원문 보기
Hello, and welcome back to the Failed Business Models channel. I'm your host, James, and in today's video, we'll be discussing Nvidia CEO Jensen Huang's recent interview. In my opinion, this is one of the most embarrassing interviews ever given by a CEO of any publicly traded company. Because, as we will show, Jensen Huang makes a number of claims that I believe are untrue .
I believe that they are significantly untrue . I am not a lawyer and I do not like to use foul language lightly . But when you are the CEO of a publicly traded company and you make statements to the public that are materially false and give investors a misleading picture about your company, I think that's a big problem and it needs to be exposed.
So the big question that came up during this interview is about all this building of AI data centers, huge amounts of money, hundreds of billions of dollars over the next few years; Hyperscale companies and neo-cloud platforms are expected to spend trillions of dollars building new data centers for AI.
And this is extremely profitable for Nvidia, because a large portion of its data center costs comes from purchasing Nvidia GPUs. And it's highly questionable whether there's actually demand to support this, given that OpenAI and Anthropologie's revenues represent only a fraction of the trillions of dollars expected to be spent on building AI data centers.
Which, of course, raises the question of how this data center construction will pay off and whether Nvidia's massive revenue growth is sustainable. So the host of the show, the host of the BG2 Pod, is this guy who owns a venture capital fund or a hedge fund or something like that. And his foundation published this chart that you see here, this bar chart , showing expected data center capital expenditures compared to the revenues of OpenAI, Anthropic, and Grok, and a few other companies that they call AI lab revenues.
If you look here, you'll see that total data center spending is expected to be nearly $800 billion in 2026. And this hedge fund expects OpenAI, Anthropic and other AI labs to generate only $105 billion in revenue. Thus , to cover the costs of building data centers, OpenAI and Anthropic's revenues will need to grow significantly in the coming years, and it is questionable whether this will actually happen.
He expects total annual revenue to reach $200 billion by the end of the year. That's $200 billion in annual AI revenue. The idea is that revenue grows rapidly throughout the year so that it reaches $200 billion by the end of the year, but in reality, only $105 billion was generated for the year . In my opinion, these figures are too optimistic .
Anthropic's S-1 recently leaked that Anthropic raised $ 33.1 billion in the first nine months of 2026 . If we annualize the third quarter figures , their annualized revenue for the third quarter would be $67 billion. As for OpenAI, information recently leaked to the media that as of October 2026, that is, just a couple of weeks ago, OpenAI's annual revenue was just under $50 billion .
A month ago, they told the media that their annual revenue was $70 billion. It turned out that this was false information. The real figure for OpenAI is $50 billion . This is the annual revenue figure as of October. In fact, it's a little less than $50 billion . Media reports say it is approaching $50 billion as of October.
It is also known that as of the end of 2025, OpenAI's annual revenue was $20 billion. If we do the math and assume a linear growth rate from the beginning of the year through October, this means OpenAI will have generated approximately $26 billion in revenue in the first 9 months of 2026. Thus, Anthropic and OpenAI received approximately $60 billion in revenue in the first 9 months of 2026.
There is also Grok, which received meager revenue. And finally, there's Google Gemini, whose revenue is very difficult to estimate because Google doesn't disclose that data separately. But Grok shows paltry revenue, and Gemini is probably much, much smaller than Anthropic or OpenAI. So if OpenAI and Anthropic only made $60 billion in revenue in the first 9 months of 2026, then annual revenue will likely be significantly less than the $105 billion that the guy from BG2 is projecting.
But even if we take his figures, which I think are too optimistic , right? OpenAI and Anthropic earned $105 billion this year. Next year, however, spending on data centers is expected to reach $1 trillion . Thus, revenue would have to increase by approximately an order of magnitude from current levels to justify building data centers.
The podcast host, this guy from BG2, asked Jensen Huang about this. Let's listen to what he said. Basically, there are three segments. Three big segments—that's probably a little more than three, but let me just break it down into three big segments. The first segment of our computing power is core computing, meaning we use it for data processing, molecular dynamics, physics simulations , fusion simulations , fission simulations—you name it, right?
Quantum chemistry, quantum physics, and so on. Fine. The world of modeling and data processing, not to mention computer graphics . And all this processing is done on our GPUs . If you were to say that there is no AI at all, and there is only classical machine learning, for example, Extreme Boost in Visa , MasterCard, fraud detection systems, recommendation systems that determine what news to recommend to you, it’s just all of this put together.
Yeah. Classical computing, not AI, not deep learning, not agent systems—I would call that a third of our business. Yes. Yeah. AI labs, people see two, but we work with all of them. And MetaMuse shows fantastic results, right? Incredible. And it's mostly videos, and, you know, last year there were zero of them.
Right. You know, a year ago there were only recommended systems, and now they, you know, are just tearing everyone apart with Meta Muse. And of course, Grok didn't exist 3 years ago. We didn't use Gemini before, we use Gemini now. And this year we started working with Anthropic. So if you just look at our computing power, people see a couple of labs, but I see more.
I see five, six labs, okay? And so all these labs are working great. Right. And if you told me that it's not quite 1/3 of our business, say a quarter of our computing power, that would be fantastic. And the other third is what's really cool. Because we are at the beginning of a new shift in the computing platform.
There's just a huge number of new startups that are popping up. So many great companies. And some of them, of course, are in the consumer goods business, you know, one of them is really doing well . Noah Instinct, I was just about to mention that. Noah excels in all types of physical AI, biological AI, and materials science AI.
Jeff Bezos's new company, Prometheus, is doing some fantastic things. And so the number of companies that have been created in a wide variety of industries is in the thousands . And what do they do ? They rent out Nvidia computers. That is, they are tenants. And so a third of our business is kind of next-generation cloud computing .
Right. Jensen Huang does try to downplay his reliance on Open AI and Anthropologie for income. And he says that he divides Nvidia's current business, their current revenue, into three categories. He says the first category, which currently makes up 1/3 of their revenue, is what he calls basic computing.
Basic computing power is power that has nothing to do with AI in the context of this whole boom in graduate programs in the humanities and social sciences over the past few years. The second category , which he attributes less than 1/3 of revenue to, is what he calls cutting-edge labs, and he specifically includes Meta, Gemini, Grok, OpenAI, and Anthropic.
And the last category, which he says accounts for 1/3 of the revenue, is all these small AI startups that are building AI-powered applications. And these small startups predominantly use open source models because open source models tend to be significantly cheaper than closed source models. So it's a separate revenue stream.
There are data center service providers that will run open source models for you on your behalf, and so these startups generate computing demand that is independent of the cutting-edge labs because they are using open source models, right ? The whole point that Jensen Huang is trying to frame this is because he wants to say: people are saying that companies like OpenAI and Anthropic are not going to meet revenue expectations, and that will be bad for Nvidia.
Jensen Huang replies, “No, these big, you know, cutting-edge labs make up less than 1/3 of our revenue. In fact, most of our revenue comes from other sources, so we are quite diversified." Now I say that Jensen Huang is lying here. In fact, a large portion, if not the vast majority, of Nvidia's revenue comes from its cutting-edge labs.
The underlying computing power in these open source startups, in my opinion, accounts for a very small percentage of Nvidia's revenue, and I'll explain why. Let's start with the cutting-edge laboratories. Well, one of these cutting-edge labs is Meta. Jensen Huang is really trying to promote the new AI assistant Meta Muse that Meta just launched.
And here's a direct quote from Jensen Huang: "Meta Muse works fantastically. Last year there was nothing. These were just recommendation systems. And now they're just tearing everyone apart with Muse." So what he's trying to say is that since Muse just launched and is so successful, that means Meta will have to buy a lot of new Nvidia GPUs just to run Muse.
Before this, Meta hadn't done anything with AI. They were only involved in recommender systems. True, Meta has been using recommender systems for a long time, and the recommender systems used Nvidia GPUs. However, it should also be noted that since 2022, Meta has begun a massive capital investment program, spending over $100 billion to build new data centers specifically for LLM.
They developed their LLMs for Llama and then Muse Spark. Jensen Huang is trying to make it sound like Meta didn't do anything with LLMs before, and just launched Muse a couple of months ago, and now that they're actively pursuing the LLM business, they'll need a lot more GPUs. In fact, it's the other way around.
Since Meta has already spent huge amounts of money on building new data centers with Nvidia GPUs, they have accumulated a huge amount of unused capacity. And we know this because a couple of months ago, in July, Meta briefly considered leasing out its unused capacity to other AI companies. But in the end they abandoned this idea.
They decided to keep their power for themselves and will use it to run Muse. They won't need to buy new Nvidia GPUs for Muse because they already have underutilized capacity. So if Muse becomes a huge success and is very popular in the future, Meta may need to buy even more GPUs. But Jensen Huang's claim that revenue was zero last year is simply a lie, because Meta is purchasing massive amounts of GPUs specifically for LLM from 2023 to 2022.
Now, I think the most egregious part of Jensen Huang's statement or narrative is the idea that 1/3 of their revenue comes from core computing, which has nothing to do with AI. This is clearly not true. Nvidia currently reports two revenue segments. One segment is called data centers, and these are essentially big GPUs that they sell to data centers that use AI.
The other segment is everything else. This includes gaming GPUs, such as those for gaming PCs, some GPUs used for automotive software, and professional visualization. If we look at the period before the launch of ChatGPT, which was in Nvidia's fiscal year 2022 (Nvidia has an unusual fiscal year, meaning 12 months ending in January 2022), ChatGPT didn't appear until November 2022, so that was the pre-ChatGPT period.
The data center segment accounted for 39% of Nvidia's revenue, while everything else accounted for 61%. The "everything else" segment is mostly gaming GPUs. So, before ChatGPT, the majority of Nvidia's revenue came from gaming, and data centers from a minority. Now, if we look at the trailing 12 months ending July 2026, 92% of Nvidia's revenue now comes from the data center segment, and everything else only comes from 8%, right?
Jensen Huang claims that 1/3 of Nvidia's total consolidated revenue is not related to AI. This requires that a significant portion of the data center segment be non- AI related. But this statement is not credible because if we look at Nvidia's data center segment, you will see when ChatGPT was released.
ChatGPT was released in November 2022 , which is the fourth quarter of Nvidia's fiscal year 2023. Almost immediately after the release of ChatGPT, Nvidia's data center segment saw a sharp increase, and this is clearly a sharp jump in growth rate. So , it's clear that the vast majority of the data center segment is tied to this AI boom.
Basic computing power accounts for only a tiny percentage. Thus, in the 12 months prior to ChatGPT's launch, Nvidia's data center segment generated $15 billion in revenue . However, some of these funds have already been directed towards meeting the demand for AI. Although ChatGPT was released in November 2022, training of AI models began much earlier.
In July 2019, Microsoft invested $1 billion in OpenAI. Much of that money ended up going to Nvidia to buy the GPUs that OpenAI used to train ChatGPT before it launched. Also in July 2021, Anthropic raised $124 million in a Series A funding round, which was also indirectly spent on GPUs. And then in 2022, Anthropic raised a much larger funding round, around $500 million , a significant portion of which also went towards GPUs.
So even before ChatGPT, a significant portion of Nvidia's data center revenue still came from training AI models. And, obviously, the explosive growth we've seen since the launch of ChatGPT is largely due to data centers using AI. Jensen Huang's claim that a third of their revenue is not related to AI is patently nonsense if you just look at the chart and see when revenue started to spike.
Now, Jensen Huang's final argument: he claims that 1/3 of Nvidia's current revenue is from open source models. The reason he tries to emphasize the benefits of open- source models is that OpenAI and Anthropic are not generating enough revenue to cover the boom in data center investment. So Jensen Huang says, "Well, it's okay, even if OpenAI and Anthropic can't pay for it, there are plenty of open-source models .
Therefore, many data centers will be needed to serve open source models . People spend a lot of money on open source models . Ultimately, this will cover the entire AI capital spending boom." However, this is not true, because if we look at the actual amounts of money spent on open source models, they are not that big.
And we can get data from two companies. There is a company called Vercel and a company called OpenRouter. Both of these companies are essentially middlemen, and if you're a developer and you want to develop some new AI-powered app, right? Or just doing something with AI, you can turn to Vercel or OpenRouter, and they'll connect you with a bunch of different data center providers that can provide you with closed-source models like ChatGPT, Claude, and Gemini.
They can also give you access to all the different open source models like DeepSeek, Kimmy and more. Now you can see that open source has really gained market share from closed source over the last year or so. However, open source models are much, much cheaper per token than closed source models. In October 2026, Vercel reported that open-source models now account for 71.5% of token consumption on their platform , but only 23% of total spending.
And that's only because open source tokens are much cheaper. So the bulk, indeed the vast majority, of the costs are incurred by closed source models. Both OpenRouter and Vercel attract developers looking to upgrade their models and chase price. Therefore, both companies likely have a higher share of open source models than the market as a whole.
Many businesses only use ChatGPT or Claude and are therefore not included in the Vercel or OpenRouter datasets. Right? So Open Router and Vercel actually favor open source because enterprises or companies only use closed source models, right? If your company only uses ChatGPT or only Claude, it will not be included in this Open Router or Vercel dataset.
Therefore, Open Router and Vercel actually overestimate the percentage of demand for open source models due to the specifics of their customers. And even for them, only 23% of their spending is on open source models . Furthermore, a very large portion of the computing resources consumed is used for training new models rather than for inferring results on existing models.
Training is overwhelmingly paid for by cutting-edge labs such as OpenAI and Anthropic . Thus, open source models probably account for much less than 1/3 of the costs of closed source models. Therefore, the total amount of money paid for open source models is a small fraction, probably less than 1/3, of the revenue generated by OpenAI and Anthropic.
So, if we go back to Jensen Huang's three main claims , we see that each of them is false. Right? Basic computing accounts for far less than a third of their income. Open source models also don't make up even a third of their income. This means that the bulk, probably the vast majority, of the income comes from these cutting-edge labs, mainly Open AI and Anthropic, which are losing huge amounts of money and will not be able to generate enough income to cover the costs of building AI data centers.
Jensen Huang knows this, but he needs to fool his shareholders into thinking that demand is more diverse than it actually is. But it's important that we look at the numbers and explain why what Jensen Huang says can't be true. When you have a guy like Jensen Huang, his net worth is in the hundreds of billions of dollars.
He is the CEO of the most valuable publicly traded company in the world . One would assume that there are some legal requirements for his honesty, which is what he does, right? Lying about your company is fraud. It would seem that he simply cannot lie to us, right? But if we compare his words with the facts that we can see from open sources, it becomes clear that he is lying.
So do n't be fooled by someone like Jensen Huang. This concludes everything we talked about today. If you have any opinions or thoughts about Jensen Huang, please share them in the comments below and see you in the next video.
Comments 0