Nothing makes sense anymore…
Looking to hide highlights? You can now hide them from the “•••” menu.
Okay, got it
Somehow, in between illegally destroying the US government, “managing” Tesla into the dirt, roleplaying as a gamer boy, and tweeting more than a cocaine-addled Trump, Musk decided to offer to buy the entirety of OpenAI for $97.4 billion. Almost as soon as the offer was posed, CEO of OpenAI Sam Altman batted it right back, tweeting, “No thank you, but we will buy Twitter for $9.74 billion if you want.” Oooofft, that must have stung. But I think Altman missed a trick here. He should have taken that offer and ran! Why? Well, OpenAI is up shit’s creek without a paddle. Let me explain.
Let’s start with why Altman likely said no.
First off, Musk’s offer was a lowball. A few months ago, OpenAI announced it had raised $6.6 billion in investment from the likes of Nvidia and Microsoft, which put the valuation of OpenAI at $157 billion!
Secondly, OpenAI has just secured the sole rights to use Project Stargate, a $500 billion project to build AI infrastructure in the US. Having exclusive access to that amount of computing power should give them a huge leg up on any competition.
Finally, Altman is convinced they are very close to creating Artificial General Intelligence, also known as superintelligent AI. In a recent blog post, he claimed that AI has broken Moore’s Law (the prediction that chip performance would double every 18 months, with no increase in power consumption). He points out that AI token costs have fallen ten times over the past year. Altman claims these falling costs suggest that artificial general intelligence (AGI) is “coming into view.”
On the surface, OpenAI looks poised to become even more of an industry titan. But, dig a little deeper, and that simply isn’t the case.
That $6.6 billion investment was needed just to keep OpenAI afloat, as they faced a $5 billion loss at the end of 2024. Moreover, they only spent $3 billion in 2024 developing new models, meaning that their wildly popular AIs are haemorrhaging cash. OpenAI is light-years away from being a profitable business.
And that $500 billion Stargate project will only compound this cash flow issue. As the exclusive user of such expensive infrastructure, how can they afford to pay for its upkeep or operation if they aren’t making any money, even with the far cheaper infrastructure they currently have? Quite simply, they can’t.
But, surely, if OpenAI can unlock superintelligent AI, then they will have far more customers, and their income will skyrocket, resolving this issue. Right?
Well, no. Altman’s blog post was even more tech-bro smoke and mirrors. You see, an AI token is how much it costs to use an AI, not how much it costs to develop an AI, which is the limiting factor of AI capability. As such, it is not comparable to Moore’s Law at all. In fact, AI already has its own Moore’s Law called the “Efficient Compute Frontier.” However, it is far less optimistic, and if you want to build anything even resembling superintelligence, you need to break it, which no one has been able to do.
The Efficient Compute Frontier is a very fancy way of saying that AI training has seriously diminishing returns. I have covered this topic several times before, so if you want a more detailed description of this issue, go here. But in short, AI has to be “trained” on data (a name that just anthropomorphises the program processing said data) to “learn” (again, a name that anthropomorphises replicating patterns in said data). So, if you want to improve your AI, you need to train it on more data, which requires more computing power. However, the relationship between the amount of training data and performance isn’t linear. There are diminishing returns. So, if you double the training data of a small AI, you might see a performance increase of 5%, but if you then double the training data again, you might only see a performance increase of 1%, and so on. OpenAI’s models are now so large that doubling their size produces barely noticeable performance increases.
Consequently, the models they can run on Stargate’s $500 billion infrastructure aren’t likely to be that much better than what we currently have. Moreover, there is mounting evidence of another limit to AI, as these larger models have started getting worse at broader, more basic tasks than their tiny cousins. So, no, OpenAI has no viable route to superintelligent AIs.
Okay, so OpenAI effectively needed a corporate bailout to stay afloat for another year; they are locked into a $500 billion project that will make cash flow even worse and have no viable route to producing significantly better AIs… Yeah, it doesn’t sound like a healthy company, does it?
But it gets worse!
Remember DeepSeek? They were able to build an AI just as efficient as the latest and greatest from OpenAI for less than 1% of the cost of OpenAI’s models, and on top of that, their AI costs 96% less to run than OpenAI’s models. To understand how they did this, and why they still haven’t broken the Efficient Compute Frontier, read my article here. But because DeepSeek is so damn cheap, they were able to offer it to the world for free! To even stay in the AI race, OpenAI has had to announce that their next models will also be totally free to use.
Now, yes, there are some security issues with DeepSeek. However, because it is an open-source model, we can study the code, find these issues, and help correct them. What’s more, nothing is stopping a Western company from copying DeepSeek and completely undermining OpenAI domestically.
In other words, DeepSeek has just destroyed OpenAI’s entire business model, leaving it high and dry with no real avenue to generate serious income, no matter how you spin it.
So, as it stands, OpenAI is set to not be able to significantly improve their AIs, have massively rising overheads, and their profits are set to fall off a cliff. Doesn’t sound like the foundations of a $157 billion company.
In fact, that makes Musk’s $100 billion offer look incredibly appetising!
Altman should have accepted this. It would have meant that companies like Microsoft, who have invested billions of dollars into OpenAI, could cut their losses and walk away clean. Moreover, OpenAI is set to go entirely for-profit this year, with Altman owning 7% equity in this new setup. So if he could push the sale past that transition (which would be easy, as Musk has been desperate to buy OpenAI for years, so Altman can drag his heels as much as he likes), it would score him a giant payday!
Furthermore, Musk is one of the few people in the world who can sink the money into OpenAI that it needs to continue. While he would almost certainly drive it into the ground, that is where OpenAI seems to be going anyway.
So, why isn’t Altman selling? Is it ego? Tech-bro stupidness? Greed? Saving face? Spiting Musk?
I don’t know, and truth be told, I don’t care. To me, this is just yet more evidence that our economy and corporations have drifted miles away from reality and are tumbling towards an existential crisis, with only us set to bear the damage.