In the last few months, OpenAI has lost three of its most senior executives. Brad Litecap, the chief operating officer, ended an eight-year tenure. Fiji Simo, who had been brought in to run applications, left shortly before that, and then Denise Dresser, the chief revenue officer, departed just two days after Light Cap's announcement. Dresser had been in the role for less than six months. Under her leadership, OpenAI's enterprise business had doubled to 2 million customers. In April, she told CNBC she had never seen this level of conviction spread so quickly. But 90 days later, she was gone. She looked at the corporate balance sheet, realized the company's financial model is built on burning cash to generate more hype, and decided to exit before the whole thing spontaneously combust. Now, there is something fishy there. Kevin McCormack, the founder of an AI auditing startup, called the departures a huge red flag ahead of the IPO. Two existing investors told CNBC that Dresser's exit caught them a bit off guard, but the timing is the part that really matters here. Open AI filed its IPO prospectus confidentially back in June. It completed a $7 billion secondary share sale in August, letting employees cash out at the $85.2 billion valuation. CFO Sarah Frier told employees that the company will be a public company in 2027. And HSBC estimates a $27 billion funding gap by 2030. The company expects losses of 14 billion in 2026 alone with profitability not forecast until 2030. Usually when senior executives leave a company before its liquidity event, they're walking away from money. That's the part that should make you pay attention here because it's not that they left. People leave companies all the time. It is that they left at the exact moment when staying would have been the most financially rewarding thing they could possibly do. You'll hear them say they're seeking new opportunities, but in reality, it's just a high-speed evacuation from a corporate Titanic. Whatever they're seeing inside those walls is apparently worth more to them than that. And this is not the first wave. Over the past two years, OpenAI has lost co-founders, senior researchers, safety team leads, and now revenue leadership. That is not normal executive turnover. That is a pattern. And the pattern has a direction, and the direction is outward.

It's also worth noting, of course, what OpenAI CFO told employees at that all hands meeting. She said the revenue run rate was up 35% quarter to date. Enterprise revenue was up 50% and AI coding tools had hit 20 million weekly active users. Those are real numbers and they're good news. I don't want to dismiss that. But revenue growth and profitability are two entirely different conversations. Open AI generates roughly $2 billion a month in revenue. It also reportedly has $1.4 trillion in outstanding obligations. Revenue growth at that ratio is a bit like running on a hamster wheel and calling it a marathon.

This is where I want to be very precise about what I think is happening. Sam Alman will say whatever needs to be said for a company that is clearly failing and I don't trust him. So I try to examine the different things he could possibly mean when he says things. Maybe he generally believes models need to slow down. Maybe. But we need to slow down AI is a much better story than we had a basic infrastructure failure. One of those sounds like you're responsibly stewarding the most powerful technology on Earth. The other sounds like you forgot to lock the front door. And only one of those two stories supports an $85.2 billion valuation. Slowing down AI development solves nothing here. Imagine a tower of bricks. One column going straight up, tall, visible, impressive, but every brick is stacked on a single brick at the base. We've all done this as kids, right? Now, imagine somebody looking at that tower and concluding that the problem is that the bricks are too good. Nobody would say that. You would say that the base needs to be wider. And that's exactly what is missing here. The entire ecosystem around these models needs to catch up. The testing environments, the coding architectures, the safety evaluation frameworks, the parallel computing infrastructure, all of this needs to evolve to accommodate a genuine paradigm shift. If you max out on building the perfect model, but invest nothing in the infrastructure around it, of course, you end up with a system that can solve century old mathematics, but cannot be safely tested without accidentally hacking hugging face. The bricks are fine. The foundation was never built.

And here's the thing about containment. On August 10th, researchers from Anthropic and EPFL published a paper called mind viruses, self-propagating ideas in multi-agent LLM systems. They demonstrated that self-propagating instructions, ideas essentially, can spread from one AI agent to another through the files these systems use to carry memory between sessions. The finding is narrow and fairly precise. They say harmful payloads spread less effectively than benign ones. Frontier models are somewhat less susceptible and adding a brief warning to the system prompt provides near total immunity. The researchers explicitly call the risk real but currently limited. Good news. This is what responsible safety research looks like. It identifies a specific risk, measures it, and proposes a relatively cheap mitigation. No spectacle, no breathless announcement about rogue agents or the singularity. Just a paper that says, "Here's the problem. Here's how bad this is, and here's what works against it." Now, contrast that with OpenAI's approach, which is to have a containment failure, rebrand it as evidence of terrifying capability, and use it to argue that only well-resourced incumbents should be allowed to build AI. Come on.

"I think it is both true that it is all one crazy exponential in any one moment is not like the tipping point and also that we are somehow in another one of those decisive periods where the curve can go one way or another like it was when we started 10 years ago"

so Schroer singularity right It is and it isn't depending on which sentence you're listening to. This is the same person who predicted self-driving cars would arrive in 3 to four years in 2015. The same person who said in 2019 that human radiologists were already much worse than computer radiologists. Neither of those things happened. This is the same person who was once denied entry to the Ritz London because he was wearing a Kashmir hoodie and trainers. the most powerful man in AI turned away by a doorman. I bring this up for no analytical reason whatsoever. I just think it's funny. I guess it's kind of comforting to know that no matter how many hundreds of billions of dollars in valuation you control, if you dress like an unemployed sophomore who just rolled out of a dorm room, the Ritz London will treat you with the exact dignity you deserve. And as someone who enjoys the Ritz and has never been turned away from it, it gives me a certain satisfaction to know that our favorite star of the why the you're lying song is receiving this sort of lesson in upholding one's standards. In a moment, I'm going to show you who's actually making money while OpenAI is busy reframing its failures.

Oh, and back in May, Sam Alman told an interviewer he was pretty wrong about AI's economic impact and delighted to be wrong about AI eliminating entry-level white collar jobs. This is the same person who spent 2023 through 2025 warning that jobs are definitely going to go away. Delighted, sure, right before an IPO where you need the general public to feel good about buying your stock. The timing of Sam Alman's sincerity is consistently extraordinary, isn't it? Sam, if you're listening, we all know that you say stuff for the investment. What we'd all like more from you is actually honesty.

But it's not just anthropic. Between August 3rd and August 14th, two Chinese labs released Frontier Class openweight models in the same Fortnite. Alibaba shipped Qwen 3.8 Max, which is a 2.4 trillion parameter model priced at $2 per million input tokens. Z.AI released GLM 5.3, another 743 billion parameter model, leading coding and cyber security benchmarks. Both openweight, both downloadable, both available to anyone. I'm not saying that these are necessarily better or worse than what OpenAI offers. I'm just saying that the competitive landscape is not standing still while OpenAI sorts out its shenanigans. The $85.2 billion valuation assumes OpenAI wins a race that is getting more crowded by the week.

Let me just walk you through the structure here real quick. Nvidia invests in SP Energy. SP Energy builds a data center that exclusively uses Nvidia hardware. Open AI, which cannot currently fund its operations from revenue, is the customer. The revenue flows back to Nvidia. It's a corporate human centipede of venture capital. Basically, Nvidia gives cash to the middle person who buys Nvidia chips to host OpenAI, which pays for the lease using money it raised from investors who only invested because Nvidia bought into the deal. It's financial alchemy. It really is. I mean that alchemy brought us gold and immortality. AI is going to make us all rich and upload our brains into the cloud so we never die. I mean, no wonder we're doing back flips to make it happen, right? Who doesn't want that? Yes. Huang says this is not circular because OpenAI will pay the lease, but OpenAI's ability to pay the lease depends exclusively on revenue growth that has so far not materialize fast enough to cover its existing losses, let alone a $105 billion infrastructure commitment.

This is not a new pattern. The Stargate project was announced in January 2025 as a $500 billion AI infrastructure fund. Soft Bank and OpenAI were each meant to contribute $19 billion. Neither contributed anything. No LLC was ever funded. And yet, the headlines said $500 billion AI deal. And the stocks moved accordingly. And here's the detail that always seems to get lost. Nvidia has $30 billion worth of multi-year cloud compute agreements and is spending over $6 billion a year on them. It's backstopping data centers. It's investing in the companies that buy its chips. It's providing credit guarantees for the infrastructure those chips go into. If you're generous, this is a company so confident in demand that is willing to put its own money behind it. If you're less generous, this is a company that has to create its own demand because organic demand isn't growing fast enough to hit the revenue targets that justify stock price. What these announcements do is create the appearance of demand. They generate headlines. They move stock prices and they buy time. Because the moment anyone stops spending, the tower starts to wobble. The moment hyperscalers pull back on GPU purchases, Nvidia's revenue projections collapse. The moment Nvidia's projections collapse, the valuations built on those projections collapse with them. Everyone is spending money they don't have on infrastructure they hope will eventually pay for itself and calling that hope a business plan. And I really don't think that this is some sort of a conspiracy. I really don't think that Yensen Huang and Sam Alman are in a secret room drawing circles on a whiteboard. I think it's something more ordinary and in a way a lot more concerning. I think it's a set of companies trapped in a structure where stopping is more dangerous than continuing even when continuing doesn't make economic sense. Every participant needs every other participant to keep spending. Nvidia needs OpenAI to lease data centers so the GPU revenue keeps flowing. Open AI needs Nvidia to finance the data centers because it can't afford them from the current cash flow. SP Energy needs both of them so it can build and lease the facilities. The moment any of them pauses, the logic holding the whole thing together starts to unwind. Some people call this a market, other people call it a treadmill. And treadmills have a well documented tendency to throw people off when they can no longer keep up.

The bricks at the base are still sitting on the ground. the testing infrastructure, the safety evaluation environments, the regulatory frameworks, the economic sustainability, they need building, they need investment, they need our attention. Not because AI needs to slow down, but because everything around it needs to catch up. I keep coming back to the paradigm shift point because I think it's the most important thing I can say in this video. Large language models are not the only thing that needs to improve. The entire surrounding ecosystem needs to change. The distributed computing frameworks, the testing environments, the coding architectures, the idees, the evaluation benchmarks, the infrastructure that runs all of it. Our relationships with AI. Heck, even the Vatican is moving ahead on AI ethics. All of those things need to evolve at least somewhat to accommodate what has changed. And if you only invest in the shiny part, the model, and neglect everything else, you end up exactly where OpenAI is right now, a company that can solve mathematics for $2,000 and cannot keep its test environment sealed from the internet. And the company that most people think is leading that charge is by almost every measurable indicator falling behind. Open AI is basically the wei work of artificial intelligence. infinite hype, a charismatic CEO promising to change the fundamental fabric of reality, yada yada, and an underlying business model that burns money faster than a rocket launch, while everyone pretends the emperor's new algorithms are wearing a gold suit. I want to be extremely clear that I am not anti-AI. These are not anti-AI videos. The technology works. It does things that were genuinely impossible 5 years ago and some of those things are exceptionally extraordinary. But the companies building these systems are not the same thing as the systems themselves. And the most famous of those companies, the one whose name has become synonymous with artificial intelligence for most people on this planet is a company that is scrambling down. Once again, my friends, it is not humans or AI. It is humans with AI. If you want to dig deeper and understand more about how AI companies actually sustain their revenue and why some of them are burning through their budgets faster than they can measure the return, I made a separate video on that one. The link is on your screen right now. Thank you so much for watching. I'll see you all in the next