In-Depth Analysis: Anthropic and Nvidia

0:00Insiders say that Anthropik is eyeing a$2 trillion valuation for their IPO, the biggest ever recorded. Nvidia is going to backstop$500 billion in AI data centers, which will be guaranteed by GPU resale values. Shield font is going to poison AI scrapers by swapping words inside of the text that humans can't see, but the AI models will. Lovable has raised$400 million at a$13.3 billion valuation. Their annual recurring revenue has tripled. Thrive Holdings has raised$2 billion at a$12 billion valuation. They're going to be essentially buying companies and injecting OpenAI inside of them. AI models are flagging fatty liver disease early as conditions are hitting about 30 % of adults worldwide.

0:50If you've ever been working with Claude and needed to generate images, audio, or video and been frustrated that it can't, I'd love for you to check out the AI Box MCP. Basically, it's a connector that connects over 80 different AI models, all the top AI models you're used to using, undoubtedly a bunch of them you already subscribe to and pulls them all inside of Claude. So when I'm using Claude and I'm trying to generate new images for Facebook ads, for example, I just say use the AI box MCP to generate an image of and I just put in the normal image prompt I would use anywhere. You can even create scheduled tasks inside of Claude that draw on the abilities of the AI box MCP and it can automatically pull an image, audio and video.

1:27It feels magical if I'm being honest, because in Claude, which I'm not used to seeing images generated, I'm seeing images that are pulled straight into the chat or the side panel, and you can visually see them all there. So if you want to give it a try, it is linked in the description, the AI box MCP, it is AI box.ai slash MCP. Before we get into a couple of the stories, I wanted to talk about a few other headlines today. There's just so much news, and I didn't want to build, you know, an hour long podcast episode. But I wanted to give you guys a couple updates. One is that Uber and Pony AI, they're going to deploy about 2000 robo taxis inside of Europe, which is really wild.

2:01Google is going to make the visible watermarks on their AI images, video and audio optional. Thank freaking goodness. I have been complaining. And in fact, I had Logan Kilpatrick on a podcast a few months ago. And I asked him about that. I'm like, man, when is Google going to get rid of if you've ever used like nano banana, there is a watermark on the images inside of nano banana. It's super annoying because you can't really use them for anything without trying to crop them out or using another AI tool to go and remove that like chat GPT. So it just was, it felt very not super usable without cropping stuff.

2:31So happy that that's going to be a possibility. Databricks has just raised$5 billion at$190 billion valuation. And finally, Cognition is in talks to raise$40 billion valuation just months after their$26 billion valuation round. Okay, let's talk about what's going on with Thrive Holdings, because this is a really interesting company. And I think the strategy is something we might see with more PE moving forward in the future. So Thrive Holdings is a company that buys traditional service businesses, and then it rebuilds them around AI. They just raised$2 billion at a$12 billion valuation. They raised the money from SoftBank and a bunch of other investors.

3:09But they now own over 70 different accounting and IT companies. And they're opening a third division, which is focused on infrastructure permitting and compliance work. Current Tax AI processed over 7 ,000 tax returns, and it did it with a 98 % accuracy, and it cut tax prep time by more than 30 % at the different firms that were using it. Shield AI's tools sped up IT help desk resolutions by 36x, and it doubled custom AI agent deployments in the last month. And OpenAI took an ownership stake in December of 2025, and they embed staff directly into portfolio companies to help accelerate the AI rollout.

3:49So I think this is interesting because they're, you know, they're working with IT companies, they're working with tax companies, I mean, something like a 98 % accuracy is obviously impressive. So you can see there's a lot of value there. But it's also interesting to me that OpenAI took a literal ownership stake in this and they're embedding their own staff to try to make this work. So it's beyond just OpenAI saying, hey, look, we're going to make AI that works great for AI or for tax companies. By taking an ownership stake, like they're becoming a tax company, which is really fascinating to me.

4:18Thrive in particular, this company is betting that the real bottleneck for AI and business isn't just the technology, it's actually getting it implemented. And we've seen a bunch of these AI implementation companies isn't the only one that opening eyes working with and Anthropics also working with some it's usually they'll kind of go with some big investment firms and create these but by owning entire service firms rather than just selling the software to them, they're capturing the full benefit of the productivity gains that AI can roll out. And I mean, this isn't anything super revolutionary.

4:48If you think about it, private equity firms oftentimes will go buy a company or a bunch of companies in the same space, and then they'll consolidate some of the back office, they'll be able to cut a bunch of staff because of this, and they'll make the companies more profitable because of it. So this is just a new element to it where we're using AI to do that. But it's really the same playbook that they've been doing for a very long time. Lovable, which is based in Sweden, is a platform. If you haven't heard it already, it turns... I've talked about it a ton on the show, so I would assume you have.

5:19It's not a company I'm actively using right now. But basically, if you want to build a web app or some sort of website or tool, it's a really easy way to get started with that. Today, I would just say go use Claude Cowork or ChatGPT Work because they can do a lot of the same stuff. but Lovable just raised$400 million at a$13.3 billion valuation. Their ARR has about tripled. So their user base is massive. Projects built on Lovable right now jumped from about$25 million to$60 million in eight months and monthly visits to those apps quadrupled to about$900 million. So not just visits exclusively to Lovable, but the apps that are built on Lovable are generating about 900 million visits, which is absolutely wild.

6:05So their annual revenue, like I mentioned, tripled, and that's just over eight months. So I mean, this is a company that is growing incredibly fast. Lovable didn't say exactly what their annual revenue number was, but they're like, look, it tripled. So I'm not going to put as much stock in that until I know the actual number. Menlo Ventures led this particular round. They also led the previous one. So I think that they're very bullish on the company on the metrics, and they think that it's got some strong growth numbers, even if they're not telling them all to us. Lovable in particular now handles security checks, they prompt users towards making safer decisions.

6:38So if you're non technical, which is basically everyone using lovable, you can ship stuff really fast. And you're not really worried about the security angle. And when I was using lovable, because honestly, lovable to me was a gateway drug into vibe coding, I'm not I wouldn't have considered myself a technical founder. And lovable is how I got started building apps by myself and kind of coding stuff. It does like your backend. It does like your login with Google. It does, you know, links to your Stripe payment. So it does a lot of this stuff for you and it makes it pretty simple. I'll say if you're willing to invest like one or two days, you can get off of Lovable completely and just build it from scratch with something like Claude Cowork or ChatGPT work.

7:18And using that, all you need to do is set up a Vercel account, a GitHub account, and you need to make a Supabase account. So if you make accounts on those three websites, you can get ChatGPT work or Claude Cowork to go get an API or an MCP, link them into their tools, and they can just build everything and work on everything for you. So if you want to spend a couple days, that's how I would do it. Otherwise, you can go with Lovable. My only downside with Lovable, if I'm being honest, which obviously they're making an insane amount of money, is it was pretty expensive. And so I was just doing so much.

7:54and what you gotta appreciate is that Claude and ChatGPT both have like$20,$100, and$200 tiers and all of those tiers, the tokens are subsidized, especially if you're doing like a$200 a month tier. You can get a ton of building, a ton of coding done and you get like thousands of dollars worth of coding for$200. If you're using Lovable, it's all API based and so you're gonna have to pay the thousands of dollars, which I was when I was using Lovable and I was, you know, once I think I spent $2 ,000 in a month on credits. I was like, okay, I need to figure out a cheaper way to do this. And that's when I moved to, to Claude and, and, but you can build a really impressive and really cool stuff there.

8:33NVIDIA is backing a$500 billion data center financing deal with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. And they are guaranteed to cover 25 % of any losses. If the GPUs use, um, and they're using the GPUs as collateral for any sort of loss in value. So that basically is letting those six different companies fund massive AI infrastructure build outs. And Nvidia is going to gain a channel to sell billions in GPU sales. They're not going to use their own cash for it. They're just kind of like a backstop on it. But they're going to get they're going to make a ton of money from this.

9:10This makes this is really interesting because Nvidia has committed about 750 billion dollars in these type of deals just this summer. They have deals with OpenAI, Anthropic, CoreWeave, and a bunch of other AI companies. Bloomberg was kind of reporting on this earlier, but the guarantee basically creates a quote-unquote wrong way risk. So NVIDIA's obligation to pay grows biggest exactly when GPU demand might get weak and NVIDIA's own revenue would slow. So that's kind of a dangerous moment, right? They're like, hey, look, like, if for some reason everything crashes, like, we'll backstop you, but like, if everything's crashing and no one's buying the GPUs, NVIDIA, like their ability to backstop them is going to be definitely strained.

9:54So it's a tricky moment. The CEO of Microsoft, Satya Nadella, recently recommended a book about railroad era financial engineering that ended in collapse. And so, you know, potentially there's other people that are kind of looking at this and saying, hey, maybe this is kind of like what happened to the railroad as far as infrastructure spending goes and saying that there are some dangerous things happening. But in any case, if NVIDIA's residual value guarantee works, it could create a kind of a really liquid secondhand market for older GPU chips. And that would let startups and smaller companies buy some of the older generations of hardware, it would be a lot cheaper instead of just always having to buy the newest stuff.

10:33But that also assumes that GPU demand is going to be really strong, right? And so I think Nvidia is trying to like, ensure it's strong by backstopping these big firms to go build out all the new stuff and then the old stuff is still going to be valuable to other people. But, you know, NVIDIA's liability is going to be massive in the case of anything going wrong. Okay, this is an interesting story. It's called Shield Font and basically what it does is it poisons AI scrapers by swapping out different words. So in the HTML, when a scraper is going and scraping your website or your blog, it's going to show random words mixed in there.

11:10To a human, you won't be able to see it but in the html it's going to mess it all up and so human readers go to your site they get everything that they want but if an ai model is trying to scrape it it's going to get really jumbled up text and it's going to mess up their llms for you know producing high quality content so they actually tested it against six major ai scraping pipelines and over 90 percent of pages that were using shield font were rejected by the scraper's quality filters so basically it's going to be, it's a technical defense against AI models coming and trying to train on your data without being authorized to do it.

11:46I love this concept if I'm being honest. And of course, I'm very bullish on AI. And I think like, you know, accelerate AI and I'm like very pro that. But at the same time, I think these AI models need to compensate people and people take measures to ensure that that happens. I'm thrilled by it. One thing that I've done is with even like this podcast, I've licensed my podcast audio to different AI models that train. They train models on, you know, just how people talk. So it's general purpose, not like cloning my voice or anything. And it's, it's been an awesome revenue source for me. It's kind of random, but it actually makes a ton of money.

12:19And I've seen a lot of other people doing this with audio, with video, with music. And by the way, if you're interested in doing that, and you have some sort of data set, go to my website, F I U N D.com. I'll leave a link in the description. It's find.com where I will give you, I'll let you know what your podcast or your YouTube channel or whatever your data set might be worth. If you're interested in that, but yeah, so I love this and I think this is a great way to help people get paid and kind of fight back against scraping that isn't authorized per se. I mean, if people want to opt into it, fantastic.

12:54If you want to opt out of it, I think there should be a way to opt out of it. And sometimes even when you have like a do not scrape, but they still go scrape you. So I think this is the next best thing and they get what they pay for. AI models that are trained on routine blood tests and chest x-rays can now spot fatty liver disease early. So this is a condition that affects about 30 % of adults worldwide. And if you can catch it early, it helps because liver damage is reversible if you have like treatment or some lifestyle changes. But yeah, three quarters of patients only get diagnosed when it is already life threatening.

13:29So there was an OSCA Metropolitan University model that identified fatty liver diseases from chest x rays, it had an 82 % accuracy. So it's able to spot spotting the liver fat in really small slices of organs that are visible on like routine scans. And if vetoes liver pro algorithm, which was built from age and nine blood biomarkers, outperformed the standard fib four test across 470 ,000 middle aged patients, and it's now being commercialized by Roche. So I love that because we kind of had the standard test, and they came up with the new algorithm, and it and it beats it. And so that's what's actually going to be commercialized and rolled out for everyone.

14:12The liver actually can regenerate, and it can also reverse fibrosis if you catch it early. So if you know, this kind of treatment is lifestyle stuff, I think we know the things that aren't good for our liver, but your doctor can definitely tell you more about what what's entitled there. But yeah, I think this is phenomenal. And this is something that is really exciting to me. I love it anytime that AI can help with healthcare can help cure diseases can help solve diseases. And so this is something that I'm really excited about. Guys, thanks so much for tuning into the podcast. If you enjoyed it, and you want to get access to over 80 different AI models inside of Claude or anywhere else, go check out AIbox.ai.

14:48And I hope to see you all on the next episode.