GP Vs. GPUs: How OpenAI Loses Money

The Tulip Folly, by Jean-Léon Gérôme, 1882

It's disgusting how much OpenAI ignores Gross Profit. GP was the bedrock of Economics as I was taught it, but Technomics hits the crack rock of ignoring it. On the street, if coke costs 9 and cutting it costs 1, you need to sell crack for 10 or else you're done. If you lose money on each rock, you're not a dealer, you're a crackhead, or a narc. On Wall Street, however, if compute costs $5 billion and you sell it for $4.3, that's somehow a galaxy brain idea. Those are actualish OpenAI numbers, check the FT, and they're actually retarded.

OpenAI is just a money laundry for Microsoft and NVIDIA and other evil there. The business never even beings to break even, according to their own projections, and yet they're writing promissory notes worth trillions for decades into the future, as if they're building pyramids. They're pyramid scheming. As the FT says in their reporting, this is not a serious chart and these are not, as Logan Roy said, serious people.

Inference is the cost of each query you make on ChatGPT. As you can see, Microsoft gets paid cash upfront, while OpenAI has to take on debt for decades

To put it in street poetry, OpenAI is violating the 4th Crack Commandment as laid down by the Notorious BIG in '97. Never get high on your own supply. The supply chains of the US of AI are disgustingly incestuous. Microsoft is OpenAI's client for AI and their supplier for compute. OpenAI gives 20% to Microsoft, and Microsoft gives 20% of some Azure and Bing revenue to OpenAI. Is it really a business if you're moving promissory notes from one pocket to another? What are we even talking about? These guys are smoking their own supply and it doesn't just violate economics, it violates Crackanomics.

Early techlords like Google could violate Crackanomics because they still respected basic economics. But OpenAI is not Google. Google's marginal cost of serving you a webpage was marginal, while OpenAI's costs on inference alone are astronomical. Every instance of ChatGPT has to reincarnate fully, which is really expensive folly. It's comically and karmically expensive. It's like rubbing a genie bottle to do the dishes. At some point, just you run out of wishes. And I, for one, am here for it. The crash of OpenAI will be delicious, and if we're lucky, it takes the whole US economy with it.

Now that AI has to ravage a rainforest to return a brainfart, Capitalism has reached terminal velocity, straight down. It cannot get stupider than this. The business model accounting for 99.9% of American growth is 99.9% a pyramid scheme. The rich are building a pyramid of GPUs over the tomb of GP.

Compare this to the recent past, which was stupid but not this stupid. While Google and, even worse, Uber might have lost money at a net income level (GP minus operating expenses), OpenAI (seems to) lose money at a gross profit level, which is very different. They're buying coke for 20 and selling crack at 15, which violates the 0th Crack Commandment, which is make money. Every query you run on OpenAI doesn't just drain the water supply somewhere, it burns money.

OpenAI loses money on a GP level, and companies that do this are not supposed to exist. They're supposed to go out of business, because selling quarters for a dime is not a business. But now they're betting the whole US economy on this. It's not the USA anymore, it's USAI. As Economist Jason Furman said when you remove data centers and AI from the US of AI, growth is only 0.1%. GPUs are the tulips for this turnt empire, grown in copious bullshit, and ultimately useless.

Let's look at the numbers, as much as we can get them.

OpenAI's Financials

My sources for this are