Latest AI News: Anthropic and Nvidia

0:00Anthropic is filing confidentially for an IPO at a 965 billion dollar valuation which is the same of their last round of funding. So what is the top of those investors get? That's a great question. Also, Microsoft is unveiling something called MAI Thinking 1. This is the very first reasoning model Microsoft has unveiled without having to rely on OpenAI or Anthropic or another player. NVIDIA has just unveiled what they're calling Cosmos 3. This is their very own model for physical AI reasoning. And Intel's Crescent Island AI chip is going to ship this year. It is undercutting NVIDIA on cost and cooling.

0:35So is Intel going to make a huge comeback and is NVIDIA going to suffer? Strava has just put their API behind a$12 a month paywall. They are blaming zero code AI apps for this. And we have some weather AI news. Windborn's Weather Mesh 6 is out forecasting the ECMWF. They have 400 balloons that are feeding this model and it has some pretty incredible results. Anthropic has just filed confidentially for an IPO on June 1st. That is today. a few hours ago before I recorded this. Also remember, this is just a few days after they closed their$65 billion Series H. And that Series H, if you do remember, valued the company at$965 billion.

1:16So this makes them one of the most valuable private companies in the entire world. Now, what's interesting is that they have an absolute battle royale going on between OpenAI and SpaceX because OpenAI is valued at$852 billion according to their last round of funding. So now they have surpassed that. They're over$100 billion more valuable than OpenAI. But SpaceX is looking for$2 trillion in valuation, apparently. So Anthropics revenue is, I mean, if I'm being honest, it's really exploding. I think they hit a$47 billion run rate. That's up from$9 billion at the end of last year, 2025. So, you know, we're not even halfway through the year yet.

1:55And they're already going from$9 billion to$47 billion. That is a massive jump. This filing that they just have done essentially allows them to refine all of their risk factors and all of their financials privately before they publicly disclose this to everyone. This is a very standard move for these mega cap IPO candidates. The company's mythos model, which they previewed in April of this year, is still under restricted access. There's thousands of high severity software bugs and the EU cybersecurity agency access is planned for this as well. If you use AI or are interested in using AI to grow and scale your business, something that I do every single week, I have a school community.

2:34It's called AI Hustle. And every week, my friend Jamie and I, we break down all of the AI tools that we're using inside of our businesses, all of the strategies we're employing, how much revenue we make from different objectives. And we post it all there. And it's exclusively over on our school community. It's$19 a month. And we basically help people grow and scale their business and grow and scale their career using AI tools. This week, we recorded an entire episode where we break down Jamie, my friend, how he got access to chat GPT ads, what it looks like inside of the platform. It's like a YouTube video kind of tutorial so you can see everything going on, what his return on his ad spend was, how much it costs per click, what he's advertising, how the targeting works, which is really fascinating.

3:16So if you're interested in how the new ads inside of ChatGPT work, how much they cost, and you want to see an inside scoop on that, go check out the link in the description to the AI Hustle School community. It's$19 a month to get access to it. And if we raise the price in the future, it will never be raised on you. So I would love for you to join the community. We have over 300 members. It is a ton of fun. People share what they're working on. We're a great community. So if you're growing and scaling your business, if you're vibe coding tools, we'd love to have you in the community. Again, there's a link in the description.

3:44So SpaceX is also, like I mentioned, planning to file their own IPO on June 12. Some people are saying this is to steal the thunder from Anthropic. I know they've been working towards this and maybe they're worried that Anthropic going first would steal their thunder. But for the June 12 IPO, SpaceX is looking to get$2 trillion. They're trying to raise$75 billion or more. But what's interesting is we also have Anthropic paying them over$15 billion annually for data center capacity. So this is a very interesting competitor customer kind of arrangement that we have going on. Anthropic has, you know, blasted past OpenAI on paper so far, but maybe OpenAI will be able to raise more money or increase their valuation by the time that they're actually trying to come to their IPO.

4:27Microsoft has just unveiled MAI Thinking 1. This is the very first reasoning model, and it's also kind of the co-pilot super app that they're building right now. it's the first time that they built something without relying on another company's AI output. So they had a big announcement at Build, which is in San Francisco, the Build conference. And right now, I think they're just basically trying to show everyone that, look, we are not going to be dependent on OpenAI for the enterprise. We can build our own fully proprietary reasoning systems. They say that they did not use model distillation for this.

5:01So I think even XAI admitted that they had used some of OpenAI's outputs to train their model with model distillation. This was kind of a trend that I think a lot of Chinese firms showed the strategy for if you use model distillation, you could get your AI models to learn much quicker and cheaper. Microsoft said that they did not do that. They built this from scratch. They weren't copying anyone else's outputs. They are also launching a Windows 11 developer mode with pre-installed tools and a distraction free environment plus a bunch of image models that the MAI Image 2.5 and MAI Image 2.5 Flash.

5:35So a bunch of exciting AI stuff coming out of Microsoft. I hope to see them as a bigger individual player. It would be great to have another competitor in the space. NVIDIA has just unveiled Cosmos 3. This is a brand new model. It's a foundational model, and it basically lets robots reason about their actions before they execute them in any predictable real world environments. This right now, I think, is putting NVIDIA against a bunch of other competitors that are doing like similar kinds of things on exactly Google DeepMind, physical intelligence, all of these are in a race to basically build this general purpose embodied AI, really something that can move robotics beyond just demos and actually doing reliable autonomous operations.

6:15I think everybody knows that this, you know, AI, chat GPT, whatever, this is all incredible technology. But the next step is robotics. So Cosmos three works by basically having robots mentally simulate outcomes against a learned world model. And then they select the highest probable action. They can then predict and act. They have that kind of capability. And this, I think, is what separates them from a lot of the household robots and kind of warehouse robots that you see around. This is really interesting, right? Like these robots are sitting there. They have these real world models in their brain.

6:47And they're like, if I take one step, this is the likely predicted outcome. If I take one step to the left, this is the likely predicted outcome or seven likely predicted outcomes. So it can choose an objective and then it can, you know, predict all the outcomes of all the ways to get to that objective and then pick the best one. This is really crazy. I'm sure in some way our brains are also kind of doing that, but we just do it so instantaneously we don't think about it. So it's really cool that we're starting to see some of that tech now move into robots. The model was presented at the International Conference on Robotics and Automation.

7:17This is basically robotics biggest, largest kind of annual gathering. They're bundling Cosmos 3 alongside their Isaac and Omniverse tooling. These are kind of synthetic data pipelines. You know, this is going to make, they're kind of focusing on chip sales and their integrated stack. There's a bunch of competitors like Physical Intelligence and Skid AI that don't have all of that. So this is a big advantage that NVIDIA has. The next thing I'm excited about is Intel because they are going to, I think Intel, it's a great American company. I have friends that have worked there for years. For the last 20 years, it's been interesting hearing stories from them on how the company has evolved.

7:56But one thing I will say, if you look at the stock price, if you look at kind of where Intel was 10 years ago till today, it feels like Intel has had a bit of a bad downward trajectory. So I'm really excited when I see Intel on the way up doing some exciting stuff. They have announced that they're going to ship their Crescent Island AI chip. There's going to be limited qualities of it, but they're going to be shipping it by the end of the year. And the big thing that they're selling is that it's undercutting NVIDIA and AMD by using cheaper LPDDR5 memory and air cooling. So they're doing that instead of some of the more expensive HBM and kind of liquid cooling systems.

8:31This right now, I mean, they're directly targeting inference workloads. So wherever like memory bandwidth demands, all of that are looser. It's going to let Intel chase more of a price sensitive market segment that NVIDIA's dominant, you know, kind of training focus chips aren't going after because NVIDIA, let's face it, it's very, very expensive. So what's interesting here is the Crescent Island. It took about 18 months for them to develop it and it's going to be manufacturing Intel's own fabs. I think a lot of people are excited about the new direction of Intel. If you look at just their share price, they've surged more than 200 % this year alone.

9:06They have a new CEO, So Lip Boutan, who replaced Pat Gesslinger last year and is aggressively rebuilding their AI Silicon roadmap. So I think that's a massive reason why the chips might actually qualify to be sold in China under the US export controls. That is a really big revenue door. It's largely closed to NVIDIA and AMD because of a lot of the trade tension with Beijing. So this could be another reason why Intel could be making a lot more money. All right, we got to talk about Strava because they have just put their API behind a$11.99 paywall. They're blaming zero code AI apps for this. This just started in June.

9:44They're ending their free tier that they used to have. Basically, they're blaming a 448 % spike in developer applications year to date. They said this is all driven by zero code AI app builders that strain their infrastructure without contributing to its cost. That is crazy. When you think of all the software and all the tools out there, all of these APIs that people are able to use. If you're giving away a free API to your data or to your tooling and you're not getting any revenue out of it, it's really hard to maintain. Those can be incredibly expensive. So honestly, this is not a shocker to me.

10:17I know for my own startup, AI Box, we have an API. So people can, we have access to over 80 different AI models. You can take our API, plug that into other tools to access what we have. But you have to have an AI Box account And you have to have a subscription because when you're using the API, you're using tokens to use AI tools. So we make money when people use our API. The companies that they're, you know, the end AI platforms and tools that they're using also make money. And that kind of makes a lot of sense to me. But if Strava was giving this away for free at zero cost, yes, I can imagine that would be very abused.

10:53And I'm not shocked. $12 a month isn't that crazy or$12 isn't that crazy. so the wearable and device integrations from garmin apple watch and wahoo are still going to be free only third-party app builders are going to have to pay that monthly fee strava also simultaneously launched a direct integration with anthropics clod so basically you can um you can create pipelines and you can see all of your workout data your heart rate your gps all that get plugged into your ai assistant you don't have to pay for that so individual users if you're just like, hey, Claude, like how fast was I running? What are you know, what's my like, what was my heart rate this morning, that kind of stuff, it will still do for free.

11:30But if you're trying to build an actual app and have that data integrated, you got to pay for it. I think it's very similar to what's going on with Reddit. In 2023, they made an API paywall, which ended their third party clients, a lot of platforms are saying, look, there's so many AI scrapers out there, they're all extracting a lot of value, a lot of data from our platforms, they're not paying us anything. This makes a lot of sense. I think Twitter did the same thing when Elon bought it. So I'm not shocked that Strava is doing this, if I'm being honest. Okay, we have to talk about AI and the weather because there was a really interesting report that said Windborne Systems, they released something called Weather Mesh 6.

12:04This is an AI weather model, and it claims to beat the European Center for Medium Range Weather Forecasting. This is like the gold standard, right? It's basically the best in the world. So it's better at surface temperatures and a bunch of other variables. But this new Stanford startup said that they are able to beat it with their AI model and their kind of secret sauce, their secret edge is that they have 400 balloons that are feeding raw sensor data directly to their transformer model. It's then bypassing the government data pipelines and all other AI forecasters have to use. So I mean, they just basically put up their own 400 balloons, and they don't have to use any of that government data stuff, which is crazy.

12:43They have six updates every hour at three kilometer resolution across Europe and continental United States. That is compared to the six hour updates from traditional kind of physics based systems. Windborn's balloons, they operate from 15 global launch sites and they supply proprietary atmospheric data that major labs. So we're talking Google DeepMind and a bunch of other people, they cannot get access to this. Only they can. So they have this kind of proprietary data because again, they're putting up the physical balloons. They actually raised$25 million in an$85 million valuation back in 2024 and their customers are NOAA, NOAA, the U.S.

13:24Air Force, U.S. Navy, and a bunch of commodity traders which I think is a hilarious customer, right? They're trying to get like forecasts to see like is the weather going to destroy the wheat yields, the wheat crop yields, and then we'll know to like short wheat or whatever. So that's hilarious to me.