The Nvidia Groq Acquisition Explained

Well, it's the last news roundup I'll be making of 2025 and I was in no way expecting anything too crazy this week. But then Nvidia went and absolutely blew the roof off with their largest acquisition in the company's history, which also makes it the largest acquisition in AI history by buying Grok for $20 billion. Or did they stay tuned to find out right now, On Christmas Eve, while most of us were distracted by the holidays, Nvidia quietly made its biggest move in its 32 year history. They agreed to pay $20 billion to the AI chip startup Groq. That's Groq, not to be confused with Gro K from Elon. But here's the thing, they're not actually calling this an acquisition. And that little distinction might be the most important detail of this entire deal. So here's some important background to really understand what's going on. Grok, these special types of AI chips called LPUs or language processing Units. Now if you've ever used Chat, GPT or like really any AI chatbot, you've noticed that sometimes it takes a little while to get a response. It's not instant. That's because most AI systems run on Nvidia's GPUs or graphics processing units. Now these are incredibly powerful, but they weren't originally designed for this specific, specific task of generating text response. They were designed for gaming. Now Gro were built from the ground up for that text generation process after you give it a prompt, also called inference. The result of this, Grox chips can run these AI models up to 10 times faster while using 10 times less energy. And the man behind this technology is someone named Jonathan Ross, and he's a key player in all of this. He's the founder and CEO of Grok. But before starting at Grok, Jonathan Ross actually worked over at Google where he helped create the TPU or Tensor Processing Unit. This is Google's own custom AI chip and it's become one of the biggest threats to Nvidia's dominance. Which actually brings us to why this deal is happening right now. You see, Google recently unveiled its seventh generation TPU called Ironwood and and then released Gemini 3 shortly after, which is their newest AI model. That kind of blew the AI world away. But here's the thing. Gemini 3 was trained entirely on Google's own TPU chips. No Nvidia hardware was involved at all. And this is huge because it proves that a company can build world class AI without depending on Nvidia. Google stock went up Nvidia stock went down, and for the first time, investors saw a credible path to to an AI future that doesn't flow exclusively through Nvidia. But Nvidia, they're smart. What did they do? They went out and they hired the guy who built Google's TPU in the first place. All right, now let's talk about how this deal was structured, because this is another thing that's super interesting about all of this. Nvidia isn't technically buying Grok. Instead, they're calling it a non exclusive licensing agreement. They're paying $20 billion to license GRO technology and hire Jonathan Ross, along with the company's president, Sonny Madra, and other key team members from Grok. So Grok on paper continues to exist as an independent company under a new CEO named Simon Edwards. And their cloud business is continuing to run. But then that begs the question, why structure it this way? Well, it all comes down to one thing. Regulation. Back in 2022, Nvidia tried to buy a chip company called ARM for $40 billion, and regulators blocked it. An outright acquisition of Grok would likely face the same kind of scrutiny, especially with Nvidia already controlling over 90% of the AI chip market. But by calling this a licensing deal instead of an acquisition, Nvidia avoids triggering that antitrust review process altogether. One analyst put it pretty, pretty bluntly. The deal structure keeps the fiction of competition alive. And Nvidia isn't the only company using this playbook. Just a few months ago, Google did the exact same thing with an AI coding startup called Windsurf. They paid $2.4 billion to license the technology and hire the CEO and key figures from that company, all while technically not acquiring the company. Microsoft did it with inflection. AI. Amazon did it with adept. The this is actually becoming a standard play in the big tech playbook. But there's also some downsides about this specific play from the playbook. And that's when it comes to what happens to the actual employees at these companies. You know, the people that aren't part of the leadership team or aren't necessarily investors. So in a traditional acquisition, when a company gets bought, employee stock options typically get converted or cashed out. Employees who joined early and took lower salaries in exchange for equity often get a life changing payday. That's why they're willing to take lower pay in exchange for equity. Work smarter, not harder. However, in a licensing deal like this one, that's not necessarily the case. So the research into some of these reverse aqua hired situations actually shows that employees with stock options, especially unvested stock options, often get nothing from deals like this. No actual acquisition means no equity conversion, no payout. The founders and executives who got hired by the acquiring company, well, they walk away with massive comp packages. The investors, well, they're getting the returns from the licensing fees. But the rank and file employees who helped build the company, well, they may just get left out entirely and be employees of what's essentially a new shell of the original company. Not a great look. And this has actually become a real pattern. Reports from the similar deals like Windsurf and some of these other startups actually suggest that a bunch of the employees didn't receive payouts, despite what actually looked like a successful exit for the company. So, yes, this deal is being celebrated as a big win for Nvidia and Grok, but there's still potentially a lot of people getting screwed over by this kind of deal. Now, another thing that I find really fascinating about this deal is how Jonathan Ross, you know, the founder of Grok, spoke about Nvidia prior to this whole acquisition thing.

Why in the world are we competing with Nvidia? We're not. Competition is a waste of money. Competition fundamentally means that you are taking something that someone else is doing and you're trying to copy it. You're wasting R and D dollars doing the exact same thing they've done. What you should be doing is you should be differentiating, you should be doing things that haven't yet been done.

So Ross always framed Grok as complimentary to Nvidia. I couldn't find any sound bites where they claimed they were competing with Nvidia. His argument was that Nvidia's GPUs are the best in the world for training AI models, while his LPUs over at Grok were the best in the world for actually getting those responses for running the inference.

Nvidia will sell every single GPU they make for training. Right now, about 40% of their market is inference. If we were to deploy a lot of much lower cost inference chips, what you would see is that same number of GPUs would be sold, but the demand for training would increase because the more inference you have, the more training you need and vice versa.

And then in another interview, when Ross was asked, why shouldn't people just use Nvidia for everything? Here's what he had to say.

Well, we are faster for inference, we're lower cost and we're lower energy. There could be something else that he's measuring for inference that we're not tracking that they're better at, but on those three, we're better.

All right, so let's put this all together. What does this all actually mean for Nvidia? It's both defense and offense. They're bringing in the architect of Google's TPU to help them compete. In a world where Google just proved you don't need Nvidia to build cutting edge AI, they're also getting technology that makes them stronger in inference, which, which is where the AI industry is rapidly heading as more companies deploy AI to actual users. For investors in Grok, particularly Black Rock, Samsung, Cisco and 1789 Capital, which Donald Trump Jr. Is one of the partners, they're seeing roughly a three times return in just 90 days. Grok was actually valued at $6.9 billion in September. Now they're getting $20 billion. Now for regulators, on the other hand, this deal represents a growing challenge among startups. These licensing structures are specifically designed to avoid antitrust review. If every major tech company starts using this playbook to absorb competitors without triggering oversight, what's the point of antitrust laws in the first place? And, well, for the broader AI industry, this raises a bigger question. Can any startup actually compete with Nvidia long term, or will they all eventually get absorbed? There's another AI chip company out there called Cerebras. They were planning to go public but recently decided to pull their IPO and they're not doing it anymore. So I don't know, maybe they're next. So whether you view this as smart business strategy or like a monopolistic consolidation, well, that probably depends on your own perspective. Nvidia would likely argue they're simply just paying top dollar for top talent as well as technology in a hyper competitive market. Critics, on the other hand, would argue they're buying silence, not innovation. They're eliminating potential rivals before they can become real threats. But what I think is completely undeniable is that Nvidia is playing a different game than everyone else. With over $60 billion in cash reserves, they can essentially write checks that make competition just disappear. And I guess the question I'm asking isn't whether or not this is all legal, because it is. The question is, should it be legal? But anyway, I wanted to break down that Nvidia Grok story for you because I've been doing some deep dive research and reading everything I can about it and I wanted to turn around and share what I learned about it with you. So hopefully you found that helpful and. But I'm not done with this video, this is my end of week roundup. So I have even more AI news to share with you. But the rest of it we're gonna do it in a rapid fire session. So let's get into it. Let's start with this new language model that came out from Liquid AI called LFM 2.2.6B exp or experimental. It's a 2.6 billion parameter model which in the world of AI is a very small model. It's basically designed to be run on device and can even run on mobile phones. Here's a quick screenshot of the benchmarks of how it compares to a bunch of other similar models that are all in that sort of 3 to maybe 4 billion parameter. I'm not sure what this Granite 4.0 is. I don't know if it's a 4 billion parameter model or if it was just the fourth iteration, but you can see that this model is the best performing model in pretty much every benchmark for a model of this size. Now I don't normally get too in the weeds with new large language models unless they're massive leaps, but I feel like Paul here really actually put this one into perspective well. So I wanted to share his X post on it where he said that's pretty crazy when you think about it. One of the biggest wow Moments with chat GPT was the update from GPT 3.5 to GPT 4. And now this open source model with 2.6 billion parameters can run locally on your phone and is better than GPT4. Like yeah, that's totally true. When GPT4 came out, it blew everybody's minds. It was like the biggest day in the AI world we've had in the last like five years. It was mind blowing. It was a huge event and now we're getting open source models with less than 3 billion parameters that outperform what GPT4 can do on pretty much every benchmark here. It's just like that is just a sign of how fast this is all moving. And since we're talking about language models, there was some news out of OpenAI this week. You can now adjust specific characteristics in ChatGPT like warmth, enthusiasm and emoji use. So if you go to your ChatGPT account, go into your settings, click on personalization, you have these various characteristics that you can change here under emoji they have more default and less. And I think I saw somebody on X bring up a good point. Why is there not an option for none? Because some of us never want emojis in our responses. But yeah, a nice little update from chatgpt that honestly I don't think most people are going to care about, but it's there. There's also been more rumblings around OpenAI and how they're going to roll out ads in their platform. They're obviously like bleeding money month over month and they need to figure something out to generate more revenue. And ads keep on coming back to the surface as one of the options just as recently as December 24th. Here we can see that it is back in the spotlight. Here's the quick takeaways. Employees discuss ways to display sponsored information in ChatGPT. Internal mockups show ads in ChatGPT sidebars, and ChatGPT shopping features could enhance commerce. Ads Again, a lot of this is just speculation, but people have been talking quite a bit recently about how OpenAI is likely going to roll ads into their platform. What I just really, really don't want to see, and I don't think the world wants to see it, is sort of sponsored. Responses I say I'm looking for a specific type of camera and I don't want it recommending the camera that's sponsoring them versus the best camera for my use case. If that starts happening. OpenAI's done right. Like people are just gonna go use Gemini or Claude or Perplexity or whatever. Model's not gonna do that to them. So yeah, something that they definitely have to tread lightly cause it could lead to an exodus if they do it wrong. ChatGPT also got like a Spotify wrapped kind of thing this year. Everybody's doing it this year. I got a wrapped for YouTube, for Spotify, for Steam, for what games I've played. And yeah, we got one for ChatGPT as well. Introducing Uber Eats Wrapped. Oh no. It's kind of fun to go back and see what chatgpt thinks of you. This is mine here. It gives you like a little poem that explains your year in poetry. Here's my big themes for 2025. Shifting from news to storytelling, building systems to buy back time, testing, tinkering and teaching the frontier. If you're really that curious about what I'm chatting with ChatGPT about, feel free to pause on any of these screens. Here's my chat stats. I was one of the first 0.1% of users, which is kind of cool, but definitely not in the top 1% of users because I'm jumping around between all the models all the time. My chat style, let's make this sound more natural based on everything you know about me. Help me flesh out this a bit more. Why does coffee make you poop so fast? And are mashed potatoes a good pre workout carb? Oh my gosh. I have kids. Okay, so kids ask questions like this. All right, just leave me alone. Don't judge. If 2025 gave out awards, you'd get most likely to Automate His Own Midlife Crisis. I mean, accurate. I'm the navigator. I don't feel unique though, because 23% of users share that there's what the Navigator consists of. Here's my year painted in pixels, still life with Steam deck and a neon coffee mug. And the funny thing is this is actually pretty accurate. I do have like multiple monitors, I've got cameras around me. I actually do have a coffee mug sitting next to me and my Steam deck is usually on my desk charging, but it's in my bedroom right now. Cause I was playing it yesterday and I guess this is like fortune cookie or something. I don't know. Anyway, go check yours out. I already immediately regret publicly sharing mine, but whatever, what's done is done. This is a pretty cool feature out of Manus. Slides created with Nano Banana Pro are now editable. And I mean it's like really truly editable. I tested this the other day. Check this out. I made a slide deck using nanobanana called the AI Power, Web Connections and Partnerships, blah blah blah. You can actually open it and see that it is an entire actually pretty decently looking slide deck that breaks down all of the sort of circular deals between all of these companies. But these were created with Nano Banana. Every single slide was created with Nano Banana. I could select any of this text and edit it however I want. Like literally any of this. Even the text over like inside of the image portions. I could retype Microsoft to Macro hard and apply the change and after like 20 seconds of loading it changed the text for me. I mean it also removed the Microsoft logo. But that might just be, you know, Google not wanting to misrepresent it or something, I don't know. But it's really cool that all of these Nano Banana texts are now fully editable. I could change any text I want now. One thing I did test was making an image over on Gemini specifically and then trying to import it into Manus. And it didn't edit properly. So it seems like you have to actually generate with Nano Banana inside of Manus. And then what you generate with Nano Banana inside of Manus becomes editable. But things that you bring in from outside aren't Editable, at least from what I can tell. Feel free to correct me if I'm wrong in the comments, but if I'm wrong, you were already planning on doing that. The company Alibaba over in China was going on a ship fest this week as well. They actually put out quite a few new things this week. They introduced Quinn Image Edit 2511. So it's a new image editing model. It's their version of Nano Banana or, you know, the new ChatGPT image editing tool. This is the Alibaba version which you can use over on Quinn Chat if you want. This one I found really interesting as well. They also released Quinn Image Layered, and this actually makes images that are layered, like Photoshop, so you can generate images that have multiple layers in them, like you can see in their little demo video here, and then pull them into Photoshop and edit the individual layers individually. That sounds like a pretty cool use case. Now, I imagine, because Alibaba figured it out, we're going to see that roll out into Google and OpenAI and all of these various platforms in the future. So don't be shocked if like every image generation platform can generate layers in the near future. But it's really cool that we've got kind of the first one I've seen so far from Alibaba. And they also released the Quinn 3 TTS lineup, which TTS is text to speech, so you type text and it gives you spoken word back, which is obviously getting more and more realistic by the day. So if you use something like 11 Labs, Quinn3tts is like Alibaba's version of what 11 Labs does. Quinn3tts family just got some new members. Quinn3tts adds voice design and voice clone and now supports 50 voices with enhanced ability on complex text. So I mean, that's a little demo of what it sounds like. I don't quite think it's up to par with 11 labs yet, but it's another option on the market that you can play with and I'm pretty sure you could test it out for free if you do want to play with it. This is actually news from a couple weeks ago, but I kind of missed it, so I wanted to share it now. Matt missed some AI news last week, so he's gonna talk about it right now. But YouTube rolled out a new feature called Playables Builder, which is a tool that allows you to create video games using AI, using Gemini, and then share those games so people can play them directly on YouTube. And YouTube reached out to me and asked if I can actually make a game. So I've had access to this for a few weeks and just like any other chat platform, you just prompt what game you want, it will build the game and then you go back and forth with it until you get what you want out of it. Now I'm not 100% sure if I'm allowed to show off the backend platform yet, but while I'm waiting to confirm that, just imagine it like a chat GPT kind of interface in the back end. You tell it what game you want, it has a whole bunch of pre written prompts to help you bug fix games and create graphics for games and things like that. So you tell it what game you want and then it suggests additional prompts to help optimize and improve the game. And you get games like this havesies game that I developed with my producer Dave, where you have to cut food in half perfectly 50, 50 to score and improve your significant other's mood by making sure you're giving nice 5050 cuts. And it's a surprisingly fun and addictive game and you get level ups every so often to improve your ability to cut food. Very, very basic, but also kind of addicting. And the whole thing was 100% built with AI. But if you head over to YouTube.com you can see the games that other people have built here. Other creators like Sambucha and my buddy Bilavel have also built games on this platform and they do have plans in the future for you to be able to monetize the game. So the games that you vibe code with AI, you'll be able to put ads in in the future. And just like YouTube ads, you'll be able to split the revenue with YouTube when people see the ads on the games that you developed. So a cool, fun sort of monetization tool for the future. Whether or not this is going to get overrun with just like AI slop games, that'll be yet to be seen. Hopefully the cream rises to the top and the people that actually put time and effort and thought into their games are the games that get the most reach. But again, this is all unfolding. It's very new. We'll see how it all plays out. Since we're talking about games, I wanted to mention this one. This game called Claire Obscure Expedition 33 swept the gaming awards this year. So just like the standard gaming awards, it won like every category including best game. There's also the indie game awards which happened recently and this game won that as well. They won best indie game. However, after they won, information came out that they used AI as part of the process to develop this game. And then the game was completely disqualified and they took away the award from them. So the game that was the runner up, which is a game called Blue Prince, actually ended up winning. But the interesting thing is there's lots of speculation that blueprints also used AI. So eventually I don't know if you're gonna be able to have a rule like this because like every game is going to have leveraged AI in some way shape or form to get to the game. Now you're probably thinking like, oh, if you vibe coded a game, let the people that actually like did the hard work to hand code games win these awards. But here's what actually happened. They used AI to generate one image that was like a poster on a signpost and they used it as a placeholder for this image and then later on hired a real artist to make that poster and they replaced the AI. So the AI didn't even ship in the game that people saw yet because they used AI to create this little, little teeny like 0.0001% of the game that they ended up removing before releasing the game. They completely disqualified them, which seems crappy to me, but you have to draw a hard line somewhere and they drew that hard line at if you used AI at all, you're disqualified. So I guess it is what it is. But I don't think rules like this are going to hold up forever because I just think more and more people are going to realize that AI is a great tool to help them get the game out quicker if you're a developer. So I don't know, we'll see how it all plays out. And then the final story, kind of a funny one. There was a power outage up in the Bay Area and when that power went out, all of the waymos just stopped causing traffic, making it hard for emergency vehicles to get around. And Elon likes to point out that all of the self driving Teslas and Robo Taxi type vehicles kept going while Waymo couldn't. So maybe AI is the future of driving, but we also need to make sure that these cars keep on moving even if there's a power outage. I don't know exactly what happened. I believe has to do with the data center that Waymo is communicating with. The data center's power went down so Waymo couldn't communicate with them, so they just stopped. That's my understanding of it, but I don't know all the details. I just think it's kind of crazy that all these waymos that are all over San Francisco, if you've ever been to San Francisco recently, they're everywhere and they were just frozen in place while this power outage happened. Which, yeah, kind of crazy story. But again, I didn't think there was going to be a whole lot to talk about this week. I was kind of right, kind of wrong. I mean, there was a handful of things to talk about, but that Nvidia story just blew my mind. Hopefully I did a decent job of breaking down everything you might want to know about it and you got a little bit of coverage of all the other little things that happened this Christmas week when people only worked like three days or two days this week. But again, that's what I got for you. This is probably going to be my last video of 2025. I will be back next week with my AI news breakdown of the week, but by that point it'll be 2026. Hopefully. I kept you informed and looped in on all the AI news and you know about all the coolest tools and you're using it to help automate more things in your life and business. That was my goal. Hopefully I did that for you in 2025 and I'm planning on doing it even more. Even better, even cooler, even smarter. I don't know, I'm planning on doing that more in 2026. So if you like stuff like this and you're not subscribed, maybe give this video a thumbs up and click that subscribe button and I'll make sure that you're always looped in with the latest AI news and tools and how to use it all and integrate it in your business and or life. If you like that kind of stuff, stick around and hopefully I'll see you in the next video. Thank you so much. I really, really appreciate you. I love you. Bye bye. Thank you so much for nerding out with me today. If you like videos like this, make sure to give it a thumbs up and subscribe to this channel. I'll make sure more videos like this show up in your YouTube feed. And if you haven't already, check out Future Tools IO where I share all the coolest AI tools and all the latest AI news. And there's an awesome free newsletter. Thanks again. Really appreciate you. See you in the next one.