Loading summary
Transcript1 lines
- [00:00]
A
Anthropic is localizing Claude's pricing in India, which is its second largest market. Interesting because people there will be able to get it cheaper than people in America or other countries that make more money historically. Apple's OpenAI complaint. They're detailing a show and tell, interviewing over 400 hires, claiming that OpenAI is copying trade secrets. World models right now are getting billions and Lacan Li and Runway are all betting against LLMs. We're going to get into the disparity there. Meta is doubling their Hyperion Data center to 5 gigawatts. It was going to be 2.5, or I thought it was maybe 2 at the beginning, but they're pushing the spending past $50 billion, increasing it there. Sam Altman and Elon Musk are having a ton of beef, which is kind of funny. We'll end the show on that. Sam Altman is saying that Elon Musk's space data centers are a short term sales pitch and Elon Musk is calling him Scam Allman. So there's lots to get into. Ever since Fable 5 came back to Claude, Anthropic has been playing this weird game where they keep saying they have a new deadline for when Claude Fable 5 is going to get pulled off of the subscription tiers and you'll only be able to access it through the API. Initially they extended it till the 12th and or I think the 7th was like the first deadline. They're like, it's going to be off after the seventh. And I remember using up all my tokens till the 7th and then they said, look, we're resetting it and you actually can use it till the 4th or the 12th, which was yeah, yesterday. And so I used up all my tokens again on the 12th and my tokens don't reset on Claude till like Wednesday. So now I'm, I literally have no more Claude tokens. Can't Claude for like three days on two separate Claude Max accounts, mind you. And then what happened? 20 hours ago, Claude came out with a tweet and said we're extending Claude Fable 5 access to all played paid plans as well as keeping Claude codes weekly limit rate limits to 50% higher through July 19th. This is the second, third time, second time that they've extended this. It's super annoying. I replied to them on Twitter and said, except I already used up all my tokens in anticipation of the deadline, so I have no more credits left. Thanks for nothing. The sentiment seems to be felt by everyone. If you look at Omasar over on X, he said. I don't think Anthropic realizes how disruptive these changes are to users. I appreciate the extension, but please stop playing games. Either keep it under the subscription or put it under the API already. I 100% agree. This is super annoying because you obviously when someone makes a rule or says that there's a deadline, you're going to plan accordingly, AKA use up your tokens ahead of time. Would I have been more frugal? Would I have spent them better, you know, if this if they were been there for two weeks? Yes, obviously. So now I'm in a pickle. And lucky for me and for anyone else that might have had an occasion where you ran out, you hit some sort of subscription limit token limit. I would love for you to check out my platform AI box AI people because when you reach a token limit on OpenAI or Claude or any other platform AI box AI for 8.99amonth, $8.99, you get access to over 80 different AI models and it's going to hold you over till your, you know your your next hit of tokens comes back in from Claude. In addition, to be honest, you get access to image, audio, video, music, text AI. So if you just want to have access to every single AI model anytime that you need it, go check out AI box AI. I'll leave a link in the description. In other news for Anthropic, they are now selling Claude in Indian rupees and they're pricing Claude Pro at 2,000 rupees. That's about $21 a month. And their most powerful model, Claude Max at $125 a month American or about 12,000 rupees. India accounts for about 5.8% of Claude's global usage. So it makes them basically the second most popular market after the United States. But they can't accept upi, which is India's dominant payment method. So this makes it tricky. But what's really interesting is a lot of people thought and you know, I kind of mentioned in the intro to the podcast that when you localize you typically will make it cheaper in the country by based off of what the income is of that country. So if India makes less money, you would make the product cheaper in India. This is what a lot of SaaS subscriptions do and in this case they did not do that. Anthropic is not heavily discounting the model for people in India and in fact it's actually more expensive because in the United States the PRO plan of Claude is $20 a month and it's roughly. The conversion rate bounces around, but it's roughly $21 if you're doing the conversion from Indian rupees right now. OpenAI added UPI support when they launched rupee pricing in August of last year. And Anthropic is still requiring credit cards or Apple or Google billing. It's a really big friction point in a market where UPI is basically the default checkout method in India. Anthropic opened a office in in February of 2026 in India and they signed some partnerships with Infosys and Tata Consulting Services to try to help reach Indian enterprises. But they're being criticized because they currently cost more than if you had tried to get the same service in the United States. Apple is filing a 41 page lawsuit alleging that OpenAI has systematically recruited over 400 Apple employees and extracted confidential hardware secrets from them, including by coaching candidates to bring actual device parts to job interviews and helping one engineer allegedly breach Apple's cloud storage after he left. The case that they're making right now is targeting OpenAI, their hardware subsidiary IO, which Apple acquired for $6.5 billion. And three current OpenAI executives. And if you guys remember, they have this huge like deal going on right now with Jony I've who formerly helped design the iPhone over at Apple and OpenAI is working on a device with him. So Apple is not happy about this. Tang Tan, who's OpenAI's chief hardware officer and a 24 year Apple veteran, allegedly instructed JOBC candidates to bring confidential parts like batteries and logic boards to show and tell interview sessions. Former Apple engineer Chang Liu downloaded dozens of confidential files weeks after leaving in January of 2026 by exploiting an authentication vulnerability and then texting a colleague. I still have another computer. Apple said that IT first contacted OpenAI about the alleged misconduct in February and received no response before they filed this lawsuit. OpenAI's hardware device is expected to launch next year. You know, Apple's obviously getting a little concerned in the run up to this. I think this is one of the most detailed evidences yet of what maybe you'd call corporate espionage. And for the AI industry what it looks like. This isn't, you know, a spy movie or something, but you're basically have a company, OpenAI, that's systematically hiring specialized engineers from Apple. Now this shouldn't be illegal. You should be able to hire people from Apple and even target Apple or another competitor if they're building something that you want and their employees have the domain expertise to do it. I Mean, there's nothing wrong with that. But if Apple's allegations hold up, I think it's going to set a precedent for how a lot of companies protect knowledge from rivals when it comes to hardware. And, you know, obviously there's some, there's some red lines to not cross, like telling people to bring actual components from the company or bringing their computer with data on it from the company. So that's all going to get litigated in court. Three different AI startups, World Labs, Advanced Machine Intelligence and a Runway, have all raised a combined $2.3 billion in early or earlier this year. And the goal of these companies, both, all three of them, is to build world models. So these are basically AI systems that are going to simulate physical environments instead of just generating text, right? LLM, ChatGPT, generate some text for you. These are going to simulate real world environments. And there's a lot of interesting use cases. It's interesting that Runway's working on this because they need video, and video you need kind of a simulated 3D environment if you want it to be really powerful so you can tell the camera to move around from all different ang. But more importantly, I think is the robotics era where this kind of falls into and, you know, you need to be able to simulate these physical environments to make these simulated environments for robots to train in. Right now, the bet that all three of those companies are making, I think, is showing a huge split in the AI field over whether LLMs alone can lead smarter, more capable. AI World Labs and AMI each raised about a billion dollars. Runway added 315 million in February this year. Google, DeepMind, World Labs and Runway have all released working versions. So there's Genie 3, there's Marble, there's GWM1, and they're all using different technical approaches. Yan Lecun, who led AI research at Meta for years, he left and he founded ami. And he is basically betting that the company is going to be able to build something better than LLMs as the path forward. And the LLMs are not the way to go. The reason why I bring this story up at all is because I think it matters a lot when it comes to robotics, video creation and scientific simulation. So we're not just talking about chatbots, but what is the next step of AI? Like, what are the next industries that this is going to go into? Okay, so Meta is doubling their Hyperion Data center in Louisiana. Louisiana to 5 gigawatts and about $50 billion. That's about double their $27 billion they announced last October if you remember, this expansion that they're doing now, I think is showing that Meta is making a huge bet that they are going to need some enormous amounts of AI computing power to stay competitive with Microsoft, Google and Amazon. And to be fair, like I would have said, I don't know, like I would have discounted Zuckerberg. But I mean, last week they came out with an absolutely incredible model. It feels like they're trying to get it, they're kind of getting a little bit of their mojo back. The project that they started back in December of 2024, they had I think $10 billion that they were going to put towards a 2 gigawatt facility. And now Meta is looking to reach 2 gigawatts by 2030 and they have not set a timeline for the full completion on that. Now I think there's a lot of reasons why they might be expanding this project. In particular, we've seen, we've all seen the headlines of people kind of opposing data centers or, or people trying to slow them down. And there's been like a lot of, I think there's like a statistic that something like, you know, six out of 10 proposed data centers gets shut down due to like community outreach, that they don't want these data centers. And so I think if you have the approval for a data center and you have the space and the land and the government connections and approval, I think we're going to start seeing some of these companies just decide to build one extra large one wherever they're at. Elon Musk is doing this with SpaceX, building the, you know, his Giga data center things, I can't remember what it's called is Colossus 3 or whatever. Meta is now apparently doing the same thing and I'd be curious to see if we see others making the same play where if you have the land next to you might as well expand the one that you currently have rather than try to get a whole bunch approved all over the place. Louisiana granted Meta a 20 year sales tax exemption on equipment purchased for data centers built before 2029, which I think lowers the cost a ton, maybe tens of billions of dollars in hardware. And local businesses have already earned $1.6 billion in contracts since the construction began. So overall for the community and the economy there, it's, you know, making a lot of money. And Met is getting this big kickback and sales tax. So they're happy to expand. I think they know that they're not going to get it cheaper anywhere else. Meta committed over a billion more for roads, water systems and other local infrastructure with this project. So hopefully this becomes a real win win for just not just meta but also the community. Okay, we got to get we got to end the podcast today on some beef in the tech world, which is Sam Altman and Elon Musk. Musk. Elon Musk was tweeting and he said that Sam Altman takes scamming to a whole new level when talking about Apple lawsuit that's going on right now. Sam Altman replied by saying, homeboy, you're the one selling public market investors on short term space data centers. Elon Musk replied and said, we start flying them next year. Maybe you can come see them if your parole officer approves. After stealing an open source AI charity, you then stole all of Apple's phone technology. Wow. What do you plan on? Plan for an encore that's tough to beat. It's always fun when the billionaires are feuding over on X. But what are the facts now? There was an article published on TechCrunch that I thought was interesting. Breaking down, I guess the case against Elon Musk and TechCrunch is I find it just technically it usually has that sort of coverage. But I do have some interesting points on all of that. So the first thing that TechCrunch says is that SpaceX told IPO investors that the starship was not going to be fully reusable near term and it's going to need to expend second stages on every launch, according to TechCrunch. They said this is going to erase the cost math for orbital data centers. They say true operational reusability requires years of iteration beyond successful test flights. Starship's near term manifest is spoken for by NASA and Starlink before Orbital Compute gets a slot. So they're saying, look like they're not even going to have the bandwidth to get this compute up. It's going to need tons of iterations. It's probably not going to be very feasible. And they say that they said scale Orbital Compute needs multi megawatt satellites, vacuum thermal management, ground station bandwidth and launch costs low enough to be Earthbound data centers, a combination experts say arrives in 2000-30s. So TechCrunch agreeing with Sam Altman and saying, look, there's no way that SpaceX is going to get the cost efficiency or have the bandwidth to put these space data centers up there. But a huge part of their $2 trillion valuation was relying heavily on Orbital, you know, this Orbital Compute thing. So, you know, it's just probably not very feasible. Now on the one hand, I think criticism of these two companies is fantastic and important, but on the other hand, for some reason, all of this just reminds me of a post I saw online. Well, there's actually been a handful of them. One post I saw online in regards to SpaceX was, you've probably all seen the post where it's like, it's like, continue iterating. It's like always a caption on this post, but it's the SpaceX rocket boosters when they first started and they were like these really complex rocket boosters with wires all over the place. They look like Frankenstein. And then it shows whatever the rocket boosters are today, which is like really sleek. It looks like they've gotten all the wires and a lot of the internal stuff moved together. I remember when this, this post first went out and they posted a picture of it. There was a guy that like commented on it and he was like, I work in aerospace and this is literally physically impossible. This is an AI generated image. You couldn't, you know, the, the whatever, the wire that goes here, XYZ and goes to this place. Like you've literally removed this part, this component of the rocket booster. That's impossible. This isn't real. And replying to that guy was someone that said, I'm an engineer at SpaceX and we worked on simplifying this. We were able to move that thing over to this other component so you can't see it. This is actually the real rocket booster. And the guy said, he's like, dang, you know, hats off to you. And that was kind of the thing. So all I'm going to say is a lot of people find it very easy to keyboard Warri criticize SpaceX in particular because it's rocket science. And you know, if someone sort of knows what they're talking about, they can say, it sounds super complicated. When SpaceX originally started, they said it would never be viable because the, to hit their financial goals and their, their launch goals because they would have to accelerate launches by a certain degree that was just too high. I remember seeing an interview by someone in the space industry just saying that there is like, you know, Elon's mask math was not possible math and that it was never actually going to happen. So all this to say, when it comes to SpaceX, they have proved time and time again to, you know, prove all the haters wrong. So maybe everyone over at TechCrunch does truly know more than what SpaceX says they're planning on doing. But I would say give them a little bit of a benefit of the doubt because they've proved the doubters wrong a couple times. But who knows? Only time will tell. If you enjoyed this episode, make sure to go check out AI box AI if you want to get access to over 80 different AI models. So the next time Claude runs out of tokens for you, you have a to go. Or if you want to just be able to access all of the different AI models, you can even pay for a higher subscription tier. We have 20 and 40 and I think $80 subscription tiers, so you can get tons of tokens, tons of usage to access all of the different audio, image and video models in one place. Go check it out. It is AI box AI.