Love him, hate him, or don’t know him, Ed Zitron dropped an exclusive last night that really puts an exclamation point on the AI bubble.
Before I get into it, saying this is a bubble isn’t saying AI is useless. It’s not even saying all companies using AI are going to decrease in value. In fact, I’d say if the bubble burst, companies like Meta would actually go up in value because Zuckerberg would be forced to stop his latest iteration of spending billions chasing something he will never win.
But nonetheless, the core of this so-called AI revolution is OpenAI/ChatGPT. Arising from that is Anthropic which basically does exactly the same thing with different PR.
Both companies have filed for IPOs valuing their companies at upwards of $1 trillion. And that is where I have a problem because, when these companies literally can’t deliver, they’re going to take a lot of great companies down with them.
First, Ed Zitron’s Exclusive
I’m not just writing this article because of Ed’s post but it was the catalyst that got me to actually open my laptop.
Ed is a consistent AI doomer who has gained a lot of traction by basically saying generative AI is a mediocre technology that has very few real use cases but, worst of all, the economics just don’t work.
That is, it costs way too much money to train and run these models than customers will ever pay for it.
For the record, he is all in on this narrative which I don’t fully agree with, but the numbers he just dropped last night are quite telling.
He was able to get access through presumably an anonymous source to OpenAI’s 2024 and 2025 numbers. Basically, OpenAI lost $5 billion in 2024…
…and $38.5 billion in 2025.
I realize that OpenAI and Anthropic losing money is not a surprise and that’s been part of the plan all along. They’ve had to get creative, even desperate, and use fear-mongering and other hype tactics to stay relevant so they can keep raising money until their models apparently become good enough that they effectively become two more Big Tech giants, making billions of dollars of actual profit.
But there are huge problems with that.
The ROI Just Isn’t Clear
Both Anthropic and OpenAI have stated their consumer-facing product (the chatbots) are not the real financial play. It’s enterprise clients incorporating the technology into their work flows.
And, until the last few months, at least on paper it looked like this was going well. Tokenmaxxing, probably the dumbest thing I’ve ever heard of, became a thing and corporations were spending endless money.
Then, finally, someone finally questioned it. The COO of Uber, Andrew Macdonald, said – after the company blew their entire annual budget in the first few months of the year – that he doesn’t see where they’re shipping more features that consumers actually want.
I’ve written many times that most people are sheep. It might seem baffling, but the industry really did need permission to finally question it. Everyone was looking around, thinking “I must be missing something… I’ll just keep waiting…. everyone can’t be wrong.” Then finally someone says, hey maybe this technology isn’t some magical thing that instantly leads to more productivity.
And now everyone’s allowed to look at their own finances and ask themselves if putting AI in every single place they can possibly put it was a wise move.
Now this is leading to the next problem.
Tokens Are Too Expensive
Sam Altman, after bewilderingly saying that nobody was questioning the cost before but now they are, is considering lowering the cost of tokens to compete with Anthropic which also plans to lower the cost.
If this isn’t a race to the bottom, what is?
And it’s a race to the bottom because there isn’t really a moat. It’s a commoditized technology which I’ll get to a minute.
But anybody who is at all considering whether to invest in Anthropic and/or OpenAI should have alarm bells going off in their heads. Can you imagine any other company preparing to go public when they lost $38.5 billion last year, all of their consumers are questioning the costs, they plan to lower prices, and their competitor has the exact same technology?
What am I missing here? This really just doesn’t work without the use of the technology showing absolutely crystal clear gains in productivity at mass scale in a wide range of industries.
And I don’t mean, AI makes it easier to create my marketing plan. Or AI lets me create summarize my emails.
The former is a race to the mediocre middle and the latter is making you a lazier, dumber person every day. In fact, I’ve been working on an article dissecting, psychologically, just how much these LLMs are warping our minds.
I know every university has a study going on, and everyone knows chatbots can be dangerous. But understanding, truly, what is going on is the only way to learn how to use it effectively to actually make money.
But, for this article, I’ll stick with OpenAI and Anthropic’s massive problems:
Open Source Models Are Good Enough
Everyone reputable, at least according to the benchmarks, says that open source models are around 6 months behind these frontier labs.
Please consider this reality: OpenAI and Anthropic have a technology that loses tens of billions a year with plans to lower costs because customers are increasingly unhappy, and open source competitors are just 6 months behind them at 90% lower costs.
I feel like I’d really have to stretch my mind to justify this without just flat out admitting I, for some reason, just love Sam and Dario.
So, to reframe this:
- Let’s say it’s January 1, 2026. Corporation X incorporates OpenAI into their work flow at $10 per unit (just using that for simplicity).
- Let’s say Corporation X does see a gain, which is a huge stretch, but let’s just say they see a 5% productivity gain.
- Now let’s say it’s July 1, 2026, and DeepSeek or some other open source model can do what OpenAI could do in July for $1 per unit.
In what world does Corporation X not switch to the $1 model and make more profit?
Even if you argued that the July 1, 2026 OpenAI is vastly better, in most work flows, why would this matter? Why would you need the latest model in every work flow, considering work flows are typically pretty much the same, so if you improved it on January 1 with this new technology, it’s still improved on July 1 all the same – just at a 90% discount.
Where is the actual path that justifies a trillion dollar valuation?
I honestly don’t care if OpenAI and Anthropic go under or, more likely, get absorbed by the hyperscalers that are funding them. But the two companies are synonymous with AI, so when this happens, the investment world of sheep are going to panic and every tech companies is going to suffer.
OpenAI and Anthropic are closer to Pets.com than Google, Microsoft, or Amazon.
Apple and Google Are The Nail In the Coffin
In typical Apple fashion, which I believe is actually smart, they’re moving at a snail’s pace when it comes to AI. It has been insane to me to watch the world react to a $4T company that prints profits very carefully enter the generative AI space.
Tim Cook’s entire legacy has even been called into question. What?
Now that Apple is entering the space with their latest announcements, I really can’t figure out where OpenAI and Anthropic fit outside of coding. And that’s one reason Dario should be slightly praised because he saw early on that the clear opportunity with LLMs, used at scale, is coding.
It’s a whole other conversation as to whether the vibe coding era, which will probably go the way of NFTs, is superficially inflating the numbers.
But, back to Apple. When Siri finally becomes actually useful for answering questions and, even more importantly, Siri can take actions across your other apps for you, where does OpenAI and Anthropic fit?
Apple and Google have actual moats. They have real products that people use every single day, so who is going to win when a commoditized technology comes along? They are, of course.
In the past few months, I moved a significant amount of my money into Apple and Google for this very reason. I am not saying you should do the same, and I’m certainly not saying both won’t drop substantially when this AI era inevitably has a downturn.
But they both have the hardware, the diehard consumers, the chips, even the frontier models. I can’t figure out why OpenAI and Anthropic are necessary at all?
Concluding My Thoughts
Again, AI is not useless. Large language models are not useless. They are, however, illusionists that have warped our minds into thinking this is somehow the future, which is no surprise considering their creators are doing exactly the same thing.
When your technology is comparable to competitors’, and your consumers will switch in a heartbeat to whichever one is cheaper, you do not have a trillion dollar company.
When larger companies with real products have the same technology, where is your moat? If Gemini / Siri can do all the same things, I can’t figure out why you need to exist.
Let me give you a real life example. I was actually writing a paragraph here about what OpenAI can do to maintain their momentum because they do actually have the massive fortune of being a household name now akin to “Googling” something.
In my example, I was writing that I recently updated my subscription to the $100 pro model for one reason: the personal finance feature. I have wanted to just be able to talk to finances for so long.
I tried to connect it with Perplexity and build my own, which failed way too many times.
Chase, led by Jamie Dimon who said AI is the next electricity, is apparently too idiotic to actually just put the technology directly in their app.
And, as stated, Apple has been slow to incorporate AI features.
But here’s the thing and the problem: I’ll cancel my $100/mo the moment Siri can communicate with Chase, which is already on my phone. The next level of consumer AI is simply being able to talk with all of our applications, and I understand this is possible right now with OpenAI and Anthropic.
But, like everyone else, I want convenience. I don’t need to run it all through ChatGPT if it’s right on my phone. And that’s where it’s all heading.
Again, there’s no moat. Worse than their being no moat is that nobody even knows how much, if at all, very expensive token usage is even helping them.
The counter to this is that there will be a breakthrough of some kind. That’s fine and possible. These research labs are not bad in themselves and I love that they’re being funded by trillion dollar companies. It leads to fun technology.
But they’re research labs in the end. They’re not trillion dollar companies and the world thinking they are is why the entirety or the markets will likely see a 30% downturn in the next couple of years. That may seem irrelevant to many people, but it’s very relevant: when public companies struggle, they cut jobs.
Look hard at large language models. Think about them. Think about what they do.
They give you the ability to communicate with large bodies of language.
It’s a phenomenal technology. It’s powerful and confusing all in one. Let’s just not confuse the technology itself with how actual business, economics, and profits work.
PS: I don’t really re-read or edit my posts anymore before publishing. This isn’t supposed to be a polished piece. These are my 4:30am thoughts before my daughter wakes up.
