Anthropic, Robots, and the Startups Caught in the Middle

Lauren: Mm-hmm. Okay, welcome back to Tech Insider Weekly. I'm Lauren, and Derek, I have to say this week's lineup is genuinely a lot. Derek: A lot is an understatement. Like, where do you even start? Speaker 3: Anthropic, you start there. Derek: Right. So Axios confirmed it this week. Anthropic just overtook OpenAI as the most valuable AI startup on the planet. $965 billion valuation. And then they filed confidentially for an IPO. Speaker 3: The Safety First AI Lab is about to ring the bell on Wall Street. We're going to dig into what that actually means for who they get to be. Get to be going forward. Derek: Should be interesting. Speaker 3: And then get this: Robots, like a lot of robots. Derek: So many robots. Speaker 3: The L.A. Times had this wild piece on the people physically puppeteering humanoids to teach them to make coffee. Sam Altman is quietly backing a startup called Alfred, and there are startups pitching battlefield humanoids on aggressive timelines, Derek: Yeah, 12 to 18 months, according to CNBC, which is either bold or terrifying, depending on your outlook. Speaker 3: Possibly both. Derek: We're also getting into the AI infrastructure stack, Nvidia chasing the $200 billion CPU market, Groq raising $650 million after its Nvidia deal, and a South Korean chip startup called Xcena making a contrarian bet that memory, not compute, is AI's real bottleneck. Speaker 3: And we close out looking at who's getting left behind. CNBC and Crypto Briefing are both reporting that AI companies captured around 80 percent of Q1 2026 venture funding, pre-ChatGPT startups are in a tough spot. Derek: It's a brutal reshuffling. Speaker 3: All right, let's get into it. Anthropic is first up and the story's bigger than the number. Okay, so get this. Anthropic just crossed $965 billion in valuation, not OpenAI. Anthropic, the safety first, slow and steady, we're worried about AI destroying the world lab, is now the most valuable AI startup on the planet. Derek: That number stopped me cold, like I had to reread the Axios piece twice. Speaker 3: Right? And then filed for an IPO on top of it. Semafor reported the confidential filing this week. They're heading to public markets! Derek: So we went from anthropic is the responsible alternative to anthropic is the biggest and going public, basically overnight. Speaker 3: Overnight is generous, but yeah, the gap closed fast. Derek: Let's get real for a second, though. $965 billion. That's walking distance to a trillion dollars for a company that was founded as a reaction to OpenAI moving too fast. Speaker 3: The irony is so thick you could cut it. Derek: And they raised $65 billion to get here, according to Axios. That's not a funding round. That's a statement. Speaker 3: It's an entire industry's worth of capital concentration, and the IPO filing per France 24 puts them alongside OpenAI and SpaceX in this coming mega IPOs. The New York Times reportedly called it a tsunami of investment and employee wealth. Derek: Which, I mean, sure, but public markets are a different animal. Speaker 3: Completely different animal. Private valuations are vibes with spreadsheets attached. Public markets want revenue, margins, growth curves. Derek: Wait, back up. That's the tension I keep coming back to. Anthropic built their entire identity around safety research, around the lab that pumps the brakes. Does going public change that? Speaker 3: I think that's the right question, and I don't think anyone has a clean answer. Derek: Because public companies answer to shareholders every quarter. After safety research doesn't have a quarterly ROI. Speaker 3: No, it doesn't. And look, I want to be fair here. They made genuine safety work central to their model. Claude has a reputation for being more measured, more careful. Derek: Sure, Speaker 3: But the pressure to ship, to grow, to justify a near trillion dollar valuation? That's not a small force. Derek: that's a gravitational force. Speaker 3: And here's the thing that actually gets me. It's not that Anthropic did something. Something wrong-the story is weirder than that-they've built the thing that was supposed to be the alternative, and now they are the thing. Derek: The most commercially dominant AI lab is the one that said, wait, maybe we should be careful. Speaker 3: Plot twist, right? Semafor noted the IPO puts real pressure on investor appetite for big spending AI labs, and these are big spending labs. The compute costs alone are staggering. Derek: So the IPO is either a validation or a stress test. Maybe both. Speaker 3: Probably both. Public market investors are going to ask hard questions about It's about path to profitability that private investors politely skipped. Speaker 4: Derek: private investors write checks, public investors write downgrades. Speaker 3: Exactly; so the valuation holds up until it doesn't, and we won't know which until the S-1 actually lands. Speaker 4: What I keep circling back to is the competitive frame here. If Anthropic is now the top dog, what does that mean for OpenAI, for Google DeepMind? Does the safety-commercial balance shift? To shift across the whole field? Speaker 3: That's the downstream question, because everyone else is watching. If being safety focused and commercially dominant works, more labs copy the positioning. If the IPO stumbles, the lesson Wall Street takes is safety is a liability. Speaker 4: Yeah, the outcome of this IPO might actually shape what AI labs are allowed to believe about themselves. Speaker 3: High stakes for a confidential filing. Speaker 4: Very quietly consequential Speaker 3: And here's where it gets interesting. All of this, the valuations, the IPO filings, the capital flooding in, that's the software and model layer. But that same wave of money is building something physical, too. The question is, what does $965 billion worth of AI ambition actually look like when it gets off the screen and into the real world? Okay, so this is where it gets genuinely weird. The LA Times ran a piece this week about people whose whole job is to stand in a room wearing motion capture sensors, pouring a cup of coffee over and over hundreds of times a day. Speaker 4: Wait, that's literally someone's job. Speaker 3: That is literally someone's job. They're called robot puppeteers. Lauren: Bernardo Flores, one of the guys the LA Times profiled, spends eight hours a day pouring the same cup of coffee to train humanoid robots, pours it, empties the mug back into the pot, does it again. The repetitiveness, it can cause some discomfort. Derek: That's almost word for word what he said. But here's why that detail matters. The only way these humanoid robots learn physical tasks right now is through massive amounts of human demonstrated motion data. data. You can't just write a prompt for pouring coffee without spilling. Lauren: Right, which makes the Sam Altman connection interesting. Business Insider reported this week he's quietly backing a startup called Alfred, run by former Tesla and Meta employees, building software for exactly this layer, not the robot hardware, the intelligence layer on top. Derek: And Khosla Ventures is also in. Alfred is physical AI, the bet is that the software stack for controlling robots across Cross cars, manufacturing, whatever. That's where the real money lands. Lauren: Here's my operator question, though. Who is actually buying these things today? Because there's a big gap between impressive demo and someone cutting a purchase order. Derek: Fair. And then there's the military angle, which is a different category entirely. Lauren: Yeah, CNBC covered Foundation Future Industries, a San Francisco startup founded in 2024, ties to the Trump family, and their pitch is humanoid robots. robots in military and hazardous environments. They're talking deployment in 12 to 18 months. Derek: 12 to 18 months. Lauren: For battlefield humanoids? Derek: I mean, I get the appeal of the framing, send a robot instead of a soldier into a dangerous situation. That argument basically writes itself. But the gap between a robot that can pour coffee after thousands of training hours and a robot that can operate autonomously in a combat environment is... Not small. Lauren: Not even close. And then you've got Shifters, the Israeli startup that just raised a $10.2 million seed round led by Ace Capital Partners. They're building AI-native autonomous ground robots for high risk environments. Total funding? $15 million since they were founded in 2023. Derek: So you've got three completely different visions of what robots are actually for. Coffee, cars, combat. Lauren: And they... They all need the same underlying thing, good enough AI to handle real-world messiness, whether that's a spilled latte or something a lot higher stakes. Derek: Which is the actual thread here. The puppeteering, the Alfred software stack, Foundation and Shifters, they're all working on the same hard problem. How do you get a physical machine to interpret an environment it's never seen before and act on it reliably? Lauren: And nobody has fully solved that. The coffee demo works because they ran it hundreds of times in a controlled setting. The battlefield does not offer controlled settings. Derek: Understatement of the year. Lauren: So I'm watching this space with genuine curiosity and some real skepticism about the timelines being pitched. 12 to 18 months for military humanoids feels more like a fundraising slide than an engineering roadmap. Derek: Yeah, and that's exactly the tension. The capital is moving fast, the technology is moving a little slower, which actually connects to something underneath all of this. Robots need chips, and right now the chip race has its own wild storylines. NVIDIA moving into CPUs, a startup raising $135 million on the argument that memory, not raw compute, is AI's real bottleneck. Lauren: Oh, that memory argument is interesting. Derek: It really is, and that's where we're going. Lauren: So on the chip side, the story just got a lot interesting. Derek: Right? NVIDIA opened Computex in Taipei and basically said, hey, we want the CPU market too. They unveiled this new chip called the RTX Spark, a one petaflop super chip designed to run AI agents locally on your PC. Dell, HP, Lenovo, Microsoft Surface. According to TechCrunch, that whole lineup is coming this fall. Lauren: And that CPU market they're going after? two hundred billion dollars. That's Derek: Wow. Lauren: not a side bet. Derek: Not even close. And the timing is interesting, because Groq, which Nvidia basically did a not acquihire on for $20 billion earlier, is now reportedly raising $650 million from existing investors. TechCrunch had that from Axios sources. Lauren: So Nvidia absorbs your best people and your IP for $20 billion, and then you go raise more money. Derek: Yeah, the inference neocloud lives on. Look, the chip market right now is just everyone is moving, everyone is raising, which brings me to the story I actually can't stop thinking about. Lauren: The memory one. Derek: The memory one. OK, so TechCrunch covered this South Korean startup called Xcena. They just raised $135 million at a $560 million valuation, and their whole thesis is that the real bottleneck in AI isn't compute, it's memory. Lauren: Walk me through that, because that's a counterintuitive pitch when everyone's obsessed. Obsessed with GPU horsepower. Derek: So here's how it actually works. Every time you ask ChatGPT something, your request goes memory, then CPU for pre-processing, then GPU for the heavy compute, then back to memory. And that round trip, it happens for every word the model generates. Lauren: Every word? Derek: Every word. So you're routing through some of the most expensive, power-hungry chips in the industry on every token. Xcena's argument is that if you fix the memory layer, There-you remove that structural inefficiency entirely. Lauren: That's a nice story, but show me the numbers. Is there evidence this actually scales? Derek: Honestly, that's the open question. A hundred and thirty five million says some serious investors think it does, but yeah, the proof is in the deployment. Lauren: Fair. Okay, now flip that on its head. While startups are betting on new silicon, Microsoft is building the connective tissue above all of it. of it. Derek: MXC? VentureBeat had a good breakdown. Microsoft launched an OS-level sandbox for AI agents built right into Windows. The idea is that when an agent goes wrong, and they do go wrong, there's a secure runtime container to catch it. OpenAI and Nvidia are already on board. Lauren: And separately, TechCrunch covered ASSERT, Microsoft's new open source framework for AI behavior testing. Developers describe what they want the AI to do in plain text and it spins up the evaluations. Derek: Which honestly has been a problem that's been quietly painful for every team shipping AI products. You build the model, you ship it, and then does it actually behave the way you intended? Lauren: Right. Microsoft is positioning itself as the layer everything runs on top of. Chips, agents, safety rails, that's a lot of surface area. area. Derek: And here's the thing, all of this infrastructure and Nvidia Lauren: The CPU push, Xcena's memory bet, Microsoft's sandbox, it's being built for AI-native companies, companies designed around these assumptions from day one. Derek: Which raises an uncomfortable question about everyone else, the companies that weren't built that way. Lauren: That's exactly where the money tells a harder story. Derek: Okay, so the infrastructure we just walked through, all of it is being built for AI-native companies, and that raises a pretty uncomfortable question about everyone else. Lauren: Yeah, because CNBC ran this piece this week, headline literally says disrupted or dead, and they're talking about an entire generation of startups that launched before ChatGPT and are now just running out of road. Derek: I've watched this pattern before, Derek. Platform shifts happen, and the companies that got... That got built on the old assumptions don't always see it coming until the funding dries up. Lauren: Right. And the numbers here are brutal. Crypto Briefing reported that in Q1 of 2026, roughly 80% of all global venture funding went to AI companies, about $242 billion out of $300 billion total. Derek: 80%. Lauren: In four mega rounds ate up 65% of the whole quarter. Whole quarter, OpenAI pulled $122 billion, Anthropic got $30 billion, xAI $20 billion, Waymo $16 billion. Derek: So what's left for everyone who built a scheduling tool or a subscription product back when cheap money made those look like great bets? Lauren: Not much. And here's the part that I find technically interesting. These companies aren't just losing funding because investors chased something shinier, their core assumptions got quietly invalidated. Derek: That's the part that should scare founders. GPT-4 didn't announce it was coming for your roadmap. It just... Arrived. Lauren: Exactly. If you built a company around automating a narrow workflow, there's now a general purpose model that does that workflow for $15 a month. Derek: Let's get real for a second. I've seen this in rooms I've actually sat in. A founder raises a Series B, builds a team, starts scaling, and then the underlying capability they were selling becomes a feature in someone's API. That's not a pivot opportunity. That's a ceiling. Lauren: And over 1,500 pre-ChatGPT unicorns are reportedly sitting at risk of down rounds right now. Derek: Over 1,500. Lauren: So here's the honest closing question. If you were starting a company today, knowing all of that, where would you even start? Derek: I'd look for the problems that get harder as AI gets better, not easier. Data quality, trust, regulation, physical world integration, the messy stuff that models can't just generate away. Lauren: Interesting. So you're betting on friction. Derek: Betting on the stuff that actually requires humans to care, which weirdly turns out to be a lot. Lauren: Friction as a moat. There are worse strategies. Derek: There really are. Okay, that was a lot in the best way possible, Lauren: Right? We went from Anthropic nearly hitting a trillion dollars to Fernando Flores pouring coffee eight hours a day. Derek: which honestly might be the most relatable job in physical AI right now. Lauren: The threat connecting it all, though, the question of who these companies are actually building for—shareholders, safety, the military, your morning routine? Derek: And that Anthropic IPO framing stuck with me. Public markets don't wait for safety research to have a quarterly return. Lauren: Yeah, that one's going to keep coming up. Derek: Warmly, if this episode got you thinking, subscribe wherever you listen. Leave us a review. It genuinely helps. And tag us on social if there's a founder or topic we should be covering. Lauren: New episodes every Wednesday. Thanks for spending the hour with us. Seriously. Derek: Smiling. We'll see you next week.