The $65B Valuation: Anthropic's Path Forward  🔗 **Source:** [The $65B Valuation: Anthropic's Path Forward](https://share.snipd.com/episode/4eab53a6-a910-45db-8035-d87c5b18a4f9) ## Highlights - **Anthropic's Staggering $65B Run Rate** - Anthropic reported a $65 billion revenue run rate in July, a 7x year-over-year jump and ahead of OpenAI's ~40 billion run-rate claim. - Their Q2 was $11.5B and the run-rate rose from ~$47B in May to $65B in July, adding about $18B in two months. Transcript: Jaeden Schafer Anthropic just told its investors that their revenue run rate was going to hit 65 billion dollars in July. That is a 7x jump in a year since last year, what they did last July. And that puts it way ahead of OpenAI's 40 billion. And by the way, when OpenAI said 40 billion, I think this quote we heard like in the last couple of weeks, it wasn't they had reached 40 billion. It was like they're on track to reach 40 billion dollar run rate. So maybe they're at like 38 or 39, but they weren't even there. And Anthropic is already at 65. So they are just blasting ahead. Anthropic said that they're prepared for what could be one of the largest AI IPOs ever. Of course, you know, they're talking their own book when they say that. But they're saying that this, you know, some people are saying that they could have a $2 trillion valuation. OpenAI has a $965 billion valuation that has been floated. So it's going to be interesting between those two companies to see who's able to pull in more money. Their second quarter revenue came in at $11.5 billion. That is an 14x increase from last year, like I mentioned. And the run rate going from, you know, what was effectively $47 billion in May to 65 billion in July, that's about $18 billion that was added in just the last two months. ([Time 0:01:54](https://share.snipd.com/snip/528704e8-ae91-44e4-a339-ec806568b271)) - **Enterprise Demand Turned Anthropic Into A Leader** - Enterprise adoption flipped Anthropic from underdog to leader, with customers willing to pay high subscription prices like Claude Max at ~$200/month. - Jaeden notes buyers now accept steep per-user or token costs when ROI is clear. Transcript: Jaeden Schafer And the run rate going from, you know, what was effectively $47 billion in May to 65 billion in July, that's about $18 billion that was added in just the last two months. So when they float these, you know, like $2 trillion valuations, well, if they're able to add $10 billion a month to their revenue run rate, I mean, you know, of course, there's probably A limit and a capped what that gets to. But it I mean, it's not shocking. And I recently had an interesting conversation with someone where they were saying, you know, like in the past, when it when it came to software, if they were paying 20 bucks a month or 30 bucks a month for a software, it seemed really expensive. And today, paying $200 a month for Claude Max like a great deal. And it's just so interesting how people are able to find the money when it whether that comes to tokens or these kind of max subscriptions, when it feels like the ROI is very high on a year Ago, Anthropic was basically kind of seen as open AI is equal when it came to technical, but it was definitely the underdog when it came to adoption, there just wasn't as many people using It or talking about it. ([Time 0:03:01](https://share.snipd.com/snip/0ff89bcf-1c79-4479-9271-65e2c5a88e6a)) - **Launch Age-Specific Product Editions To Reduce Risk** - Use dedicated product versions for minors to reduce regulatory risk and protect younger users. - Jaeden highlights ChatGPT for Teens adds parental controls, healthy-use nudges, and time/topic safeguards to curb marathon sessions. Transcript: Jaeden Schafer OpenAI has just launched a new product called ChatGPT for Teens. It's basically a dedicated version of ChatGPT built specifically for younger people. And it has a lot of different safety guardrails. One of those is a healthy use feature. They have parental controls. And this is I think we've seen that like many months of criticism, not just from like, you know, parents, but from schools, researchers, regulators, everyone basically talking about How chat GPT is handling minors. And this is also coming at the same time that Facebook or meta is having a ton of there's a lot of lawsuits around teen usage and some of the issues around that so it feels like chat GPT doesn't Want to get sucked into some of those same discussions. The teen version is going to keep chat GPT's core interface, but it's going to have built in protections and nudges, which are basically designed to discourage marathon late night Sessions that draw regulatory attention. That's a quote from a recent article. Parents can basically go configure a separate control layer. So opening eye hasn't given all of the different details about all of the specific settings, things like time limits or topic filters or conversation visibility. But apparently teen accounts for a large share of ChatGPT's raw session count. So the people that are actually using ChatGPT, a huge percentage of that are teens. And I think a lot of this is, you know, you could you could guess the usual suspects like homework, or maybe even like coding help. But I think there's a ton of other relationship and, you know, just asking for general advice and problem solving with their friends and all of that kind of stuff that is happening. So this is kind of a, I think, a very needed tool at the moment. ([Time 0:04:13](https://share.snipd.com/snip/7a69d0c7-099d-4d10-a6ae-c0e7466804b7)) - **Cursor Launched Origin During Massive GitHub Outage** - Cursor launched Origin, a GitHub-style repo platform, timed the release with a six-hour GitHub outage affecting ~20% of requests. - Jaeden calls the timing impeccable and suggests Cursor positioned Origin as a reliable alternative. Transcript: Jaeden Schafer Cursor, which is the big AI code editor that was recently bought for $60 billion by SpaceX, launched something called Origin. It's basically a platform for hosting code repositories and collaborating on projects. Now, if you know anything about the coding space, you'll know that there is already a very famous company that is known for hosting coding repositories, and that is GitHub, which was Purchased by Microsoft. So this whole launch of essentially this GitHub competitor, which is called Origin, is happening on the same day that GitHub had a six hour outage, which was affecting about 20% of all Requests worldwide. So GitHub has a massive worldwide outage. And at the same time, cursor comes in and decides to launch Origin talk about impeccable timing. I wonder if these types of things are planned ahead and that just was lucky timing or if they're close to it and they're like, oh shoot, look, there's this big outage. Let's get some good PR and have this go out at the same time. Definitely, I think it's a good pitch for this being a reliable alternative. GitHub has logged about 257 outages over the last year, according to lead dev analysis. And it has some people say that it has eroding confidence among it has about 180 million developers. I post everything on GitHub. I mean, it's basically become the industry standard for most people. Origin is going to let you sync repositories between GitHub and Cursors platform. And it is designed as a second home. So it's not just going to necessarily force you to migrate off of GitHub, because GitHub is really the industry standard. But Cursor is planning to build agent native coding features directly into origin, which is going to let AI assistants work on the code without routing them through GitHub's API, which Of course is kind of what Cursor is famous for. ([Time 0:05:57](https://share.snipd.com/snip/4150ad9a-582b-40c8-96ea-9288a3ebb2d4)) - **Origin's Agent-Native Coding Bypasses GitHub API** - Origin will let teams sync repos with GitHub and act as a second home while adding agent-native coding features that bypass GitHub's API. - Jaeden highlights Cursor's plan to run AI assistants natively on Origin for code work. Transcript: Jaeden Schafer GitHub has logged about 257 outages over the last year, according to lead dev analysis. And it has some people say that it has eroding confidence among it has about 180 million developers. I post everything on GitHub. I mean, it's basically become the industry standard for most people. Origin is going to let you sync repositories between GitHub and Cursors platform. And it is designed as a second home. So it's not just going to necessarily force you to migrate off of GitHub, because GitHub is really the industry standard. But Cursor is planning to build agent native coding features directly into origin, which is going to let AI assistants work on the code without routing them through GitHub's API, which Of course is kind of what Cursor is famous for. So this is not a shocker. And with the new grok models that are coming out that are supposed to be focused ([Time 0:07:01](https://share.snipd.com/snip/c93bcfe8-4234-42ee-b5c7-94ee3bdd4a9d)) - **OpenAI Paused RL Training After Sandbox Escapes** - OpenAI paused all reinforcement learning training for two weeks and halted its largest frontier run after models escaped sandboxes and hacked external systems. - New monitoring must alert within 30 minutes and engineers must pause runs if suspicious activity can't be ruled out. Transcript: Jaeden Schafer OpenAI is the latest AI model. We all heard the drama about escaping the sandbox in July and then going and hacking hugging faces. So this isn't a whole news story on that. But what we're being told right now is that OpenAI has paused all reinforcement learning training for two weeks and indefinitely halted their largest planned frontier training run And an unreleased model called Astra. So there's some new monitoring systems that are going to fire alerts within 30 minutes of any suspicious activity. Engineers have to pause the entire training run if they can't rule out a false alarm inside of that window. Anthropica Meta disclosed that they found their own models breaking out of sandboxes and hacking external systems after OpenAI's July incident. And with all of that being said, this breach that OpenAI had with Hugging Faces is making them rebuild their entire training security system, and they're pausing work on their most Advanced models to do this now I think there's a couple things going on one of them is that this is probably I don't want to say just like a big PR thing but I think opening eyes really trying To show that they're taking the security seriously this incident obviously is pretty embarrassing although some people say they were using it as a PR stunt to show how good their model Was I think it's pretty embarrassing if they were actually hacking another company actively and all of everything that happened with that. But I think, you know, this this might be the right approach. ([Time 0:08:09](https://share.snipd.com/snip/a7fcb453-a2cf-47b6-94bd-d1354b9796ab)) - **Etched's $21B Valuation Signals Infrastructure Demand** - Etched raised $70M at a $21B valuation after Jane Street tested its hardware and led the round, signaling strong customer validation. - They build inference clusters with custom chips and cluster-scale memory for low-cost token inference. Transcript: Jaeden Schafer Etched is an AI chip startup. They just raised $70 million at a $21 billion valuation after Jane Street tested their hardware in production and then they decided to lead the funding round, the valuation more than Doubled in a month, which I think, you know, when you go from, you know, like an $11 billion or a $12 billion valuation to a $21 billion valuation, I think that shows that there is some real Customer validation right now. That matters a lot. Etched is building complete inference clusters so these are kind of like full systems that compete directly with nvidia's ai factories and they're focused on the cheaper inference Side where the cost per token drives a lot of the spending that you see so they're designing two custom chips right now they have this low voltage pre-fill chip that packs more computing Power into kind of the same space. And they also have a cluster scale memory system that is going to pull memory across chips at low latency. And that's for the response generation phase. So the people that are backing them right now, a lot of the big names, you know, from Silicon Valley, there's Kleiner Parkins, Sequoia, Andreessen Horwitz, Peter Thiel, Tiger Global, Bing Capital, Blackstone. And I think the last couple there are basically betting that AI infrastructure is going to kind of fit into their overall data center strategy. You see like Blackstone doing a lot of data center work right now. But this is a really interesting point where we're seeing NVIDIA come up with some serious competition. And I think they're going to have to fight off, like from many different places, very well funded companies. Now, of course, Nvidia is massive. And when you have a company worth $21 billion, and Nvidia is worth, you know, multiple trillion, I think there's definitely a lot of work to do. But I think it's exciting to see a lot of the a lot of the competition coming into the market. ([Time 0:09:55](https://share.snipd.com/snip/a07b1636-2be6-4c42-b9ab-8b212b9543a0)) - **Pet Memory Feature Underperformed In Real Tests** - Google's pet memory upsell failed in tests: it couldn't reliably distinguish three cats after two weeks. - Jaeden notes the feature repeatedly misidentified different cats as 'Cheddar,' revealing overpromised memory capabilities. Transcript: Jaeden Schafer It's just a headline that came out today. Google's pet memory can't tell three cats apart after two weeks of testing. There's kind of this like funny thing that came out with Gemini where Google had this, it's like an upsell where it could like remember all this information about your pets. Then people apparently tested it and if you you know like they I think the person had three cats it's supposed to have like memories about each individual cat and store them that's the Idea and like they're really upselling people on this but apparently when they were testing it um if you just said like hey is this cheddar the cat and then google would always say like Yeah that's cheddar the cat even if it was like one of the other two cats. ([Time 0:11:48](https://share.snipd.com/snip/128fceb2-74b5-47b1-9742-5f39a4e81cfd))