OpenAI's Problem

On my personal blog/newsletter, I often write about AI. Sometimes it is positive posts, and sometimes it is negative. I often speculate at how does all this money being poured into Data Centers, GPU's and a lot of marketing ever makes a return on this oversized investment.

In 2024, it was estimated that about $394B waw spent chasing the GenAI dragon, and this year, it is estimated that the hyperscalers, the neoclouds, and the tech elite tier companies will spend near $450B when all is said and done.

The most generous estimate of OpenAI's revenue is about $13B, and their revenue burn is much more than that. (read Ed Zitron's Where's Your Ed At for some outstanding bottom's up analysis)

And OpenAI is the big dog, #2 in the space, Antropic is less than half that in revenue.

The rest is noise.

So, clearly, OpenAI is going to be the victors in this race, right?

Time to put your thinking cap on, and let's see what we can divine.

The Concept of a 'Moat'

Warren Buffett was famous for his newsletters, and his business practices. One of the things that was primary in Buffett's criteria for investing in a company (or acquiring it) is the "moat" they have around their business.

From Investopedia:

An economic moat is more than just a fleeting competitive edge—it's a sustainable advantage that allows a company to outperform its rivals over an extended period. These can take many forms, but they all serve the same purpose: protecting a company's market share and profitability from competitive forces. Warren Buffett has explained the idea many times. Here's one example from when he answers an investor's question about how he looks for winning companies:

We're trying to find a business with a wide and long-lasting moat around it, surrounding and protecting a terrific economic castle—with an honest lord in charge of the castle … For one reason or another, it can be because it's the low-cost producer in some area. It can be because it has a natural franchise [or] because of its service capabilities, its position in the consumer's mind, [or] because of a technological advantage. For any kind of reason at all, it has this moat around it.

You do not need an MBA to understand the concept. Moats come in many forms. Apple has one that is iOS and MacOS. Google has one that is their ecosystem of Android, gmail, G-Suite and their real business that is their Ad tech stack, Amazon has AWS and their e-commerce empire.

Hell, Facebook has its ad-stack (and let me tell you, you have to block about 1,200 URLs of servers to prevent them from tracking you across the net) and the inertia of the most customers of their digital properties (Facebook, Insta, Threads) that even Google failed to take a bite out of with their Google+ stab.

All of these are moats in the classic sense.

OpenAI? They really have no moat. Sure, they are the first mover, the first to capitalize on the Generative AI revolution. And they have the most monthly users (MAU). They are the "Kleenex" of AI for brand recognition. Everybody thinks of ChatGPT, even if they are using other services.

And while ChatGPT does some things amazingly well, the other providers have their strengths. Claude, the LLM from Anthropic, is believed to be the best for coding use. Google's NanoBanana is the current best at image generation. Facebook seems to use their LLM to help people flood their properties with AI Slop[1].

But one datum stands out. A few weeks back, Google released their version 3 of Gemini, and the public perception was that it had (slightly) better performance than ChatGPT 5.1.[2]

This caused Sam Altman to declare a "Code Red" as immediately the WAU metric dropped 6 or 8%, the first drop in their growth curve.

Additionally, when ChatGPT 5 was released, the critical response was largely "meh". Sure, it was better than the prior 4.5 release, but it wasn't a massive jump in performance. The web was awash with speculation that the foundational models of LLMs had hit a scaling wall. While 6 months before, there was confidence that just tossing more GPU's and training data would be enough to achieve the holy grail, AGI, even the most devoted believers began to wonder if the GPT train was running out of steam.

And since then, OpenAI has been tossing things at the wall to see what sticks. Sora2, the short form video generator that flooded the web with clips of Sam Altman stealing GPU's and Spongebob running from the law. Funny, but not really useful.

Then there was the announcement that they would let users who are verified adults be able to use ChatGPT to write Erotica.

And then there is all the circular financing that has been announced that while it doesn't look like Enron, it sure smells fishy.

In fact, OpenAI is negotiating about $1.4T in infrastructure build out over the next few years, and has been fishing (with their friend David Sacks) for a government guarantee on the loans/financing.

That feels like a business that has no defensible moat.

Apple, Google, Meta, and even Anthropic (they are aligned to Amazon) are better positioned. OpenAI is beginning to feel very vulnerable to competition.

And that lack of a moat should make all their employees nervous.

Final Notes

None of these players in this space seem to have a path to profitability. The infrastructure spend and the commitments to build data centers filled with compute and GPU's is just bonkers. Apparently in 2025, the spend on the infra was more than the growth in consumer spending, and many smart financial analysts reckon that the spending by the Magnificent 7 is carrying both the economy and the stock market.

That wild abandon in build-out can't continue.

Also, the sycophantic tenor of the standard ChatGPT interaction has already led to a few lawsuits from families of people who committed suicide after going down the rabbit hole with ChatGPT. The fact that people are turning to a chatbot to be their therapist is just ugly.

At the end of the day, these models don't think, they are elaborately complex neural networks that are training and tuned to offer up rational guesses. In other words, a super autocomplete.

When this shakes out (and it will) OpenAI's lack of a defensible moat will lead to them becoming an also-ran.

1 - I am not on any Meta properties, as I block all their tracking in my hosts file. But my wife is still on Facebook and Insta, and she shows me some of the boomer-bait that the AI generated slop shows her.

2 - That led OpenAI to rush the 5.2 update to market.