Dissecting Anthropic's $965 Billion Valuation Growth  🔗 **Source:** [Dissecting Anthropic's $965 Billion Valuation Growth](https://share.snipd.com/episode/a392aa46-2722-4381-9724-b9fe5e8ab0b3) ## Highlights - **Anthropic's Leap Past OpenAI** - Anthropic raised $65 billion at a $965 billion valuation, overtaking OpenAI's prior $950+ billion mark. - The raise follows a reported $47 billion revenue run rate and 4.7x growth in 12 months driven by enterprise adoption of Claude features like Claude Code and Cowork. Transcript: Jaeden Schafer Anthropic has raised $65 billion at a $965 billion valuation. That is surpassing OpenAI's $950 billion valuation from their last raise. So Anthropic, not only is their usage skyrocketing, but they have officially passed the valuation of OpenAI, which is insane considering how many users have OpenAI. ChatGPT has become synonymous with AI chatbots, but Anthropic is officially pulling ahead because the enterprise is really paying off. Shift, a new company, is offering free home cleanings, so they'll actually send someone to your house, they'll clean your house for free in exchange for robot training footage. So the person that comes to your house to clean it is going to wear a little camera on their hat that points down and watches everything they do while they clean, and this is to train robots To clean. Boston Children's is using OpenAI and all of their tools to diagnose 40 plus rare disease cases. Grok is looking for 650 million dollars from existing investors after NVIDIA's 20 billion dollar licensing deal. ([Time 0:00:03](https://share.snipd.com/snip/1ecd16df-f514-4933-acdb-2b3cd375b673)) - **Host's AI Site Hit 65,000 Impressions** - Jaeden shares that his site AIChatDaily.com passed 65,000 Google impressions and provides daily AI deep dives and a newsletter. - He built the site himself, publishes ~20 articles daily, and sees growing traffic and search rankings for AI terms. Transcript: Jaeden Schafer I want to kick this episode off by saying I've officially passed 65,000 impressions on Google for my vibe coded project, AI chat daily.com. Essentially it was, it's a website that covers all of the same stories I cover on the show here. And it gives deep dive articles with all of the information, way more info on what actually happened. Um, but it's kind of just the show notes to this, to this podcast. And I've created this site and essentially it is, uh, I think it's going to become a competitor to tech crunch and the verge and a lot of other big, uh, companies. It's very specifically on AI news and we have a newsletter. Anyways, I've been blown away by how popular it has become and how many people are starting to go there for their daily AI news. People are tagging me on LinkedIn posts that don't even listen to the podcast that somehow have found it. If you search for AI chat daily on Google, it's the top result, which I'm super excited about. And it's just something I'm thrilled with. I think it's beautiful. It's been so much fun. If you're interested in getting daily AI news to your inbox, go subscribe on AI chat daily to our newsletter. And if you want to see the latest in AI news or just what my latest vibe coded project looks like, I 100% built this thing myself. It's an amazing tool. It has like 20 articles every day, the top 20 stories that are happening in AI, and it does really good deep dives. Go check it out. I look at it every day. I'm tweaking this thing every day. I'm working on it every day. It's so much fun, but I'm blown away. ([Time 0:01:35](https://share.snipd.com/snip/5e20a74f-fd75-499d-b640-dc74206a0e9b)) - **Opus 4.8 Boosts Browser Agents Without Raising Price** - Opus 4.8 improves browser-agent performance (84% benchmark) while keeping API pricing equal to 4.7. - New CloudCode dynamic workflows spawn hundreds of parallel sub-agents for large code migrations, speeding huge projects for enterprise customers. Transcript: Jaeden Schafer They've just released the latest, which is Opus 4.8. It hit a 84% on browser agent benchmarks, and it is beating GPT 5.5. It's also holding Opus 4.7's API pricing, and it's not going to actually increase. So the API pricing isn't increasing for the new model, but it is significantly better on what I think is the most important metric today, which is browser agent. And the reason why I think this is so important is because right now we really need these computer use agents to do well. Cloud Cowork, I have been absolutely blown away by basically every task I give it, every website I send it to, every insane spreadsheet I download within a ton of data, it's able to parse And give me the feedback and the information I need. I'm blown away by it on a daily basis. You don't need me to continue to pump it all day long. But I think this is impressive that it's getting better and it's the same price. CloudCode's new dynamic workflows have also been announced with this kind of in tandem. And this is going to let Opus 4.8 spawn hundreds of parallel sub agents in a single session, which all of them are capable of handling code-based migrations, spanning hundreds of thousands Of lines of code. So you're really truly going to be able to give Opus 4.8 way bigger projects. It's going to be able to take those and get it done, not just, I wouldn't say cheaper, but faster. So all the sub-agents are still going to be burning through tons of tokens. I think this is actually awesome because I have a lot of projects where it will take an entire day to get something done. It does them, you know, it's like spinning up, I think it right now, it spins up like 10 sub agents is what I can get Claude cowork to do for me. Anything more sometimes 20, but anything more than like 20, I think. And it's like, it just sacrifices quality, I think it goes to cheaper models. It's just not as good. And so when I have, you know, a huge project, like recently I was vibe coding a Bible study app and I needed to write, you know, 1300 Bible study pages for every chapter in the Bible, which Was like crazy. It took an entire day to do that. And, you know, it would, it would run for like half an hour and I'd be like, Hey, I did like this batch. And I'd be like, continue. And just every 30 minutes I'd say, continue. I was like working on my normal stuff. And on the side, I was just saying like, continue, continue, continue. It burnt through all of my credits. So I think that's also something to take into consideration. I got locked out of Claude Code and Co-Work for two days because of that. So there's a lot of new problems. But if you're on the API, you wouldn't get that problem. You can go buy extra credits, which is what they want you to do. And for big enterprises with massive budgets, being able to get something done in one day that you would have taken you a week before because you have 100 sub agents, I think is some real Value. ([Time 0:05:19](https://share.snipd.com/snip/0c5358cd-a311-456f-9518-a52edccd1daf)) - **Daylong Cowork Run Burned Credits And Got Locked Out** - Jaeden describes using Claude Cowork to parse complex websites and massive spreadsheets, praising its ability to deliver actionable results repeatedly. - He recounts a day-long job creating 1,300 Bible study pages where he repeatedly told Cowork to 'continue' and burned through credits, even getting locked out for two days. Transcript: Jaeden Schafer Cloud Cowork, I have been absolutely blown away by basically every task I give it, every website I send it to, every insane spreadsheet I download within a ton of data, it's able to parse And give me the feedback and the information I need. I'm blown away by it on a daily basis. You don't need me to continue to pump it all day long. But I think this is impressive that it's getting better and it's the same price. CloudCode's new dynamic workflows have also been announced with this kind of in tandem. And this is going to let Opus 4.8 spawn hundreds of parallel sub agents in a single session, which all of them are capable of handling code-based migrations, spanning hundreds of thousands Of lines of code. So you're really truly going to be able to give Opus 4.8 way bigger projects. It's going to be able to take those and get it done, not just, I wouldn't say cheaper, but faster. So all the sub-agents are still going to be burning through tons of tokens. I think this is actually awesome because I have a lot of projects where it will take an entire day to get something done. It does them, you know, it's like spinning up, I think it right now, it spins up like 10 sub agents is what I can get Claude cowork to do for me. Anything more sometimes 20, but anything more than like 20, I think. And it's like, it just sacrifices quality, I think it goes to cheaper models. It's just not as good. And so when I have, you know, a huge project, like recently I was vibe coding a Bible study app and I needed to write, you know, 1300 Bible study pages for every chapter in the Bible, which Was like crazy. It took an entire day to do that. And, you know, it would, it would run for like half an hour and I'd be like, Hey, I did like this batch. And I'd be like, continue. And just every 30 minutes I'd say, continue. I was like working on my normal stuff. And on the side, I was just saying like, continue, continue, continue. It burnt through all of my credits. So I think that's also something to take into consideration. I got locked out of Claude Code and Co-Work for two days because of that. So there's a lot of new problems. But if you're on the API, you wouldn't get that problem. ([Time 0:05:50](https://share.snipd.com/snip/84bedb2e-edd2-41cd-9f96-30bbdbdabda9)) - **Shift Trades Free Cleanings For Robot Training Footage** - Shift offers free home cleanings in exchange for POV footage from workers wearing 'magic hats' to train robots for household tasks. - Faces, screens, and IDs are blurred before data enters training pipelines; Shift plans expansion and future verticals like plumbing and cooking. Transcript: Jaeden Schafer Okay, we have to talk about the new company Shift. They are offering free home cleanings in exchange for robot training footage. I think this is interesting. The company is giving what they're calling magic hats to all of their cleaners, all of their workers, and it basically has a POV, a point of view footage. So it's a camera pointing straight down. So it's kind of looking at your hands and like right in front of you of what you're actually seeing. They say that all of the faces that this camera is going to capture, because I mean, it's inside of people's homes, So many faces, screens, ID cards, all that's gonna get blurred out before The video goes into the AI training pipeline. They've already paid 10s of 1000s of people across 15 different countries to record activities with their app. I think I saw on their website, you can go and sign up to be like a worker on shift. And it will give you like little jobs like, hey, go film a video of XYZ thing, or you picking up a stick, or you doing this thing or, or something, and they'll pay you for all of those. So this is really interesting having this cleaning right now, I think what they've done is identified area where where it was going to pay a lot in this data is going to be very valuable. All of these humanoid robots that are rolling out right now, they want that data to train the robots to be able to actually go and do cleaning tasks. So I think they're getting ahead of the data set right now. They are expanding, they have some plans to expand to San Francisco, London, Zurich and Munich very soon. And they said that plumbing, cooking and building ([Time 0:07:57](https://share.snipd.com/snip/ab28fb82-30ae-4f6a-8732-c1178e8a41ea)) - **AI Helps Speed Rare Disease Diagnosis** - Boston Children's is using OpenAI tools to help diagnose 40+ rare disease cases, targeting the long diagnostic lag for rare conditions. - Host highlights rarity complexity: ~7,000 known conditions often taking years and multiple specialists before diagnosis. Transcript: Jaeden Schafer Very, very wild. Boston Children's is using open AI tools to diagnose 40 plus rare diseases. This is one of the areas I am the most excited about AI is when it's actually able to help with healthcare. This problem in particular, I think is a big deal because rare diseases take patients an average of five or more years to diagnose. And so when you have AI and it has its pattern matching, and you're putting this on structured patient data, I think this is going to reshape how a lot of hospitals are trying to approach This, which I am thrilled by. This is really one of the hardest things to diagnose. And a lot of these really rare diseases, right? Like you go to your doctor, I've seen so many people, they go to the doctor. It takes them years. They say all of the symptoms. They try all of the most common things. When it's none of that, you really have to go down a rabbit hole. And sometimes I've heard of people getting very lucky. Personally, I feel like it's blessed, right? I'm like, I'm a person of faith. So I'm like, okay, God help that person in a lot of these kind of disease, rare disease diagnoses. But I don't want that to have to happen to everyone. This is a terrible place to be at where there's all these rare diseases. It's very hard to diagnose and it takes years. People have to suffer with them before they can find a diagnosis. Sometimes it's too late. And so I'm really excited to have AI with all of its pattern matching be able to do this, especially when you consider the fact that rare diseases are affecting right now about 7000 known Conditions and each is hitting fewer than 200,000 people. So patients typically see multiple specialists over years before they're diagnosed. Anyways, OpenAI has withheld a lot of the details about this particular study. Obviously, it's kind of healthcare, so there's a lot of confidential stuff. They didn't say what models were being used. They didn't say how many total cases were being screened to give those 40 different diagnoses. And they also didn't say what the false positive rates were. But in any case, I think we can be pretty confident that it is making some big strides. ([Time 0:09:26](https://share.snipd.com/snip/1aebd67e-5fe7-4218-abc9-12534aff773a)) - **Grok Raises To Become An Inference Cloud** - Grok is raising $650 million from existing investors to build an inference cloud after NVIDIA's $20 billion licensing deal. - The move shifts Grok from hardware toward token-based cloud inference to compete with hyperscalers on latency-sensitive workloads. Transcript: Jaeden Schafer They are looking for $650 million from their existing investors. This is after Nvidia did a $20 billion licensing deal. So they're raising this new, they're raising these new funds to build out their inference cloud service. And this is just five months after Nvidia give them, you know, that they did the $20 billion deal. So this isn't very, you know, this isn't very long after. I think it's especially interesting when you consider the fact that when NVIDIA did that deal, they licensed all of their chip technology, which is what they're really after. And they also hired away all of their senior leadership. But this new $650 million that they're raising is essentially going to actually keep Grok independent. So it wasn't straight up acquisition. And I think a lot of people were like, oh, they're going to do a licensing deal because then it can't get blocked, right? No one wants acquisitions and mergers to get blocked by the government or yeah, basically by the government. So they're like, oh, it's just a licensing deal. It's easier. It's much more palatable that way. But what's interesting is that also gives the opportunity for Grok to remain completely independent. It's pivoting them away from the hardware sales and more towards a token based cloud business. So they're going to be competing directly with Cerebrass, which is the hyperscaler that recently went public. They've also essentially went out to all of their investors and said, hey, like we're raising the $650 million, but is it basically guaranteed that they're going to close it at this Point? Because Disruptive and Infinitum, two of their investors have guaranteed the full $650 million and they've agreed to cover any shares that other investors are declining on, which Is basically making this a guaranteed. But I think they want to get investment from all of the previous investors. So everyone's kind of united on this new move forward. And there is so much potential here because inference workloads now dominate production AI spending for most companies. And it's, you know, it's really kind of passing a lot of the training compute. And it's making per token revenue way more attractive than just per chip sales. So I think this is a good move on Grok's part. They know that they can't outcompete NVIDIA in the GPU market. So they're betting on custom silicone for latency sensitive high volume inference. I'm excited to follow along. ([Time 0:11:22](https://share.snipd.com/snip/8b8cf389-ca07-4c37-a521-1c976e37b0b6)) - **Asana Buys Stack AI To Own Work Automation** - Asana acquired no-code agent builder Stack AI for $75 million to own the work-management layer and add automation agents. - StackAI founders join Asana; acquisition values Stack at ~3.75x its prior funding after Y Combinator. Transcript: Jaeden Schafer The final story of the day, Asana is buying the no code agent builder Stack AI for $75 million. This is something I'm personally very excited about considering Stack AI is one of the original competitors to AIbox.ai, my own startup. So I'm always thrilled to follow along and see these companies continue to grow and be successful. And by the way, if you want to get access to over 80 different AI models in one place and build tools with them, like you do with Stack, go check out my startup AI box.ai. I felt like that's a pretty natural plug on this story. But in any case, this move right now, I think is showing that Asana is betting that controlling the work management layer where a lot of teams are actually assigning tasks and tracking Progress is a lot more defensible against AI disruption than trying to compete with OpenAI or Anthropic directly, especially when it feels like Anthropic is eating everybody's lunch In every single category. StackAI has raised under $20 million total. And so the $75 million price is about a 3.75x markup. And the most recent $16 million Series A that they did was backed by Gradient, Epiklon Capital, and Vercel's CEO, Guillaume Rauch. Right now, Stack AI's co-founders, Tony Russell and Bernard Asituno, they're going to be joining Asana. The startup was part of Y Combinator's winter 2023 cohort. So they've been through Y Combinator. They built agents that basically automate processes. They pull data from Salesforce, Slack, and Google Suite, and they let you build agents with all of that. Asana has lost more than half of its market cap since Chad Chippity launched. So I think this is some good context to put in there. I think this was all kind of accelerated after their founder, Dustin Moskivitz, stepped away as the CEO back in March of last year. Time will tell if all of these acquisitions they're lining up right now are going to help turn the company around. I really hope it does. And I'm rooting for Stack AI and for Asana at this point. ([Time 0:13:18](https://share.snipd.com/snip/6524d9e9-fb1a-4567-bbd0-f4755b6725c5))