This was excellent - thorough and objective. There’s been so much hype that it’s good to see sober eyes analyzing the underpinning economics of OpenAI as well as the progress of its competitors. What strikes me as interesting is that Microsoft hasn’t been eying the solid position that Google is in with the integration of AI into their cloud ecosystem. For those large potential customers who need proven FED RAMP or HIPAA level environments, it’s quite a combination to offer. I most likely may be missing Microsoft’s offering in this arena, all I know about is CoPilot, but Microsoft has a similar environment that could, with the right AI offering, be very compelling. They seem to be content to just supply the shovels and pans will OpenAI hunts for gold, but I would think they would pull OpenAI in, pair with their cloud tools and compete directly with Google. State and federal govt contracts using AI as solutions would be quite lucrative.
Great analysis! Like I said in my latest piece, this won’t be a “rising tide lifts all boats” scenario; we’ll see the winners and losers more clearly. Right now we’re in a “shoot first, ask questions later” type of market but I’m glad people like you, Gary Marcus, Ed Zitron, etc and even myself (who is nowhere near as known) are questioning the economics of AI.
I have a theory that there’s a non-zero probability OpenAI gets bought out by Microsoft in the future. The thing is with OpenAI relative to other LLM platforms & tech companies, they seem to be taking so many shortcuts in a bid to become profitable and it’s just messy. The other thing that made me question everything is Sam’s answer to Brad in his Bg2 podcast episode. It’s not reassuring to investors at all, and then Sarah’s comment about needing a government backstop was even worse. It became a PR nightmare.
It’d be silly for investors or anyone really to underestimate China, especially DeepSeek and Alibaba. Deepseek’s business model is self funding which puts them in a far more unique position than most AI startups. The people who run DeepSeek also run a quant hedge fund, and they made a 57% return in 2025 iirc. Those returns are then used to fund their operations and this is why they don’t rely on outside capital (VC et al).
Thanks for your comment Les, I'm going to have to read your article. Now that China has h200 access, they're going to push both open weight models and Enterprise AI much harder. DeepSeek is certainly a wild card. They can't be judged until their next flagship model is out. They're coding model in recent paper indicate significant innovation and new emerging software and model methods.
China made an incredible amount of progress in 2025 against fantastic odds with far less funding and vastly inferior GPUs. That's their startups have gone public before the big AI IPOs in the west is telling.
Everyone from economist to hobbyists are analyzing the business model of OpenAI, and it's only going to get worse for them.
More signs of desperation - signing up 30+ universities to pay site license fees that cover use of 'ChatGPTEdu' by all students and faculty ("Amazing and welcome move!!", said no faculty member ever), and, considering taking a cut of any IP resulting from inventors using OpenAI!
The scale of the pretending is reaching epic and fraudulent levels. But distribution and the trust of businesses isn't something you cheat to win, it's something you earn. There's something clearly wrong with the business strategy of this leadership team and the products they are building.
The scale of the capital funding isn't proportionate to the end result of the products, and it's alarming a world that's moving faster in AI than ever before. A world that isn't sure generative AI is even a game changer.
This was excellent - thorough and objective. There’s been so much hype that it’s good to see sober eyes analyzing the underpinning economics of OpenAI as well as the progress of its competitors. What strikes me as interesting is that Microsoft hasn’t been eying the solid position that Google is in with the integration of AI into their cloud ecosystem. For those large potential customers who need proven FED RAMP or HIPAA level environments, it’s quite a combination to offer. I most likely may be missing Microsoft’s offering in this arena, all I know about is CoPilot, but Microsoft has a similar environment that could, with the right AI offering, be very compelling. They seem to be content to just supply the shovels and pans will OpenAI hunts for gold, but I would think they would pull OpenAI in, pair with their cloud tools and compete directly with Google. State and federal govt contracts using AI as solutions would be quite lucrative.
Great analysis! Like I said in my latest piece, this won’t be a “rising tide lifts all boats” scenario; we’ll see the winners and losers more clearly. Right now we’re in a “shoot first, ask questions later” type of market but I’m glad people like you, Gary Marcus, Ed Zitron, etc and even myself (who is nowhere near as known) are questioning the economics of AI.
I have a theory that there’s a non-zero probability OpenAI gets bought out by Microsoft in the future. The thing is with OpenAI relative to other LLM platforms & tech companies, they seem to be taking so many shortcuts in a bid to become profitable and it’s just messy. The other thing that made me question everything is Sam’s answer to Brad in his Bg2 podcast episode. It’s not reassuring to investors at all, and then Sarah’s comment about needing a government backstop was even worse. It became a PR nightmare.
It’d be silly for investors or anyone really to underestimate China, especially DeepSeek and Alibaba. Deepseek’s business model is self funding which puts them in a far more unique position than most AI startups. The people who run DeepSeek also run a quant hedge fund, and they made a 57% return in 2025 iirc. Those returns are then used to fund their operations and this is why they don’t rely on outside capital (VC et al).
Thanks for your comment Les, I'm going to have to read your article. Now that China has h200 access, they're going to push both open weight models and Enterprise AI much harder. DeepSeek is certainly a wild card. They can't be judged until their next flagship model is out. They're coding model in recent paper indicate significant innovation and new emerging software and model methods.
China made an incredible amount of progress in 2025 against fantastic odds with far less funding and vastly inferior GPUs. That's their startups have gone public before the big AI IPOs in the west is telling.
Everyone from economist to hobbyists are analyzing the business model of OpenAI, and it's only going to get worse for them.
Nice analysis!
More signs of desperation - signing up 30+ universities to pay site license fees that cover use of 'ChatGPTEdu' by all students and faculty ("Amazing and welcome move!!", said no faculty member ever), and, considering taking a cut of any IP resulting from inventors using OpenAI!
The scale of the pretending is reaching epic and fraudulent levels. But distribution and the trust of businesses isn't something you cheat to win, it's something you earn. There's something clearly wrong with the business strategy of this leadership team and the products they are building.
The scale of the capital funding isn't proportionate to the end result of the products, and it's alarming a world that's moving faster in AI than ever before. A world that isn't sure generative AI is even a game changer.