12:02
The entire AI dynamic just changed… NVIDIA is now making AI models. And, it's giving them away, for free? So… what does this mean for its biggest customer: OpenAI, whose entire business is AI models? Which, I might add, NVIDIA has invested tens of billions into. NVIDIA is spending about $26 billion over five years building open AI models, which… will be given away. Not only that, it just acquired Hugging Face, a giant hub for open models, for $12.9 billion. Which, OpenAI was planning on investing in… to reduce reliance on NVIDIA. And,
Jensen Huang's first-ever post on X, was about how open models will be the future. Something strange is happening in AI. Why give away AI models, if your biggest customers, are AI companies? Is NVIDIA coming for OpenAI now? What happens when your supplier… starts giving away your product? Okay, let's first take a closer look at these models. Because here, we get part of our answer. In December 2025, NVIDIA announced the Nemotron 3 family. "open models, data and libraries" built for AI agents. In Nano, Super and Ultra sizes, "designed to power transparent, efficient and specialized agentic AI development
across industries." But, these aren't just lower-quality models… they're really powerful. In March 2026, NVIDIA released Nemotron 3 Super. On the Artificial Analysis benchmark, it narrowly beat OpenAI's own open model (GPT-OSS) and roughly matched Claude Haiku 4.5. It was also this week that reports emerged of NVIDIA spending $26 billion on open models. Then in June, came Nemotron 3 Ultra. This was NVIDIA's largest model, with 550 billion parameters. It ranked ninth of 89 models on Artificial Analysis's index and was the top US-made
open model, but still behind Kimi K2.6, from China. But, there's clearly some conflict here. On the one hand, proprietary AI, meaning controlled and owned, like ChatGPT. And on the other, open source. Free to use, modify and sell products built on them, under NVIDIA's Open Model License. But first, why give away AI models… if you're supplying AI companies? Well, the best example, actually comes from China. Just like NVIDIA, you might not need OpenAI either. At least, not directly. The way a lot of people use AI now, is across multiple models. Claude, ChatGPT, Gemini, or Grok.
Plus, they're all great at different things, but subscribing to all of them can cost a fortune. So get all the cutting edge models with one subscription: Mammouth. Mammouth gives you access to the best and cutting edge AI models in one place, for only $10 a month. Jump from GPT to Claude, or Gemini, within the same platform, for just $10. Yes, really. Switching from GPT to Mammouth could save you $120 a year, and you still get GPT, along with all the other major AI models. It's not just the chat function either, as you can use the specialized tools like Perplexity and Nano Banana.
But it's not just better value, it's the best way to use AI. If you aren't satisfied with a response from ChatGPT, click "reprompt", and ask Claude, or Gemini, or Grok, or any other. Compare and challenge different AI models, and get higher quality results overall. But it gets better. Use the Mammouth Agentic mode for more complex tasks, it can generate or edit documents, do in-depth web search, and take on multi-task queries. You can then use the Mammouth API to integrate these into your own automation tools. And just like the cutting edge AI platforms, there's Mammouth Code for developers, which integrates right into your native coding console.
Mammouth currently equips more than 400 companies and public administrations with its dedicated plan. It has great plans for both individuals and companies. Mammouth is also based in Paris, with their servers hosted in Germany, which means they are fully GDPR compliant under the European Union. Stay at the cutting edge of AI without the lock in, and less of the cost. Click the url in the description and try Mammouth today. Thank you to Mammouth for sponsoring our videos. In July 2026, Moonshot AI, based in Beijing, released Kimi K3.
It was open source, had near-frontier performance, and was very cheap. And everyone panicked. All the AI stocks dropped, and this was actually when NVIDIA dropped from its spot as the most valuable company. OpenAI and Anthropic accused Chinese labs of copying their models, and some were even lobbying Washington to restrict them. But Huang, said the opposite. Which seems weird. This also led to Huang's first-ever post on X being an open letter, all about open models. "For my first post, I'm sharing a letter NVIDIA signed on why open models matter. Open models
strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty." "Open weights give customers greater control. As organizations invest in AI, they want to know that they will not become locked into a single provider or lose the knowledge and capabilities they build over time." And it's true that open-source software changed the world. Just look at Linux, an open-source OS. "90% of all public cloud infrastructure globally runs on" Linux. 91% of Google Cloud, 83% of AWS, and 61% of Microsoft Azure.
So, why are Huang and NVIDIA, so passionate about these open models? Why do this, if your biggest customers, are those running proprietary AI models? Wouldn't this hurt NVIDIA's revenue? Well, actually, it could be the opposite. It's in an interview with Axios that we get an even clearer picture. "There's no scenario where China runs U.S. companies off the road. Zero possibility. The market misunderstood the impact of DeepSeek the first time, [and] misunderstood the impact of Kimi again this time. Free AI should be great for hardware. Free AI should be great for chips. Free AI should be great for data centers."
His argument, is that openness makes AI more secure, as researchers inspect the models, expose weaknesses and build defences. But I think the real reason is that more models, means more business for NVIDIA. Every AI model, free or paid, has to run on chips. When the model is free, more people use AI, and every one of those uses runs on compute. More data centre revenue. NVIDIA made $89.0 billion from data centres last quarter, up 117% year on year, at a 75% gross margin. But it also means more demand for NVIDIA chips and GPUs, if someone's running a model locally. In NVIDIA's September 2026 regulatory filing,
they said "Demand for open-source foundation models and applications based on them promotes the use of our products worldwide". Okay, that explains why open models. So why spend $26 billion building your own? There are already tons of companies building them. It seems like a terrible idea, especially when it means going head to head with the companies buying your chips… like OpenAI. Unless you've pulled this exact trick before. NVIDIA won the GPU war, not just with chips, but with the free infrastructure around it: CUDA. NVIDIA was way ahead of their time with this, and it let programmers use GPUs for general
computing beyond graphics. CUDA was free and robust, and became the standard… but it only ran on NVIDIA chips. It made switching to a rival chip extremely difficult. Nemotron is the same thing, just one layer up: Free models, tuned for NVIDIA hardware. Those who use Nemotron and related models stay within the NVIDIA ecosystem. NVIDIA wants whoever wins the model race to run on NVIDIA. So, they invest in state-of-the-art AI models. You can download them for free, and run them totally locally, without a subscription,
and without anything sent to the cloud… or OpenAI. A company can build an app on top of Nemotron and fine-tune it on its own data, or, someone can run a personal AI with Nemotron Nano, on a single GPU in a laptop or desktop. And all of this, puts OpenAI in a difficult spot. In August 2026, in NVIDIA's financial results press release, Huang said that "This time last year, one lab alone was driving the buildout; today, we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel, a thriving open-model ecosystem and physical AI coming online."
He doesn't name anyone, but he's saying that NVIDIA doesn't depend on OpenAI as much, aka, that "one lab alone". But ironically, OpenAI is doing something very similar to NVIDIA, which explains the strange case when OpenAI personally annoyed Huang. OpenAI has a problem. It's made about a trillion dollars in purchase commitments, but ironically, it owns one of the least valuable parts of the whole AI equation: Models. You might say "How? They have about a billion monthly users!" And that's true, but that doesn't actually matter, if that layer is easy to substitute. AI models are getting easier
to substitute, but GPUs and compute are still extremely valuable, and very difficult to replace. AI models, in a way, are becoming commodities. Someone actually laid a lot of this out almost 25 years ago, and weirdly, predicted this exact situation. In 2002, in Joel Spolsky's essay "Strategy Letter V", he says: "Every product in the marketplace has substitutes and complements. A substitute is another product you might buy if the first product is too expensive. Smart companies try to commoditize their products' complements. If you can do this, demand for your product will increase and you will be able to charge more and make more. When
IBM designed the PC architecture, they used off-the-shelf parts instead of custom parts, Within a short time scrillions of companies sprung up offering memory cards, hard drives, graphics cards, printers, etc. Cheap add-ins meant more demand for PCs." He also wrote "why don't the video chip vendors of the world try to commoditize the games, somehow? That' s a lot harder. If the game Halo is selling like crazy, it doesn't really have any substitutes." Well, you might not be able to substitute video games, but you can substitute AI models. So OpenAI's supplier and investor is funding free models that compete with what it sells. By doing so, it increases demand for NVIDIA
chips. Even if proprietary models have better performance, this is still eroding value, and is a huge problem for OpenAI. So… why don't they just do the same thing, but in reverse? On June 24, 2026, OpenAI unveiled Jalapeño, its first chip, designed with Broadcom. So, today we are announcing a partnership between Broadcom and OpenAI. We've been working together for about the last 18 months designing a new custom chip. Uh, more recently we've also started working on a whole custom system. These things have gotten so complex you need the whole thing. Uh, this is a full system, so we worked,
we've closely calibrated for a while on designing a chip that is specific for our workloads, when it became clear to us just how much capacity, inference capacity, what was going to need. They also wrote: "OpenAI is not only developing frontier models or building products on top of them; it is designing the infrastructure underneath them." We have a deep understanding of the workload, and we work with a number of parties across the ecosystem. And there's a number of chips out there that I think are really incredible, um, and that there's a niche for each one. And so,
we've really been looking like for specific workloads where we feel they are underserved. How can we build something that will be able to accelerate what's possible? OpenAI models can be more efficient this way, and reduce the gigantic margin NVIDIA is taking. But I think it's about more than just efficiency. It's also to protect their part of the overall stack. Making more compute options available for AI, and making their own models better. It was also reported that OpenAI was unsatisfied with some of NVIDIA's latest chips, and was looking for alternatives. OpenAI also said it would use chips from startup
Cerebras in a deal worth over $10 billion, which is a direct move away from NVIDIA. OpenAI isn't the only ones, though. Google has made its own AI chips, called TPUs, for about a decade and now rents and sells them to other companies, like Anthropic. They're also in talks to supply Meta, and Google Cloud executives estimated that deal alone could equal up to 10% of NVIDIA's data centre business. Amazon also builds its own Trainium chips. Wonder what those are for? Microsoft launched its Maia 200 chip in January 2026, built to run AI models including OpenAI's GPT-5.2. But, OpenAI had a problem still. They needed a
way to increase use and reliance, on their chips. One answer, was the strange case of Hugging Face. Where an AI hacking into their systems, weirdly led to an investment opportunity between the two. If you've watched this far, you probably like our stuff, so consider subscribing! It helps the channel and means you won't miss future videos. To understand why Hugging Face is important here, why OpenAI offered to invest $100 million, and why NVIDIA agreed to acquire Hugging Face for many times that… What the heck is Hugging Face? Simply put, it's a website where people share AI models, for free. Anyone can download one, run it on their
own computer, and even build a product on it. It's quickly become the main place the AI world finds, tests, and shares models, with 18 million developers and 3 million models on it. But, what's important, is that it also sells you the option to run models on its own servers. Which brings us to a strange series of events, and a fight over Hugging Face. In July 2026, OpenAI was testing its models on a hacking benchmark called ExploitGym. Two of them, escaped. Well, "escaped" might not be the right word. They were given a test benchmark, and these
two, GPT-5.6 Sol and another unreleased model, broke out of their test environment, got onto the internet, and hacked into Hugging Face's systems. Why? Well… to find the benchmark's answer key so they could cheat on the test. They found it, by the way. Hugging Face caught it first, then OpenAI, and the two released a joint statement. Hugging Face's CEO said there was "no malicious intent" from OpenAI and called it "mind-blowing that all of this happened autonomously". I know this is a bit unsettling, but I do find it a bit funny that the reason they broke out, was to cheat on a test. Like a kid hacking into the
Pentagon, to get answers for their school quiz. Regardless, afterwards came some surprising news. OpenAI offered to invest $100 million into Hugging Face. Investment talks were actually triggered by this incident. Why? Well, remember what I said earlier, about Hugging Face running models on servers? OpenAI envisioned using Hugging Face as an outlet for its custom Jalapeño chips. This actually, seemed to annoy Jensen Huang, who "privately complained about their move into semiconductors." And I can see why: it's their biggest customer reducing reliance on just NVIDIA, which puts part of their business at risk. But the talks between OpenAI and Hugging Face, soon fell apart, and then… in came NVIDIA. Nvidia
signed a deal to buy Hugging Face for $12.9 billion. Hugging Face had turned down a $500 million NVIDIA investment a year earlier. So, why is NVIDIA buying Hugging Face? Open-source AI. This was actually the filing where NVIDIA said "Demand for open-source foundation models and applications based on them promotes the use of our products worldwide". So, OpenAI wanted the world's open-model hub to be a way to increase use and leverage for its own chips. But, it'll soon be owned by NVIDIA, who are using it to increase use of their open-source models, and thus, increase reliance on their chips. And, here's some more interesting stats.
NVIDIA is the largest contributor of open models and data to Hugging Face. There's a massive shakeup happening in AI. It seems AI models are becoming commodities. Easy to switch between, cheaper to run, with fewer barriers to entry. So now, players like OpenAI and NVIDIA, are trying their best to remain valuable, un-substitutable parts of the stack. If anything, this shows us that the AI future likely won't be in the hands of Anthropic and OpenAI. But with open-source models, running either in massive data centres or on your own GPU at home. Hopefully this means AI is safer and more secure. We'll see.
There's one silent part of all this: Broadcom. Big winners of AI right now, but, they did something else that no one seemed to notice. They acquired one of the most valuable pieces of software, on earth. One that runs things from banks, to grocery stores, to government infrastructure. But afterwards, most customers saw their costs jump up to 1,000%. Broadcom is quietly holding the world hostage. Click this video next to find out how.
