AI profitability

[mike]

Yesterday I heard a podcaster who has had a pessimistic view of the economy, partly because of fears of an AI bubble, start to change his tune a bit. In his view, for some time the market has been inflating an AI bubble similar to the dot-com bubble of the late 90s or the housing/credit bubble leading up to 2008.

With AI being a new technology, it has attracted huge amounts of venture capital, and the idea was that eventually the better technologies and companies would be sorted out, and there would be huge losses on the part of those who don't succeed, driving a downturn in the markets.

But what he said was that industries connected to AI are beginning to make real profits. According to what I found by poking around a bit, OpenAI (ChatGPT) is seeing the fastest growth, and is making some of the greatest investments, but is only on a path to profitability by 2029-2030.

However, Anthropic (Claude) is nearing break-even status, seeing extremely high growth, and should be profitable by 2027-2028. And Google, for which AI is only a part of its business, is of course hugely profitable. And these are just a few of the larger AI companies of course.

What all of this means is that the actual productivity of AI is likely gaining momentum, and that what we are seeing in the growth of AI in terms of stock value may actually reflect that rather than a bubble ready to burst.

Just thought it was interesting.

[JohnH]

But what he said was that industries connected to AI are beginning to make real profits. According to what I found by poking around a bit, OpenAI (ChatGPT) is seeing the fastest growth, and is making some of the greatest investments, but is only on a path to profitability by 2029-2030.

Interestingly, OpenAI is legally a nonprofit, so one wonders what exactly the purpose even is of laws about taxes and nonprofits and for-profit businesses something like that exists. (OpenAI regularly pays million-dollar salaries packages.)

However, Anthropic (Claude) is nearing break-even status, seeing extremely high growth, and should be profitable by 2027-2028. And Google, for which AI is only a part of its business, is of course hugely profitable. And these are just a few of the larger AI companies of course.

Anthropic has massively raised prices and also lowered the quality of the product they are offering, with many complaints their service is no longer effective for what it used to do. That means competitors are going to start eventually getting their customers, and they will not be able to charge the highest prices in the industry anymore.

Midjourney achieved profitability (as far as we know) quite a while ago.

Google as a whole is profitable, but it is unclear if Google's AI division is profitable at all. It is doubtful.

What all of this means is that the actual productivity of AI is likely gaining momentum, and that what we are seeing in the growth of AI in terms of stock value may actually reflect that rather than a bubble ready to burst.

If the model for profitability is Anthropic, that's a very bad omen for the AI industry as whole.

[mike]

But what he said was that industries connected to AI are beginning to make real profits. According to what I found by poking around a bit, OpenAI (ChatGPT) is seeing the fastest growth, and is making some of the greatest investments, but is only on a path to profitability by 2029-2030.

Interestingly, OpenAI is legally a nonprofit, so one wonders what exactly the purpose even is of laws about taxes and nonprofits and for-profit businesses something like that exists. (OpenAI regularly pays million-dollar salaries packages.)

Elon Musk of course has sued OpenAI for this very reason. Trial is going on as we speak. The outcome will be quite interesting.

I'm not familiar with Claude/Anthropic, but with the wild growth and path to profitability, it doesn't seem to experiencing the consequences of the alleged price-jacking and lesser quality quite yet.

I'm more familiar with Gemini, Grok, and ChatGPT, although since I've gotten a paid sub to ChatGPT I rarely use the others.

[JohnH]

I'm not familiar with Claude/Anthropic, but with the wild growth and path to profitability, it doesn't seem to experiencing the consequences of the alleged price-jacking and lesser quality quite yet.

Yes, one can often achieve short term profitability by simply raising prices and lowering quality. Whether or not this works depends on how much competition there is, generally speaking.

I don't see how the current American players can compete with Chinese prices, long-term. DeepSeek V4 is 10 - 20 times cheaper than Anthropic, and is adequate for most tasks.

[mike]

Today it was announced that Anthropic is leasing an entire SpaceX data center. For awhile now, Apple's Siri has been using Google's Gemini as its AI engine. Some really strange AI partnerships are developing among fierce competitors.

[Soloist]

Today it was announced that Anthropic is leasing an entire SpaceX data center. For awhile now, Apple's Siri has been using Google's Gemini as its AI engine. Some really strange AI partnerships are developing among fierce competitors.

Claude is excellent as a tool with careful use but lately they've struggled with the influx of users who fled Chatgpt. This means on occasion its brain dead.

[mike]

Today it was announced that Anthropic is leasing an entire SpaceX data center. For awhile now, Apple's Siri has been using Google's Gemini as its AI engine. Some really strange AI partnerships are developing among fierce competitors.

Claude is excellent as a tool with careful use but lately they've struggled with the influx of users who fled Chatgpt. This means on occasion its brain dead.

Even on the paid plans?

I have a $20 or so per month ChatGPT subscription, and when processing a lot of images, eventually it forces timed breaks like 5-6 minutes before doing more.

[Soloist]

Today it was announced that Anthropic is leasing an entire SpaceX data center. For awhile now, Apple's Siri has been using Google's Gemini as its AI engine. Some really strange AI partnerships are developing among fierce competitors.

Claude is excellent as a tool with careful use but lately they've struggled with the influx of users who fled Chatgpt. This means on occasion its brain dead.

Even on the paid plans?

I have a $20 or so per month ChatGPT subscription, and when processing a lot of images, eventually it forces timed breaks like 5-6 minutes before doing more.

I have the 20$ pro plan with Claude. For a while I was using chatgpt. I haven't tried the programming with OpenAI's new thing but I code regularly with Claude. They have it as 5 hour sessions with caps and weekly caps. If you start new conversations/coding sessions you can get a fair bit. If you try to cram it in one, you run out.

[JohnH]

Today it was announced that Anthropic is leasing an entire SpaceX data center. For awhile now, Apple's Siri has been using Google's Gemini as its AI engine. Some really strange AI partnerships are developing among fierce competitors.

Claude is excellent as a tool with careful use but lately they've struggled with the influx of users who fled Chatgpt. This means on occasion its brain dead.

Even on the paid plans?

I have a $20 or so per month ChatGPT subscription, and when processing a lot of images, eventually it forces timed breaks like 5-6 minutes before doing more.

OpenAI/ChatGPT hasn't been having any problems. Anthropic/Claude has been having major problems due to oversubscription, which they have dealt with by raising prices, suddenly imposing usage limits on those paid plans, and then decreasing the quality of what you get for that.

So Anthropic is pretty desperate, and decided to see if SpaceX (formerly xAI) could help them out a bit. Grok is very fast and has lots of unused capacity and no real usage limits for paid plans.

[Soloist]

Claude is excellent as a tool with careful use but lately they've struggled with the influx of users who fled Chatgpt. This means on occasion its brain dead.

Even on the paid plans?

I have a $20 or so per month ChatGPT subscription, and when processing a lot of images, eventually it forces timed breaks like 5-6 minutes before doing more.

OpenAI/ChatGPT hasn't been having any problems. Anthropic/Claude has been having major problems due to oversubscription, which they have dealt with by raising prices, suddenly imposing usage limits on those paid plans, and then decreasing the quality of what you get for that.

So Anthropic is pretty desperate, and decided to see if SpaceX (formerly xAI) could help them out a bit. Grok is very fast and has lots of unused capacity and no real usage limits for paid plans.

Yeah Anthropic wasn't counting on how badly OpenAI would screw up their customer opinions and their ads may have been too successful.