WEBVTT 00:00:00.000 --> 00:00:03.500 A single recruitment decision beat. It ended 00:00:03.500 --> 00:00:07.160 a legendary billionaire friendship. It sparked 00:00:07.160 --> 00:00:10.460 a 10 -year silence. Think about that for a second. 00:00:10.599 --> 00:00:12.960 We assume the future is built on pure science. 00:00:13.339 --> 00:00:16.739 On logic. Exactly. We assume the trajectory of 00:00:16.739 --> 00:00:18.719 artificial intelligence is driven by cold, hard 00:00:18.719 --> 00:00:21.359 math and objective algorithms. But sometimes 00:00:21.359 --> 00:00:24.620 the entire timeline of human innovation hinges 00:00:24.620 --> 00:00:29.280 on, well, on one deeply personal betrayal. It 00:00:29.280 --> 00:00:31.140 is a really messy reality. I mean, you look at 00:00:31.140 --> 00:00:32.820 the neural networks and the staggering compute 00:00:32.820 --> 00:00:35.640 clusters, and it all feels so systematic. But 00:00:35.640 --> 00:00:37.859 behind the code, you just have people. You have 00:00:37.859 --> 00:00:40.640 incredibly powerful, deeply flawed individuals 00:00:40.640 --> 00:00:45.079 making decisions based on fear and pride. Welcome. 00:00:45.140 --> 00:00:47.100 You're joining us for a deep dive into the raw 00:00:47.100 --> 00:00:50.509 reality of AI today. No fluff. Not at all. We 00:00:50.509 --> 00:00:52.329 are mapping out a fascinating journey through 00:00:52.329 --> 00:00:55.710 a stack of newly unsealed documents, market data, 00:00:55.850 --> 00:00:58.270 and engineering specs. We are moving from the 00:00:58.270 --> 00:01:00.810 bruised egos and boardroom paranoia of the AI 00:01:00.810 --> 00:01:03.409 elite straight into the physical footprint of 00:01:03.409 --> 00:01:05.930 these models. Which is huge. It is. And that 00:01:05.930 --> 00:01:08.890 footprint is expanding from gas -guzzling data 00:01:08.890 --> 00:01:12.230 centers in Mississippi all the way into literal 00:01:12.230 --> 00:01:16.290 orbit. And we're wrapping up with a massive kind 00:01:16.290 --> 00:01:19.310 of quiet shift in the market. The loudest voices 00:01:19.310 --> 00:01:21.209 dominating the headlines are not actually the 00:01:21.209 --> 00:01:23.689 runs winning the enterprise war. Okay, let's 00:01:23.689 --> 00:01:25.489 unpack this. We have to start with the newly 00:01:25.489 --> 00:01:28.129 revealed legal filings from the ongoing lawsuit 00:01:28.129 --> 00:01:31.849 between Elon Musk and OpenAI. Legal discovery 00:01:31.849 --> 00:01:34.989 is always illuminating because it strips away 00:01:34.989 --> 00:01:37.390 all the PR messaging. Right. We are finally getting 00:01:37.390 --> 00:01:40.170 a raw window into their internal communications. 00:01:40.909 --> 00:01:43.189 from around 2018. I was reading through those 00:01:43.189 --> 00:01:45.829 documents and a very specific theme emerges immediately. 00:01:46.390 --> 00:01:49.170 There is this deep -seated, almost obsessive 00:01:49.170 --> 00:01:52.409 preoccupation with Demis Hassabis. The CEO of 00:01:52.409 --> 00:01:54.730 Google DeepMind. Yeah. At the time, DeepMind 00:01:54.730 --> 00:01:56.810 had just conquered the game of Go. Yeah. They 00:01:56.810 --> 00:01:58.790 were achieving technological leaps that most 00:01:58.790 --> 00:02:01.230 experts thought were decades away. And internal 00:02:01.230 --> 00:02:03.629 communications show that Musk and his inner circle 00:02:03.629 --> 00:02:06.549 viewed Hassabis as a literal threat. Like they 00:02:06.549 --> 00:02:08.909 didn't just see a business rival. No. They actively 00:02:08.909 --> 00:02:17.310 believed he endangered That level of existential 00:02:17.310 --> 00:02:20.909 dread fundamentally alters how a company operates. 00:02:21.110 --> 00:02:23.449 I mean, it shifted their entire strategy from 00:02:23.449 --> 00:02:26.409 a methodical research pace into an absolute sprint. 00:02:28.360 --> 00:02:31.139 That fear destroyed major personal relationships. 00:02:31.400 --> 00:02:33.680 Yeah, Musk actually testified about the fallout. 00:02:33.780 --> 00:02:37.219 He recognized that to counter DeepMind, he needed 00:02:37.219 --> 00:02:39.939 the absolute best talent on the planet. So he 00:02:39.939 --> 00:02:42.879 recruited a brilliant researcher named Ilya Sutskiver. 00:02:43.060 --> 00:02:45.300 Right. He poached him directly from Google. And 00:02:45.300 --> 00:02:47.719 that specific poaching incident was the breaking 00:02:47.719 --> 00:02:50.319 point. It completely ended Musk's friendship 00:02:50.319 --> 00:02:53.500 with Larry Page. Page tried desperately to keep 00:02:53.500 --> 00:02:56.539 Sutskiver. Sergey Brin tried. Hassabis tried. 00:02:56.759 --> 00:02:58.400 They threw everything at him. They really did. 00:02:58.479 --> 00:03:00.340 They threw everything at him to stop him from 00:03:00.340 --> 00:03:03.259 leaving. But Sud Skiver chose the vision at OpenAI, 00:03:03.539 --> 00:03:06.199 and the fallout was immediate. Larry Page felt 00:03:06.199 --> 00:03:08.500 completely betrayed. He stopped speaking to Musk 00:03:08.500 --> 00:03:11.259 entirely, and that silence has lasted for over 00:03:11.259 --> 00:03:14.199 a decade. Wow. Two titans of the tech industry, 00:03:14.379 --> 00:03:17.379 completely estranged over a single hire. It's 00:03:17.379 --> 00:03:20.180 like stacking Lego blocks of data, but the architects 00:03:20.180 --> 00:03:22.080 are burning down the house fighting over the 00:03:22.080 --> 00:03:25.060 instruction manual. Two sec silence. That's a 00:03:25.060 --> 00:03:27.770 brilliant way to picture it. The math is perfect, 00:03:27.949 --> 00:03:30.770 but the architects are deeply unstable. And what 00:03:30.770 --> 00:03:33.370 becomes apparent in these documents is the stark 00:03:33.370 --> 00:03:35.789 contrast in what was driving these architects. 00:03:36.050 --> 00:03:39.030 Right. Musk was entirely consumed by the threat 00:03:39.030 --> 00:03:42.009 of a Google monopoly on intelligence. He wanted 00:03:42.009 --> 00:03:44.689 to shatter that monopoly. But when you look at 00:03:44.689 --> 00:03:47.090 Sam Altman's private messages from the same period, 00:03:47.270 --> 00:03:49.310 you see a completely different psychological 00:03:49.310 --> 00:03:52.050 drive. Altman was analyzing corporate moats. 00:03:52.210 --> 00:03:55.270 His messages reveal a meticulous pursuit of industry 00:03:55.270 --> 00:03:57.449 dominance. Like he was focused on the actual 00:03:57.449 --> 00:04:00.169 structure of power. Securing compute. Yes, securing 00:04:00.169 --> 00:04:03.090 compute and positioning open AI as the gatekeeper 00:04:03.090 --> 00:04:05.710 of the next computing paradigm. And that boardroom 00:04:05.710 --> 00:04:08.310 paranoia is still dictating the company's structure 00:04:08.310 --> 00:04:12.129 today. Even in late 2023, open AI executives 00:04:12.129 --> 00:04:15.330 were sending frantic messages. Mira Marotti was 00:04:15.330 --> 00:04:18.069 emailing Microsoft leadership, actively panicking 00:04:18.069 --> 00:04:20.209 about losing their top researchers back to Hassabis. 00:04:20.449 --> 00:04:23.009 Right, and pleading for more computing resources 00:04:23.009 --> 00:04:25.829 just to keep their talent happy. The ghost of 00:04:25.829 --> 00:04:27.689 DeepMind never actually left their operations. 00:04:28.430 --> 00:04:30.850 B, I have to ask you something about this dynamic. 00:04:31.209 --> 00:04:34.009 Is the race toward artificial general intelligence 00:04:34.009 --> 00:04:36.709 actually being driven by science or just the 00:04:36.709 --> 00:04:40.279 fragile egos of billionaires? That raises a critical 00:04:40.279 --> 00:04:42.180 question about the foundation of this technology. 00:04:42.519 --> 00:04:45.600 The algorithms themselves are undeniably rigorous. 00:04:46.079 --> 00:04:49.860 But the timeline, the aggressive funding rounds, 00:04:50.019 --> 00:04:53.660 and the ruthless poaching, that is entirely driven 00:04:53.660 --> 00:04:57.040 by human flaws. It is propelled by a paranoid 00:04:57.040 --> 00:04:59.519 need to be first, regardless of the scientific 00:04:59.519 --> 00:05:02.939 readiness. So billionaire bruised egos are literally 00:05:02.939 --> 00:05:05.519 steering the future of human technology. They 00:05:05.519 --> 00:05:07.480 absolutely are. And those bruised egos demand 00:05:07.480 --> 00:05:10.220 more than just market share. They require a staggering 00:05:10.220 --> 00:05:12.170 amount of physical infrastructure. They need 00:05:12.170 --> 00:05:14.250 compute power today, which leads to drastic, 00:05:14.350 --> 00:05:16.589 often dirty physical measures. Because these 00:05:16.589 --> 00:05:18.750 billionaires cannot wait five years to build 00:05:18.750 --> 00:05:21.089 sustainable infrastructure, which brings us to 00:05:21.089 --> 00:05:23.449 the actual physical reality of this technological 00:05:23.449 --> 00:05:26.269 arms race. Right. The energy requirements for 00:05:26.269 --> 00:05:29.209 training frontier models are hitting a hard physical 00:05:29.209 --> 00:05:32.699 wall. You cannot just plug 100 ,000 GPUs into 00:05:32.699 --> 00:05:35.819 a standard city grid. Look at what Elon Musk's 00:05:35.819 --> 00:05:38.480 XAI is doing right now. They are running a massive, 00:05:38.560 --> 00:05:41.660 colossus data center down in Memphis and Mississippi. 00:05:42.000 --> 00:05:45.500 Yeah. To power it quickly, they utilized a regulatory 00:05:45.500 --> 00:05:48.220 loophole. They bypassed the years -long wait 00:05:48.220 --> 00:05:50.120 for grid interconnects by bringing in nearly 00:05:50.120 --> 00:05:53.720 50 massive gas turbines. 50 gas turbines burning 00:05:53.720 --> 00:05:56.459 fossil fuels around the clock just to push electricity 00:05:56.459 --> 00:05:58.889 into a neural network. The local pushback has 00:05:58.889 --> 00:06:01.230 been severe. Community groups are protesting 00:06:01.230 --> 00:06:04.149 because the sheer volume of emissions is drastically 00:06:04.149 --> 00:06:06.509 worsening the local air quality. Understandably. 00:06:06.589 --> 00:06:08.889 Yeah. It is a brute force engineering solution 00:06:08.889 --> 00:06:11.730 to a grid that simply cannot support the AI timeline. 00:06:11.949 --> 00:06:14.810 It exposes a massive structural bottleneck. The 00:06:14.810 --> 00:06:17.029 planet's transmission lines take a decade to 00:06:17.029 --> 00:06:19.129 upgrade, but these models are doubling in size 00:06:19.129 --> 00:06:22.089 every few months. And that bottleneck is directly 00:06:22.089 --> 00:06:24.470 fueling an explosion in specialized hardware. 00:06:25.290 --> 00:06:28.110 Companies realize that brute forcing energy isn't 00:06:28.110 --> 00:06:30.949 sustainable long term. There is a London -based 00:06:30.949 --> 00:06:34.389 startup called Fractile that just closed a $220 00:06:34.389 --> 00:06:37.230 million funding round. That is massive capital 00:06:37.230 --> 00:06:40.430 for a hardware startup trying to compete with 00:06:40.430 --> 00:06:43.290 NVIDIA. I mean, NVIDIA still dominates this entire 00:06:43.290 --> 00:06:48.089 sector with a $5 .49 trillion market cap. Fractile 00:06:48.089 --> 00:06:50.759 is taking a different approach. entirely focused 00:06:50.759 --> 00:06:54.519 on building faster AI inference hardware. They're 00:06:54.519 --> 00:06:56.819 trying to make the end user experience drastically 00:06:56.819 --> 00:06:59.160 cheaper and faster. And for you listening to 00:06:59.160 --> 00:07:00.480 this right now, when we talk about inference 00:07:00.480 --> 00:07:02.579 hardware, meaning the specific computer chips 00:07:02.579 --> 00:07:04.540 that run in AI after it's trained, that's really 00:07:04.540 --> 00:07:06.779 where the everyday battle is fought. Training 00:07:06.779 --> 00:07:08.959 a model is like sending a kid to college. Inference 00:07:08.959 --> 00:07:11.500 is that kid going to work every single day. Beautifully 00:07:11.500 --> 00:07:14.120 put. The demand for that daily computation is 00:07:14.120 --> 00:07:16.579 skyrocketing, which leads us to the ultimate, 00:07:16.660 --> 00:07:18.879 almost science fiction escalation in this arms 00:07:18.879 --> 00:07:21.050 race. I was reading the engineering proposals 00:07:21.050 --> 00:07:24.350 for this, and it is staggering. Google and SpaceX 00:07:24.350 --> 00:07:27.529 are seriously considering moving AI servers into 00:07:27.529 --> 00:07:30.350 actual orbit. It sounds absurd until you look 00:07:30.350 --> 00:07:33.189 at the thermodynamics. Musk argues that orbit 00:07:33.189 --> 00:07:35.990 is the fastest way to scale compute without fighting 00:07:35.990 --> 00:07:39.230 terrestrial power companies. Google is already 00:07:39.230 --> 00:07:42.970 planning prototype data satellites by 2027. Data 00:07:42.970 --> 00:07:45.269 centers on Earth spend roughly half their energy 00:07:45.269 --> 00:07:47.949 budget just on cooling systems. Right. In deep 00:07:47.949 --> 00:07:50.920 space. you have access to the endless ambient 00:07:50.920 --> 00:07:54.019 cold of the vacuum for heat dissipation. Furthermore, 00:07:54.319 --> 00:07:57.139 you have unmetered constant solar energy. You 00:07:57.139 --> 00:08:00.000 completely bypass the planetary power grid. Whoa! 00:08:00.480 --> 00:08:03.259 Imagine scaling to a billion queries from literal 00:08:03.259 --> 00:08:05.600 space. It completely shatters our traditional 00:08:05.600 --> 00:08:08.019 understanding of digital infrastructure. You 00:08:08.019 --> 00:08:10.420 beam the processing requests up, the orbital 00:08:10.420 --> 00:08:12.980 cluster performs the inference, and it beams 00:08:12.980 --> 00:08:15.079 the result back down. But looking at the Mississippi 00:08:15.079 --> 00:08:17.660 gas turbines and the plans for orbital servers, 00:08:17.980 --> 00:08:19.779 are we fundamentally running out of resources 00:08:19.779 --> 00:08:22.680 on Earth to sustain this arms race? If we connect 00:08:22.680 --> 00:08:25.360 this to the bigger picture, the answer is yes. 00:08:25.839 --> 00:08:29.850 Our physical reality grows linearly. We can only 00:08:29.850 --> 00:08:32.610 pour concrete and string copper wire so fast. 00:08:33.070 --> 00:08:35.909 But algorithmic demand is compounding exponentially. 00:08:36.490 --> 00:08:39.850 Space solves the thermodynamic and energy constraints 00:08:39.850 --> 00:08:42.250 of Earth, assuming the launch economics hold 00:08:42.250 --> 00:08:45.009 up. Got it. We're outgrowing the planet's power 00:08:45.009 --> 00:08:47.970 grid, forcing this wild hardware race. Precisely. 00:08:48.169 --> 00:08:50.490 And all of this massive infrastructure, from 00:08:50.490 --> 00:08:53.710 gas turbines to space servers, exists to power 00:08:53.710 --> 00:08:56.730 a new generation of tools. And these tools require 00:08:56.730 --> 00:08:59.149 an incredible amount of behavioral data to function. 00:08:59.370 --> 00:09:01.950 Because we are shifting away from simple chatbots 00:09:01.950 --> 00:09:04.429 toward autonomous agents, agents that need to 00:09:04.429 --> 00:09:06.690 know exactly how you navigate your daily workflows. 00:09:07.070 --> 00:09:09.210 Which lands this technology squarely in the regulatory 00:09:09.210 --> 00:09:11.350 crosshairs. Yeah. The data collection required 00:09:11.350 --> 00:09:13.330 to build these agents is becoming aggressively 00:09:13.330 --> 00:09:15.490 invasive. Let's examine the internal culture 00:09:15.490 --> 00:09:17.889 war happening at Meta right now. employees recently 00:09:17.889 --> 00:09:20.669 organized a widespread internal protest over 00:09:20.669 --> 00:09:23.070 new data collection policies. They discovered 00:09:23.070 --> 00:09:25.230 the company had deployed internal mouse tracking 00:09:25.230 --> 00:09:27.909 tools. They were logging the micro -movements 00:09:27.909 --> 00:09:31.389 of employee mouse cursors to train future AI 00:09:31.389 --> 00:09:34.809 agents. To build a large action model, the AI 00:09:34.809 --> 00:09:37.350 doesn't just need text. It needs to know how 00:09:37.350 --> 00:09:39.730 a human physically navigates a user interface. 00:09:40.149 --> 00:09:42.509 The employees were deeply unsettled. I mean, 00:09:42.509 --> 00:09:45.049 it represents a profound shift in corporate surveillance. 00:09:45.269 --> 00:09:48.110 An algorithm is quietly mapping your hesitation, 00:09:48.570 --> 00:09:51.710 your scrolling speed, your exact workflow patterns. 00:09:51.830 --> 00:09:53.769 It feels incredibly dystopian. You're just trying 00:09:53.769 --> 00:09:56.149 to clear your inbox. And an invisible architecture 00:09:56.149 --> 00:09:58.610 is learning how to replicate your physical inputs. 00:09:58.950 --> 00:10:02.129 Meta is facing intense friction everywhere right 00:10:02.129 --> 00:10:04.779 now. Over in Europe, they are navigating a regulatory 00:10:04.779 --> 00:10:07.519 minefield with the Digital Markets Act. They 00:10:07.519 --> 00:10:10.399 are desperately trying to dodge a potential multi 00:10:10.399 --> 00:10:13.360 -billion dollar fine from the EU regarding their 00:10:13.360 --> 00:10:15.600 closed ecosystems. Their strategy is actually 00:10:15.600 --> 00:10:18.080 quite clever. To avoid being labeled an illegal 00:10:18.080 --> 00:10:21.299 gatekeeper, Meta is temporarily opening up WhatsApp. 00:10:21.559 --> 00:10:23.980 Right. They're allowing free direct access to 00:10:23.980 --> 00:10:27.179 rival AI chatbots like ChatGPT and Claude right 00:10:27.179 --> 00:10:29.529 inside the WhatsApp interface. They are essentially 00:10:29.529 --> 00:10:32.830 turning their most guarded messaging app into 00:10:32.830 --> 00:10:36.350 an open port. It is a wild concession, but it 00:10:36.350 --> 00:10:39.309 highlights how crowded and desperate this ecosystem 00:10:39.309 --> 00:10:41.909 is getting. Everyone is fighting for space on 00:10:41.909 --> 00:10:44.970 your devices. Notion recently announced a massive 00:10:44.970 --> 00:10:47.889 platform pivot. They're no longer just a workspace. 00:10:48.190 --> 00:10:51.230 No. They're a full hub for AI agents that interact 00:10:51.230 --> 00:10:53.710 with live company databases. Then you have the 00:10:53.710 --> 00:10:56.350 explosion of ambient hardware. There is a new 00:10:56.350 --> 00:10:59.370 device called the Momokit Gem. It is a tiny all 00:10:59.370 --> 00:11:01.830 -day AI wearable. It passively captures your 00:11:01.830 --> 00:11:04.049 meetings, your phone calls, and your casual conversations. 00:11:04.009 --> 00:11:07.590 It runs localized inference to record and analyze 00:11:07.590 --> 00:11:10.029 your daily decisions. The developer tools are 00:11:10.029 --> 00:11:12.710 getting equally invasive. A platform called Latitude 00:11:12.710 --> 00:11:15.389 now traces every single session inside Claude 00:11:15.389 --> 00:11:17.870 Code. It maps exactly where developers are burning 00:11:17.870 --> 00:11:20.309 tokens so companies can audit the thought process 00:11:20.309 --> 00:11:23.070 of the AI. We are also seeing the rise of self 00:11:23.070 --> 00:11:25.649 -building tools like CraftBot. It allows a general 00:11:25.649 --> 00:11:28.850 AI agent to write, test, and evolve custom internal 00:11:28.850 --> 00:11:31.860 applications entirely on its own. Even traditional 00:11:31.860 --> 00:11:34.139 hardware is adapting to the surveillance layer. 00:11:34.559 --> 00:11:37.379 The new Google book launching this fall is built 00:11:37.379 --> 00:11:40.799 entirely around native Gemini intelligence, constantly 00:11:40.799 --> 00:11:43.259 reading screen context. The integration into 00:11:43.259 --> 00:11:45.259 our lives is becoming completely frictionless. 00:11:45.580 --> 00:11:48.120 But the reality of that friction disappearing 00:11:48.120 --> 00:11:51.200 is what worries me. I see the utility in a system 00:11:51.200 --> 00:11:53.960 that knows my preferences perfectly. But is the 00:11:53.960 --> 00:11:56.659 sheer convenience of these pervasive AI tools 00:11:56.659 --> 00:11:59.840 blinding us to the reality of constant workplace 00:11:59.840 --> 00:12:02.399 surveillance? This raises one of the most important 00:12:02.399 --> 00:12:05.279 questions of this decade. The tradeoff is absolute. 00:12:05.659 --> 00:12:09.279 We all want bespoke AI assistance. We want an 00:12:09.279 --> 00:12:11.960 agent that preemptively drafts the perfect email 00:12:11.960 --> 00:12:13.840 because it knows exactly how we communicate. 00:12:14.059 --> 00:12:16.799 Right. But to achieve that level of personalization, 00:12:17.019 --> 00:12:19.419 the AI requires total behavioral transparency. 00:12:19.779 --> 00:12:22.200 It has to watch you work. We are voluntarily 00:12:22.200 --> 00:12:24.779 trading our behavioral autonomy just to save 00:12:24.779 --> 00:12:27.019 a few keystrokes a day. Right. We're normalizing 00:12:27.019 --> 00:12:29.299 intense surveillance just to build slightly smarter 00:12:29.299 --> 00:12:31.759 digital assistants. And corporate adoption is 00:12:31.759 --> 00:12:34.000 the primary engine forcing this normalization. 00:12:34.419 --> 00:12:36.379 Businesses are eager to deploy these agents. 00:12:36.639 --> 00:12:38.919 Which brings us to the actual market reality. 00:12:39.279 --> 00:12:41.700 We're going to dive into exactly which businesses 00:12:41.700 --> 00:12:44.179 are adopting what right after this quick break. 00:12:44.559 --> 00:12:48.059 Mid -roll sponsor, Reed Placeholder. And we are 00:12:48.059 --> 00:12:51.059 back. We have spent the first half of this deep 00:12:51.059 --> 00:12:53.820 dive looking at the noisy, headline -grabbing 00:12:53.820 --> 00:12:57.799 chaos. We explored Elon Musk's paranoia, Meta's 00:12:57.799 --> 00:13:01.039 internal tracking protests, and OpenAI's boardroom 00:13:01.039 --> 00:13:03.840 drama. It is very easy to get distracted by the 00:13:03.840 --> 00:13:07.220 billionaire drama. It is loud by design. But 00:13:07.220 --> 00:13:08.860 when you strip away the noise and look at the 00:13:08.860 --> 00:13:11.250 actual enterprise deployment data... A completely 00:13:11.250 --> 00:13:13.750 different narrative emerges. Someone else is 00:13:13.750 --> 00:13:16.210 quietly winning the business war. The telemetry 00:13:16.210 --> 00:13:18.289 data tells a story of hyper -focused execution. 00:13:18.710 --> 00:13:22.509 We just received the fresh May 2026 enterprise 00:13:22.509 --> 00:13:25.909 software data from Ramp, and it reveals a massive 00:13:25.909 --> 00:13:28.549 fundamental shift in the market hierarchy. Anthropic 00:13:28.549 --> 00:13:31.250 has officially surpassed OpenAI in verified business 00:13:31.250 --> 00:13:34.009 customers. That is a staggering milestone. The 00:13:34.009 --> 00:13:37.509 data shows that 34 .4 % of businesses are now 00:13:37.509 --> 00:13:40.240 paying for Anthropic's enterprise tier. OpenAI 00:13:40.240 --> 00:13:44.000 has steadily dropped down to 32 .3%. OpenAI certainly 00:13:44.000 --> 00:13:46.779 maintains a volume lead in broader non -technical 00:13:46.779 --> 00:13:49.720 consumer applications. Sure. But Anthropic has 00:13:49.720 --> 00:13:52.299 captured the high -value, high -adoption enterprise 00:13:52.299 --> 00:13:55.799 sectors. They have completely locked down finance, 00:13:56.100 --> 00:13:58.960 technology, and professional services. And if 00:13:58.960 --> 00:14:00.659 you look at developer platforms like the open 00:14:00.659 --> 00:14:03.679 router Leaderboard, Anthropic has consistently 00:14:03.679 --> 00:14:08.399 ranked above OpenAI since December of 2025. Ramp 00:14:08.399 --> 00:14:11.360 economists attribute this victory to a very deliberate 00:14:11.360 --> 00:14:14.600 technical first strategy. While OpenAI was building 00:14:14.600 --> 00:14:17.159 consumer -friendly voice modes and image generators, 00:14:17.519 --> 00:14:20.399 Anthropic ignored the broad consumer market. 00:14:20.539 --> 00:14:23.039 They focused entirely on developers and specialized 00:14:23.039 --> 00:14:25.559 technical professionals. And you can see the 00:14:25.559 --> 00:14:28.059 results in their feature rollouts. Look at Claude 00:14:28.059 --> 00:14:31.139 Code. They just upgraded it with a native multi 00:14:31.139 --> 00:14:33.639 -agent view. Which is a massive paradigm shift 00:14:33.639 --> 00:14:35.700 for software engineering. Yeah. Instead of a 00:14:35.700 --> 00:14:38.299 developer prompting one AI to write code, the 00:14:38.299 --> 00:14:40.820 multi -agent view allows an entire fleet of specialized 00:14:40.820 --> 00:14:43.500 agents to check each other's work, run tests, 00:14:43.700 --> 00:14:45.840 and debug in a continuous loop. Professional 00:14:45.840 --> 00:14:48.480 services are deploying tools like Cloud Cowork 00:14:48.480 --> 00:14:51.320 for incredibly dense, complex document analysis. 00:14:51.720 --> 00:14:54.200 And in the finance sector, the demand for precision 00:14:54.200 --> 00:14:57.350 is absolute. Finance has a massive demand for 00:14:57.350 --> 00:15:00.490 a large context window size. It means exactly 00:15:00.490 --> 00:15:03.549 how much text the AI can remember at once. If 00:15:03.549 --> 00:15:06.470 a financial analyst feeds a 500 -page corporate 00:15:06.470 --> 00:15:09.970 prospectus into a model, the AI cannot drop the 00:15:09.970 --> 00:15:12.809 thread. No. It cannot hallucinate a single digit 00:15:12.809 --> 00:15:15.590 on page 400 because it forgot the parameters 00:15:15.590 --> 00:15:17.929 set on page one. And I have to admit, I still 00:15:17.929 --> 00:15:20.159 wrestle with prompt drift myself. I'll start 00:15:20.159 --> 00:15:22.419 a long, complex research conversation with the 00:15:22.419 --> 00:15:25.299 chatbot, and an hour later, it completely forgets 00:15:25.299 --> 00:15:27.539 my initial formatting instructions. It starts 00:15:27.539 --> 00:15:30.159 hallucinating broad summaries. It is the most 00:15:30.159 --> 00:15:32.759 frustrating bottleneck in consumer AI right now. 00:15:32.919 --> 00:15:36.179 The models degrade over long sessions. It completely 00:15:36.179 --> 00:15:38.679 ruins professional workflows. So it makes total 00:15:38.679 --> 00:15:41.220 sense why enterprise users are abandoning general 00:15:41.220 --> 00:15:43.659 chatbots. They are migrating to Anthropic because 00:15:43.659 --> 00:15:45.700 the structural integrity of the long context 00:15:45.700 --> 00:15:47.840 memory is simply superior. Corporate clients 00:15:47.840 --> 00:15:57.230 do not care about a model. Does this mean the 00:15:57.230 --> 00:15:59.730 era of the general purpose chatbot is already 00:15:59.730 --> 00:16:02.330 dead in the corporate world? I believe that phase 00:16:02.330 --> 00:16:05.649 of the market has permanently closed. Businesses 00:16:05.649 --> 00:16:08.629 demand precision execution. They have no use 00:16:08.629 --> 00:16:10.690 for conversational parlor tricks in a boardroom. 00:16:10.730 --> 00:16:13.610 If a model cannot reliably parse a massive legal 00:16:13.610 --> 00:16:17.129 contract or autonomously debug thousands of lines 00:16:17.129 --> 00:16:19.990 of legacy code without supervision, it offers 00:16:19.990 --> 00:16:23.330 zero enterprise value. They need focused utility, 00:16:23.590 --> 00:16:26.840 not a generic talking companion. Basically, businesses 00:16:26.840 --> 00:16:30.139 want a specialized execution tool, not a jack 00:16:30.139 --> 00:16:32.580 -of -all -trades chatbot. Exactly. The enterprise 00:16:32.580 --> 00:16:35.039 market has finally matured past the novelty phase. 00:16:35.200 --> 00:16:37.000 This has been a wild journey through the stack 00:16:37.000 --> 00:16:39.379 today. We covered a tremendous amount of ground, 00:16:39.480 --> 00:16:42.059 from orbital cooling systems to pixel -level 00:16:42.059 --> 00:16:44.039 mouse tracking. If we connect this to the bigger 00:16:44.039 --> 00:16:46.840 picture, a profound irony emerges from all this 00:16:46.840 --> 00:16:49.480 data. The loudest voices in this industry, the 00:16:49.480 --> 00:16:52.340 pioneers like Musk and Altman. are fighting bitterly 00:16:52.340 --> 00:16:54.679 over theoretical power. They're burning bridges, 00:16:54.879 --> 00:16:57.779 building 50 -turbine power plants, and planning 00:16:57.779 --> 00:17:00.480 servers in space to win a perceived monopoly. 00:17:00.919 --> 00:17:04.099 But while they wage this loud, chaotic war fueled 00:17:04.099 --> 00:17:07.700 by ego, they're being quietly outflanked. They're 00:17:07.700 --> 00:17:10.380 losing the actual, highly lucrative business 00:17:10.380 --> 00:17:13.519 sector to Anthropic's quiet, hyper -focused, 00:17:13.619 --> 00:17:16.500 drama -free execution. The enterprise market 00:17:16.500 --> 00:17:20.309 ultimately rewards utility, not paranoia. So 00:17:20.309 --> 00:17:23.289 what does this all mean, Beat? Think about this 00:17:23.289 --> 00:17:25.490 as you go about your week. If the foundational 00:17:25.490 --> 00:17:28.269 pioneers of this technology are so deeply driven 00:17:28.269 --> 00:17:31.430 by paranoia, ego, and the pursuit of power that 00:17:31.430 --> 00:17:33.210 they will stop speaking to their best friends 00:17:33.210 --> 00:17:36.130 for a decade, what inherent human biases are 00:17:36.130 --> 00:17:37.950 being permanently baked into the foundations 00:17:37.950 --> 00:17:40.170 of the objective AI agents we will soon rely 00:17:40.170 --> 00:17:42.410 on for every decision we make? There's a deeply 00:17:42.410 --> 00:17:44.289 unsettling thought to leave on. Something to 00:17:44.289 --> 00:17:46.170 mull over as these agents integrate into your 00:17:46.170 --> 00:17:47.950 daily life. Thank you for joining us on this 00:17:47.950 --> 00:17:50.690 deep dive. We will see you next time. Outro music.