WEBVTT 00:00:00.000 --> 00:00:03.319 Intro music. So let's talk about a specific week, 00:00:03.439 --> 00:00:08.900 September 11th to 18th, 2025. Right. Because 00:00:08.900 --> 00:00:11.580 something really shifted in AI funding then. 00:00:11.619 --> 00:00:13.240 It wasn't just about bits and bytes anymore. 00:00:13.400 --> 00:00:16.539 It got physical. Yeah, exactly. That was the 00:00:16.539 --> 00:00:19.920 week we saw that huge headline. Figure AI, you 00:00:19.920 --> 00:00:21.519 know, the robotics company, they pulled in a 00:00:21.519 --> 00:00:24.579 Series C round of over a billion dollars. A billion 00:00:24.579 --> 00:00:27.399 dollars for a humanoid robot. I mean, that's 00:00:27.399 --> 00:00:29.260 such a clear signal, isn't it? It really is. 00:00:29.559 --> 00:00:32.119 It says the next big step for AI, it isn't just 00:00:32.119 --> 00:00:34.359 about like thinking faster inside a computer. 00:00:34.520 --> 00:00:36.420 It's about doing things, actually interacting 00:00:36.420 --> 00:00:39.320 with the real world. Welcome to the deep dive. 00:00:39.479 --> 00:00:41.719 We've basically spent time digging through what, 00:00:41.799 --> 00:00:44.219 like 20 major funding announcements just from 00:00:44.219 --> 00:00:46.200 that one week. Yeah, quite a stack. And we're 00:00:46.200 --> 00:00:48.079 trying to see, you know, where's the smart money 00:00:48.079 --> 00:00:50.299 actually going? What are the real trends here 00:00:50.299 --> 00:00:52.899 for you? So our plan for this deep dive, we've 00:00:52.899 --> 00:00:55.460 grouped these these huge investments into three 00:00:55.460 --> 00:00:58.140 main buckets, three themes that seem to define 00:00:58.140 --> 00:01:00.990 what's coming next. First, we're looking at AI 00:01:00.990 --> 00:01:04.189 literally taking physical shape, you know, robotics, 00:01:04.370 --> 00:01:06.090 advanced manufacturing. Right. They're moving 00:01:06.090 --> 00:01:08.250 to the physical. Then second, we've got to look 00:01:08.250 --> 00:01:10.489 at the infrastructure needed for all this. We're 00:01:10.489 --> 00:01:13.450 talking specialized chips, security layers, the 00:01:13.450 --> 00:01:15.870 stuff that makes it all run. Exactly. And third, 00:01:15.989 --> 00:01:19.269 how AI is starting to automate really specialized 00:01:19.269 --> 00:01:22.920 high stakes work. Think science, complex coding. 00:01:23.099 --> 00:01:25.040 Got it. So let's start with that billion dollar 00:01:25.040 --> 00:01:27.280 bet, the physical world stuff. Yeah, let's dive 00:01:27.280 --> 00:01:29.680 in. The biggest signal has to be that capital 00:01:29.680 --> 00:01:32.219 flowing into AI that can, you know, actually 00:01:32.219 --> 00:01:35.040 touch and manipulate things. Figure AI's deal 00:01:35.040 --> 00:01:37.799 on September 16th was, well, it was the headline. 00:01:38.000 --> 00:01:40.260 It really was. And they're building general purpose 00:01:40.260 --> 00:01:43.310 humanoid robots. So basically robots shaped like 00:01:43.310 --> 00:01:46.269 us. Two arms, two legs, a torso. The idea being 00:01:46.269 --> 00:01:48.489 the sort of ultimate general purpose machine. 00:01:48.569 --> 00:01:50.629 Right. Something that can work alongside us. 00:01:50.750 --> 00:01:53.689 Yeah. And critically, operate in our world. Use 00:01:53.689 --> 00:01:56.670 the tools we use. Navigate spaces designed for 00:01:56.670 --> 00:01:59.069 humans. That's the key. Which, you know, directly 00:01:59.069 --> 00:02:01.709 hits things like labor shortages, automating 00:02:01.709 --> 00:02:04.150 jobs that are maybe dangerous or just hard to 00:02:04.150 --> 00:02:06.349 fill. Logistics comes to mind immediately. For 00:02:06.349 --> 00:02:08.909 sure. But OK, let's think about that billion 00:02:08.909 --> 00:02:11.330 dollars. That's a massive amount for a robot 00:02:11.330 --> 00:02:14.069 that's, let's be honest, still mostly in development, 00:02:14.210 --> 00:02:16.949 right? Yeah. What are the risks investors are 00:02:16.949 --> 00:02:19.069 taking on there besides just the tech not working 00:02:19.069 --> 00:02:22.270 out? That's a good point. I think maybe the biggest 00:02:22.270 --> 00:02:25.860 hurdle isn't always the tech itself. It's regulation. 00:02:25.879 --> 00:02:29.180 It's public acceptance. This kind of money, it's 00:02:29.180 --> 00:02:32.819 not just funding lab work. It's pushing to scale 00:02:32.819 --> 00:02:35.159 up manufacturing, deploy with partners, like 00:02:35.159 --> 00:02:37.819 now. It feels like investors are betting the 00:02:37.819 --> 00:02:39.939 regulatory side will sort itself out quickly, 00:02:40.039 --> 00:02:42.400 or maybe they intend to push it. Okay, that makes 00:02:42.400 --> 00:02:44.300 sense. It's a bet on deployment happening soon, 00:02:44.360 --> 00:02:46.360 not just maybe someday. And it connects to other 00:02:46.360 --> 00:02:48.159 things happening that week, right? It's not just 00:02:48.159 --> 00:02:51.120 the robot. Exactly. We also saw Divergent Technologies 00:02:51.120 --> 00:02:54.060 raise $290 million. They're building something 00:02:54.060 --> 00:02:57.069 called DAPS, the Digital Adaptive Production 00:02:57.069 --> 00:03:01.069 System. This is like manufacturing 2 .0. Yeah, 00:03:01.189 --> 00:03:04.590 DAPS. It mixes AI design, 3D printing, and then 00:03:04.590 --> 00:03:06.270 robotic assembly. It's a completely different 00:03:06.270 --> 00:03:08.129 way of thinking about production lines, moving 00:03:08.129 --> 00:03:11.250 away from those rigid old school factories. Totally. 00:03:11.349 --> 00:03:14.009 Instead of a factory built for just one specific 00:03:14.009 --> 00:03:17.330 part, DAPS is modular. It's defined by software. 00:03:17.530 --> 00:03:19.849 That means you can make complex things, maybe 00:03:19.849 --> 00:03:23.090 aerospace parts, custom car components, way faster 00:03:23.090 --> 00:03:25.800 and cheaper. Oh. Okay, just imagine scaling that 00:03:25.800 --> 00:03:28.699 DAPS tech. You could rapidly spin up production 00:03:28.699 --> 00:03:32.159 for custom parts for almost anything. Cars, planes, 00:03:32.360 --> 00:03:34.740 spacecraft. That software -defined flexibility 00:03:34.740 --> 00:03:37.680 is... That speed. It really is. And then there's 00:03:37.680 --> 00:03:40.219 passive logic. They pulled in 74 million in the 00:03:40.219 --> 00:03:42.620 same space. They call what they do generative 00:03:42.620 --> 00:03:45.439 autonomy. Generative autonomy. Their system stands 00:03:45.439 --> 00:03:47.520 a whole building like a commercial office building, 00:03:47.620 --> 00:03:50.919 creates this super detailed digital twin. And 00:03:50.919 --> 00:03:52.979 then the AI generates the entire control system 00:03:52.979 --> 00:03:56.740 for it. HVAC, lighting, everything. It just designs 00:03:56.740 --> 00:03:58.919 and optimizes it automatically, replacing a ton 00:03:58.919 --> 00:04:01.860 of manual engineering work. So it's complex physical 00:04:01.860 --> 00:04:04.819 system control, but done by software. Based on 00:04:04.819 --> 00:04:07.819 a scan. Yeah. It's pretty advanced. So back to 00:04:07.819 --> 00:04:10.740 figure AI for a second. If that billion dollars 00:04:10.740 --> 00:04:13.240 means investors expect mass production and deployment 00:04:13.240 --> 00:04:15.979 soon. Yeah. What does that signal about their 00:04:15.979 --> 00:04:18.600 confidence in these other pieces like DPS and 00:04:18.600 --> 00:04:20.879 pass logic, the things needed to support those 00:04:20.879 --> 00:04:23.339 robots? It tells me they think the support infrastructure 00:04:23.339 --> 00:04:26.269 problem is basically. Solve or solvable very 00:04:26.269 --> 00:04:28.350 quickly. They believe we can make the custom 00:04:28.350 --> 00:04:30.649 parts the robots need and manage the complex 00:04:30.649 --> 00:04:33.050 environments they'll work in. OK, so that covers 00:04:33.050 --> 00:04:35.629 the what's AI getting physical. But that kind 00:04:35.629 --> 00:04:37.810 of ambition, it needs some serious power behind 00:04:37.810 --> 00:04:39.629 it. Right. Which brings us to the infrastructure 00:04:39.629 --> 00:04:42.050 investments. Exactly. Theme number two. Yeah. 00:04:42.129 --> 00:04:45.290 The infrastructure, speed, security, making sure 00:04:45.290 --> 00:04:47.930 AI can be everywhere. And the big name here was 00:04:47.930 --> 00:04:51.670 Groke. G -R -O -Q. They raise a huge seven hundred 00:04:51.670 --> 00:04:54.579 and fifty million dollars. Right, Groke? They 00:04:54.579 --> 00:04:56.699 build specialized chips for AI inference. They 00:04:56.699 --> 00:04:59.500 call them LPUs, language processing units. Maybe 00:04:59.500 --> 00:05:01.540 just quickly clarify inference for everyone. 00:05:01.779 --> 00:05:04.740 Sure. Inference is basically just the AI using 00:05:04.740 --> 00:05:07.600 its training, taking its learned model and applying 00:05:07.600 --> 00:05:09.639 it to give you an answer or perform an action 00:05:09.639 --> 00:05:12.860 in real time. LPUs are specifically designed 00:05:12.860 --> 00:05:15.860 to run those big language models very, very fast. 00:05:16.199 --> 00:05:19.319 Less lag. So they're tackling the latency problem, 00:05:19.519 --> 00:05:21.920 the spinning wheel of death when you're waiting 00:05:21.920 --> 00:05:24.300 for the AI. Pretty much, yeah. Because slow AI 00:05:24.300 --> 00:05:26.220 is just a bad experience, especially if you're 00:05:26.220 --> 00:05:28.420 trying to use it for real work. Groke's tech 00:05:28.420 --> 00:05:31.560 aims for that smooth, real -time interaction 00:05:31.560 --> 00:05:34.240 you need for widespread adoption. It makes you 00:05:34.240 --> 00:05:37.139 wonder, though, if Groke really nails this instant 00:05:37.139 --> 00:05:40.100 response time for powerful AI, what does that 00:05:40.100 --> 00:05:42.019 do to the established players like the big GPU 00:05:42.019 --> 00:05:44.800 makers? Does the whole competition shift from 00:05:44.800 --> 00:05:47.990 training speed to, like... user experience speed. 00:05:48.170 --> 00:05:50.029 I think it suggests exactly that. The market 00:05:50.029 --> 00:05:51.629 seems to be saying, OK, the models are smart 00:05:51.629 --> 00:05:53.370 enough for now. Let's make them fast and cheap 00:05:53.370 --> 00:05:55.709 enough to actually use everywhere. The bottleneck 00:05:55.709 --> 00:05:58.170 has definitely moved. Yeah, that tracks. And 00:05:58.170 --> 00:06:01.269 we also saw money going into the plumbing, the 00:06:01.269 --> 00:06:03.949 system infrastructure. Upscale AI raised 100 00:06:03.949 --> 00:06:06.730 million. They're building an open standard AI 00:06:06.730 --> 00:06:09.670 networking platform. Open standard being key 00:06:09.670 --> 00:06:11.889 there. It stops companies getting locked into 00:06:11.889 --> 00:06:14.540 one vendor. It's like creating a common language. 00:06:14.699 --> 00:06:18.360 So AI systems running in the cloud or on a company's 00:06:18.360 --> 00:06:21.420 own servers or even on tiny devices out in the 00:06:21.420 --> 00:06:24.879 field can all talk to each other easily. Interoperability. 00:06:25.240 --> 00:06:27.519 Super important for big companies. And if you're 00:06:27.519 --> 00:06:29.279 connecting all these systems, you absolutely 00:06:29.279 --> 00:06:32.360 need security. We saw two big plays there. ARIA 00:06:32.360 --> 00:06:34.860 raised $100 million for general enterprise AI 00:06:34.860 --> 00:06:37.439 security. Kind of the essential tools for managing 00:06:37.439 --> 00:06:39.860 access, checking for threats across lots of different 00:06:39.860 --> 00:06:43.300 AI models. The basic blocking and tackling. Irregular 00:06:43.300 --> 00:06:45.980 raised 80 million, and they focus specifically 00:06:45.980 --> 00:06:48.879 on securing the frontier AI models. Right. The 00:06:48.879 --> 00:06:50.839 really big cutting edge ones from places like 00:06:50.839 --> 00:06:53.699 OpenAI or Google. Exactly. And they're worried 00:06:53.699 --> 00:06:56.209 about. New kinds of attacks, things like data 00:06:56.209 --> 00:06:58.370 poisoning, where someone messes up the training 00:06:58.370 --> 00:07:01.709 data. Right. Or prompt injection, which is basically 00:07:01.709 --> 00:07:04.730 tricking the AI into ignoring its safety rules 00:07:04.730 --> 00:07:06.889 and doing something it shouldn't. Yeah. I still 00:07:06.889 --> 00:07:09.389 wrestle with understanding prompt injection vulnerabilities 00:07:09.389 --> 00:07:11.810 myself sometimes. It's genuinely hard to make 00:07:11.810 --> 00:07:14.850 these flexible models consistently stable and 00:07:14.850 --> 00:07:17.569 safe. So, yeah, I appreciate companies focusing 00:07:17.569 --> 00:07:19.709 just on that defense layer. It feels really necessary. 00:07:20.129 --> 00:07:24.189 So when you add it up, Groke's speed focus. plus 00:07:24.189 --> 00:07:26.310 the security funding from area and irregular, 00:07:26.629 --> 00:07:29.089 that's over a billion dollars just on inference 00:07:29.089 --> 00:07:32.689 and safety in one week. Does that suggest that 00:07:32.689 --> 00:07:35.550 the main thing holding back wider AI adoption 00:07:35.550 --> 00:07:38.730 isn't that the models aren't smart enough, but 00:07:38.730 --> 00:07:40.889 that they're perceived as too risky or too slow? 00:07:41.360 --> 00:07:43.800 I think, yes, it feels like the intelligence 00:07:43.800 --> 00:07:46.060 is largely there, at least for many tasks. The 00:07:46.060 --> 00:07:47.860 money is now flooding in to make that intelligence 00:07:47.860 --> 00:07:50.860 fast, reliable and secure enough for businesses 00:07:50.860 --> 00:07:53.000 and people to actually trust it everywhere. That's 00:07:53.000 --> 00:07:54.500 the foundation for the physical stuff we talked 00:07:54.500 --> 00:07:56.660 about first. OK, which leads us nicely into the 00:07:56.660 --> 00:07:59.259 third major trend. automating the experts. Right. 00:07:59.360 --> 00:08:02.459 Using AI not just for general tasks, but to speed 00:08:02.459 --> 00:08:04.959 up discovery and efficiency in really complex, 00:08:05.040 --> 00:08:07.899 specialized fields. And the standout here, maybe 00:08:07.899 --> 00:08:10.920 the biggest surprise for some, was Leela Sciences. 00:08:11.579 --> 00:08:15.779 They raised a huge $235 million in a Series A. 00:08:15.819 --> 00:08:18.180 That's massive for an A round. It really is. 00:08:18.300 --> 00:08:20.079 And what they're building is, well, they call 00:08:20.079 --> 00:08:22.959 them self -driving labs, AI -powered autonomous 00:08:22.959 --> 00:08:26.139 research labs. Self -driving labs. Okay, break 00:08:26.139 --> 00:08:28.920 that down. It's like a closed loop. The AI designs 00:08:28.920 --> 00:08:31.480 an experiment. Robots physically carry it out. 00:08:31.620 --> 00:08:34.860 Then the AI analyzes the results and uses that 00:08:34.860 --> 00:08:37.620 to design the next experiment. It just runs continuously, 00:08:37.779 --> 00:08:40.659 247. Wow. So it takes the human scientists out 00:08:40.659 --> 00:08:42.639 of the loop for the day -to -day bench work. 00:08:42.840 --> 00:08:45.139 Essentially, yeah. It lets discovery happen at 00:08:45.139 --> 00:08:47.379 machine speed. You can see the potential impact 00:08:47.379 --> 00:08:49.580 on things like discovering new drugs or materials. 00:08:49.799 --> 00:08:52.139 It's potentially profound. But building a fully 00:08:52.139 --> 00:08:54.580 automated lab sounds incredibly expensive and 00:08:54.580 --> 00:08:57.179 complicated. What's the immediate hurdle they 00:08:57.179 --> 00:08:59.559 have to jump over to make that huge Series A 00:08:59.559 --> 00:09:02.019 make sense? Beyond just the pure engineering 00:09:02.019 --> 00:09:04.779 challenge, I think it's standardization. Real 00:09:04.779 --> 00:09:07.720 world labs are often messy, full of custom equipment. 00:09:08.080 --> 00:09:11.320 To make a self -driving lab work at scale, you 00:09:11.320 --> 00:09:13.080 need to standardize everything, the hardware, 00:09:13.240 --> 00:09:16.759 the processes. That's a massive task. But the 00:09:16.759 --> 00:09:19.820 payoff, if they can do it, removing that human 00:09:19.820 --> 00:09:23.220 bottleneck from discovery. Yeah, I see why investors 00:09:23.220 --> 00:09:24.980 are interested. And sticking with health, we 00:09:24.980 --> 00:09:27.700 saw conceivable life sciences get 50 million. 00:09:28.480 --> 00:09:31.120 They're automating the IVF lab using AI and robotics. 00:09:31.580 --> 00:09:34.019 Okay, so taking those complex manual steps and 00:09:34.019 --> 00:09:36.220 fertility treatments and automating them. Right. 00:09:36.279 --> 00:09:39.139 The aim is better consistency, maybe higher success 00:09:39.139 --> 00:09:41.639 rates, and making these treatments more accessible 00:09:41.639 --> 00:09:44.179 because automation could lower costs. Makes sense. 00:09:44.340 --> 00:09:47.620 And then in a totally different vertical, software 00:09:47.620 --> 00:09:49.419 development, which AI has been targeting for 00:09:49.419 --> 00:09:51.879 a while, CodeRabbit raised $60 million. Yeah, 00:09:51.980 --> 00:09:54.779 an AI code review platform. So the AI reads new 00:09:54.779 --> 00:09:57.230 code developers right. finds errors, suggests 00:09:57.230 --> 00:10:00.110 fixes, even summarizes what changed. Exactly. 00:10:00.169 --> 00:10:02.730 It speeds up the whole development process, improves 00:10:02.730 --> 00:10:05.090 the quality of the code, and hopefully frees 00:10:05.090 --> 00:10:07.129 up senior engineers to think about bigger picture 00:10:07.129 --> 00:10:08.889 problems instead of just line -by -line reviews. 00:10:09.230 --> 00:10:11.450 And there were others too, right? Like Apex raising 00:10:11.450 --> 00:10:14.389 $200 million for satellite platforms, Greenlight 00:10:14.389 --> 00:10:16.409 getting almost $50 million for construction permits. 00:10:16.649 --> 00:10:19.129 Yeah, lots of vertical -specific plays getting 00:10:19.129 --> 00:10:22.899 serious backing. Looking at this whole category, 00:10:23.120 --> 00:10:26.059 automating science labs, IVF procedures, code 00:10:26.059 --> 00:10:28.639 reviews, these are things we used to think needed 00:10:28.639 --> 00:10:31.840 deep human expertise and judgment for every step. 00:10:32.200 --> 00:10:35.059 Is this story here just replacing those experts? 00:10:35.610 --> 00:10:38.129 Or is it changing what the expert does? I think 00:10:38.129 --> 00:10:40.370 it's fundamentally changing the role. It's freeing 00:10:40.370 --> 00:10:42.549 up the experts, the scientists, the doctors, 00:10:42.649 --> 00:10:45.389 the top developers, letting them focus on strategy, 00:10:45.470 --> 00:10:47.570 on their really hard problems, on creativity. 00:10:47.929 --> 00:10:50.809 While the AI handles the more routine execution 00:10:50.809 --> 00:10:53.970 parts, it lets the humans move up the value chain. 00:10:54.230 --> 00:10:56.769 Mid -role sponsor, Reed. All right, so let's 00:10:56.769 --> 00:10:58.289 try and pull this all together. Looking back 00:10:58.289 --> 00:11:01.450 at that week in September 2025, what did these 00:11:01.450 --> 00:11:03.840 big funding rounds really tell us? I think the 00:11:03.840 --> 00:11:06.279 biggest message is that AI is moving out of the 00:11:06.279 --> 00:11:09.360 purely digital realm. It's stepping into the 00:11:09.360 --> 00:11:11.559 physical world in a serious way. Yeah. First, 00:11:11.679 --> 00:11:13.840 you saw that huge push towards commercializing 00:11:13.840 --> 00:11:16.460 physical AI, that billion for figures robots, 00:11:16.700 --> 00:11:19.159 plus the money for advanced manufacturing like 00:11:19.159 --> 00:11:22.340 DPS. It signals readiness for actual deployment. 00:11:22.720 --> 00:11:25.039 Second, you can't have physical AI without the 00:11:25.039 --> 00:11:27.559 right engine. So a massive commitment to the 00:11:27.559 --> 00:11:30.200 infrastructure, the need for speed like Grok's 00:11:30.200 --> 00:11:32.820 chips, and the absolute necessity of security 00:11:32.820 --> 00:11:35.320 and stability from companies like Irregular and 00:11:35.320 --> 00:11:39.259 Aria, making AI trustworthy at scale. And third, 00:11:39.419 --> 00:11:41.779 the money flowing into what you call closed -loop 00:11:41.779 --> 00:11:44.500 automation, systems that can run entire complex 00:11:44.500 --> 00:11:46.240 workflows themselves, whether it's discovering 00:11:46.240 --> 00:11:49.019 drugs or reviewing code, supercharging specialized 00:11:49.019 --> 00:11:51.659 fields. Yeah, the big picture takeaway feels 00:11:51.659 --> 00:11:53.919 like a clear pivot. We're moving from AI that 00:11:53.919 --> 00:11:56.779 just thinks or predicts to AI systems that operate, 00:11:56.879 --> 00:11:59.080 that build, that manage things in the tangible 00:11:59.080 --> 00:12:02.419 physical world. So that phrase from the source 00:12:02.419 --> 00:12:06.019 material about the figure AI robot, calling it 00:12:06.019 --> 00:12:08.500 the ultimate general purpose machine, if they 00:12:08.500 --> 00:12:10.840 actually achieve that, if it gets real dexterity, 00:12:10.840 --> 00:12:13.379 real autonomy, what's the final implication? 00:12:13.799 --> 00:12:16.200 Well, think about it. If that machine can truly 00:12:16.200 --> 00:12:19.669 operate anywhere a human can, What space that 00:12:19.669 --> 00:12:22.169 we've designed, a factory floor, a building site, 00:12:22.330 --> 00:12:24.750 even just a regular kitchen, won't eventually 00:12:24.750 --> 00:12:27.210 be designed with autonomous machines in mind 00:12:27.210 --> 00:12:29.509 or be occupied by them. That's a pretty profound 00:12:29.509 --> 00:12:31.710 thought. That feels like the core message behind 00:12:31.710 --> 00:12:34.210 that billion dollar signal. That's the future 00:12:34.210 --> 00:12:37.129 these massive investments are betting on a deeply 00:12:37.129 --> 00:12:40.029 physical, increasingly automated world. Thanks 00:12:40.029 --> 00:12:42.350 for joining us for this deep dive. Yeah, I hope 00:12:42.350 --> 00:12:44.110 this gives you a clearer picture of where things 00:12:44.110 --> 00:12:46.190 are heading in this incredibly fast moving space. 00:12:47.470 --> 00:12:48.230 Outro music.