, GCP and Azure are all growing 30%+ yoy, are profitable and if anything supply constraint in that they can’t build more capacity fast enough to meet demand. So it seems to me that to some degree they are already recouping some of those investments. I don’t see a drop in demand for compute, and even if using/training ai would become less resource intensive, Jevons paradox would just lead to more demand.
Of course they also burn a lot of money as anytime a new model gets trained and beats the older ones, it kind of renders the resources spend on the previous one worthless. But to me that seems like the cost of doing business.
The current investments they can afford. What would actually lead to shedding huge amounts of marketcap is, if they’d let a rival establish themselves. Similar to how the movie studios didn’t get into streaming early (mostly to not hurt their cable business) and gave Netflix enough time to establish themselves.
To comment on something you mentioned in another reply below:
I just don’t see a world where most people are coughing up more than $10 a month for AI.
I think the big money will be in the business world, where salaries for actual people are high enough that saving a person even a few hours/week or replacing a single employee saves so much money that even expensive subscriptions would easily be worth it.
On the consumer side as you say running smaller models locally will likely be the norm. But that means it would be free for both the likes of Deepseek and Google. And then it’ll just come down to who has access to personal information and is better embedded, which would be likely be whoever also controls other aspects of a users life, such as Goole with Android, gmail etc. Money here will be made just as it is done with other free services.
You could have made this same analysis in 2000 and it would be equally valid.
Yes, the business world is willing to pay big bucks to reduce labour costs and that business case is solid.
But we already see that success is not determined by the size of the model, but by the data and providing and processing that data in a smart way to the AI. And the companies that are successful in this area are model agnostic. They can, and will, switch to cheaper to run models that are good enough for their purposes.
So the dogma that whoever has the biggest model wins, just doesn’t apply. AI is already hitting diminishing returns.
Once the investment money pumping the hype is gone, there will be a glut of capacity and a heavy price competition, which will drive down margins.
Who do you mean with that? Companies like OpenAI or Anthropic, or do you also include the likes of Google/Amazon/Microsoft?
With the former I can see it, but the later also profit from providing the infrastructure (and have other profitable business), so imo those will be just fine.
I definitely see Google/Amazon/Microsoft shedding a huge amount of market cap when the time comes to write-off the 100s of billions they invested the past two years.
They just don’t have any feasible path to recouping those investments.
Sure, they’ll never go fully broke, that’s just a nice word for emphasis.
I actually think Google is going to win this one, they’re the only ones making their own hardware to run their own models. Open ai are starting down that road but they’re years behind.
When you look at the pricing of all the AI companies, Google’s is so much cheaper (orders of magnitude) and they’re not having to pay Nvidia billions to do it.
I actually agree with you, they definitely have the edge, but I still am skeptical that they will be able to maintain their valuation.
I just don’t see a world where most people are coughing up more than $10 a month for AI.
Most people will only use free AI and companies will mostly buy cheap AI.
Running Deepseek locally is basically free. That’s the competition.
As i understand it most of the money they are investing goes into new datacenters. So when a model gets outdone by a new one they still have those, unlike e.g. OpenAI that use other companies resources (i think microsoft and oracle mostly?). In a way companies that use those external clouds to train their own models are financing the investments needed for the big players.
AWS, GCP and Azure are all growing 30%+ yoy, are profitable and if anything supply constraint in that they can’t build more capacity fast enough to meet demand. So it seems to me that to some degree they are already recouping some of those investments. I don’t see a drop in demand for compute, and even if using/training ai would become less resource intensive, Jevons paradox would just lead to more demand.
Of course they also burn a lot of money as anytime a new model gets trained and beats the older ones, it kind of renders the resources spend on the previous one worthless. But to me that seems like the cost of doing business.
The current investments they can afford. What would actually lead to shedding huge amounts of marketcap is, if they’d let a rival establish themselves. Similar to how the movie studios didn’t get into streaming early (mostly to not hurt their cable business) and gave Netflix enough time to establish themselves.
To comment on something you mentioned in another reply below:
I think the big money will be in the business world, where salaries for actual people are high enough that saving a person even a few hours/week or replacing a single employee saves so much money that even expensive subscriptions would easily be worth it.
On the consumer side as you say running smaller models locally will likely be the norm. But that means it would be free for both the likes of Deepseek and Google. And then it’ll just come down to who has access to personal information and is better embedded, which would be likely be whoever also controls other aspects of a users life, such as Goole with Android, gmail etc. Money here will be made just as it is done with other free services.
You could have made this same analysis in 2000 and it would be equally valid.
Yes, the business world is willing to pay big bucks to reduce labour costs and that business case is solid.
But we already see that success is not determined by the size of the model, but by the data and providing and processing that data in a smart way to the AI. And the companies that are successful in this area are model agnostic. They can, and will, switch to cheaper to run models that are good enough for their purposes.
So the dogma that whoever has the biggest model wins, just doesn’t apply. AI is already hitting diminishing returns.
Once the investment money pumping the hype is gone, there will be a glut of capacity and a heavy price competition, which will drive down margins.