WEBVTT 00:00:00.000 --> 00:00:02.500 We are building a trillion -dollar future right 00:00:02.500 --> 00:00:05.219 now. We are building it on the backs of a few 00:00:05.219 --> 00:00:10.320 AI companies. Beat. But what happens if those 00:00:10.320 --> 00:00:13.099 foundational pillars suddenly collapse? The whole 00:00:13.099 --> 00:00:15.759 system shakes violently. It's a massive structural 00:00:15.759 --> 00:00:18.260 vulnerability for the global economy. Welcome 00:00:18.260 --> 00:00:20.239 to the Deep Dive. I'm very glad you're here with 00:00:20.239 --> 00:00:22.800 us today. Let's unpack this highly complex landscape 00:00:22.800 --> 00:00:25.079 together. We have some truly fascinating ground 00:00:25.079 --> 00:00:27.739 to cover today. First, we're analyzing the $900 00:00:27.739 --> 00:00:30.879 billion house of cards in AI infrastructure. 00:00:31.440 --> 00:00:34.280 Then we track the fierce platform war happening 00:00:34.280 --> 00:00:37.240 between anthropic and open AI. Right. We'll also 00:00:37.240 --> 00:00:39.920 explore the sudden shift to off -grid local AI 00:00:39.920 --> 00:00:42.700 video. Yeah. And finally, we look at... how AI 00:00:42.700 --> 00:00:45.579 is actively rewiring our jobs, our dating lives, 00:00:45.840 --> 00:00:47.899 and even our pets' healthcare. I have to offer 00:00:47.899 --> 00:00:49.859 a vulnerable admission right up front, though. 00:00:49.960 --> 00:00:52.520 I still wrestle with prompt drift myself. Oh, 00:00:52.539 --> 00:00:54.799 sure. Prompt drift is when the same prompt suddenly 00:00:54.799 --> 00:00:56.700 gives you different answers over time. Yeah, 00:00:56.740 --> 00:00:59.380 it happens constantly. So trying to wrap my head 00:00:59.380 --> 00:01:02.420 around a $900 billion infrastructure gamble feels 00:01:02.420 --> 00:01:05.239 completely surreal. The sheer scale of it breaks 00:01:05.239 --> 00:01:07.920 our standard economic models. Let's start with 00:01:07.920 --> 00:01:10.569 that massive macro picture. We're looking at 00:01:10.569 --> 00:01:12.790 the foundational hardware driving this entire 00:01:12.790 --> 00:01:15.450 boom. We're looking at severe systemic risk. 00:01:15.650 --> 00:01:18.409 A recent Reuters report highlights a very serious 00:01:18.409 --> 00:01:21.909 and growing concern. Major labs like OpenAI and 00:01:21.909 --> 00:01:24.950 Anthropic might be too big to fail. That specific 00:01:24.950 --> 00:01:27.909 phrase carries a lot of dark historical weight. 00:01:28.049 --> 00:01:30.189 We all remember what that meant for the big banks. 00:01:30.620 --> 00:01:33.500 It really does. Today, these few companies are 00:01:33.500 --> 00:01:35.920 the main buyers driving the global compute surge. 00:01:36.239 --> 00:01:39.579 Tech giants are spending about $650 billion this 00:01:39.579 --> 00:01:42.159 year alone. That's just this year alone. It is 00:01:42.159 --> 00:01:44.540 a staggering amount of capital flowing into one 00:01:44.540 --> 00:01:47.500 single sector. Exactly. And much of that massive 00:01:47.500 --> 00:01:50.019 physical expansion relies on a huge assumption. 00:01:50.299 --> 00:01:53.060 It assumes these AI labs will keep demanding 00:01:53.060 --> 00:01:56.060 massive computing power forever. It's like stacking 00:01:56.060 --> 00:01:58.959 Lego blocks of data or building millions of gas 00:01:58.959 --> 00:02:01.540 stations for a... very specific car, but... But 00:02:01.540 --> 00:02:03.540 the car manufacturer suddenly goes bankrupt, 00:02:03.739 --> 00:02:05.180 right? Exactly. You don't just lose the car. 00:02:05.340 --> 00:02:07.739 You have billions of dollars in useless concrete 00:02:07.739 --> 00:02:10.340 and cables. Yeah, and they fall hard. If a major 00:02:10.340 --> 00:02:13.319 lab fails, that massive compute demand disappears 00:02:13.319 --> 00:02:16.580 instantly. And then the financial dominoes start 00:02:16.580 --> 00:02:19.219 falling everywhere. The impact spreads outward 00:02:19.219 --> 00:02:21.680 through the entire market. It cascades completely 00:02:21.680 --> 00:02:24.120 through the global supply chain. You lose the 00:02:24.120 --> 00:02:27.000 constant need for both AI training and inference. 00:02:27.659 --> 00:02:29.939 And inference is running a trained model to make 00:02:29.939 --> 00:02:32.460 live predictions or decisions. Right. Without 00:02:32.460 --> 00:02:35.280 that constant load, the explosive growth of data 00:02:35.280 --> 00:02:38.240 centers could slow sharply. Cloud providers and 00:02:38.240 --> 00:02:40.439 chip makers would suddenly struggle. They simply 00:02:40.439 --> 00:02:42.599 wouldn't find enough new customers. And this 00:02:42.599 --> 00:02:45.419 massive risk touches completely normal financial 00:02:45.419 --> 00:02:47.939 institutions, too. We're talking about traditional 00:02:47.939 --> 00:02:51.039 cornerstone finance. Banks and private lenders 00:02:51.039 --> 00:02:55.860 have financed about $900 billion in AI infrastructure. 00:02:56.650 --> 00:02:58.509 That is almost a trillion dollars in combined 00:02:58.509 --> 00:03:02.650 debt and equity. Cusack silence. So what exactly 00:03:02.650 --> 00:03:04.849 triggers that collapse? Do servers just power 00:03:04.849 --> 00:03:07.469 down overnight? No, it's a sudden drop in market 00:03:07.469 --> 00:03:09.569 confidence. The financial markets would freeze 00:03:09.569 --> 00:03:12.330 long before any server actually turns off. Cloud 00:03:12.330 --> 00:03:14.949 providers would panic instantly. So market confidence 00:03:14.949 --> 00:03:17.710 collapses long before the actual technology does. 00:03:17.789 --> 00:03:19.830 Let's transition to the direct head -to -head 00:03:19.830 --> 00:03:23.110 battle here. If the global economy rests on these 00:03:23.110 --> 00:03:26.189 massive labs, which one is actually winning? 00:03:26.650 --> 00:03:30.509 This brings us to RAMP's March 2026 AI Index. 00:03:30.789 --> 00:03:33.330 It tracks how companies actually spend their 00:03:33.330 --> 00:03:35.990 real money. This is actual corporate credit card 00:03:35.990 --> 00:03:39.090 data. It's not just surveys or optimistic boardroom 00:03:39.090 --> 00:03:41.770 projections. The data reveals a fascinating trend. 00:03:42.419 --> 00:03:44.759 Anthropic is pulling far ahead of the competition. 00:03:45.219 --> 00:03:48.020 They're winning 70 % of head -to -head deals 00:03:48.020 --> 00:03:51.719 against OpenAI for new businesses. That is a 00:03:51.719 --> 00:03:54.060 massive shift in market dominance. It gets even 00:03:54.060 --> 00:03:56.020 crazier when you look at broad usage. Nearly 00:03:56.020 --> 00:03:59.360 one in four businesses on ramp now pays for Anthropic. 00:03:59.500 --> 00:04:02.139 Just a year ago, it was only one in 25. That's 00:04:02.139 --> 00:04:04.319 truly explosive growth. Meanwhile, what is the 00:04:04.319 --> 00:04:07.599 actual corporate data showing for OpenAI? OpenAI's 00:04:07.599 --> 00:04:10.680 adoption rate actually fell 1 .5 % in March. 00:04:10.860 --> 00:04:12.819 That represents their biggest single -month drop 00:04:12.819 --> 00:04:14.819 so far. I have to push back on this logically. 00:04:15.000 --> 00:04:17.500 Why on earth would businesses flock to a platform 00:04:17.500 --> 00:04:20.319 that actively throttles their usage over a cheaper, 00:04:20.439 --> 00:04:23.540 widely available one? It's deeply counterintuitive 00:04:23.540 --> 00:04:25.660 at first glance. The core developer products 00:04:25.660 --> 00:04:28.600 are actually very close in performance. Cloud 00:04:28.600 --> 00:04:31.240 Code and OpenAI Codex are broadly comparable 00:04:31.240 --> 00:04:33.920 coding tools. Codex might even be cheaper with 00:04:33.920 --> 00:04:36.220 higher limits. You'd literally get more output 00:04:36.220 --> 00:04:39.019 for less money. Right. And Anthropic has a very 00:04:39.019 --> 00:04:41.800 strange operational constraint right now. They 00:04:41.800 --> 00:04:44.199 cannot fully meet their current enterprise demand. 00:04:45.230 --> 00:04:47.709 Many plans still have strict rate limits attached. 00:04:48.050 --> 00:04:50.230 In almost every software market in history, the 00:04:50.230 --> 00:04:52.490 cheaper product usually wins. You normally just 00:04:52.490 --> 00:04:55.149 choose the frictionless available option. But 00:04:55.149 --> 00:04:58.029 here, that normal economic pattern is breaking 00:04:58.029 --> 00:05:01.730 entirely. Choosing an AI is becoming a powerful 00:05:01.730 --> 00:05:04.689 cultural identity marker. Like the green bubble 00:05:04.689 --> 00:05:07.670 versus blue bubble divide an iMessage. It signals 00:05:07.670 --> 00:05:10.269 exactly who you are to the world. Exactly. And 00:05:10.269 --> 00:05:13.269 Topic uses constitutional AI to guide its models. 00:05:13.649 --> 00:05:15.769 Which means an AI... trained to follow a strict 00:05:15.769 --> 00:05:18.250 set of built -in rules. Right. That specific 00:05:18.250 --> 00:05:20.490 approach gives enterprise boards a deep sense 00:05:20.490 --> 00:05:23.370 of safety. They want predictable behavior over 00:05:23.370 --> 00:05:26.250 raw processing power. It signals a responsible 00:05:26.250 --> 00:05:29.269 corporate identity to their shareholders. But 00:05:29.269 --> 00:05:31.769 if it's just about brand safety, what does this 00:05:31.769 --> 00:05:34.339 say about tech loyalty right now? It tells us 00:05:34.339 --> 00:05:36.860 that businesses are choosing based on brand alignment. 00:05:37.040 --> 00:05:39.240 They are picking a philosophical side in the 00:05:39.240 --> 00:05:42.420 AI race, not just optimizing for raw volume. 00:05:42.660 --> 00:05:44.740 It's not just utility anymore. It's a corporate 00:05:44.740 --> 00:05:47.819 identity. So to escape those bottlenecks, rate 00:05:47.819 --> 00:05:50.480 limits, and systemic risks, the industry is pushing 00:05:50.480 --> 00:05:53.300 back. We're seeing a huge shift toward decentralization. 00:05:53.680 --> 00:05:56.339 Developers are pushing high -end AI directly 00:05:56.339 --> 00:05:58.800 to personal hardware. This is where the launch 00:05:58.800 --> 00:06:01.959 of LTX Studio 2 .3 comes in. Until now, the most 00:06:01.959 --> 00:06:04.860 powerful video tools lived entirely in the cloud. 00:06:05.120 --> 00:06:08.500 Sora, Runway, Kling AI, and Seedance all require 00:06:08.500 --> 00:06:11.480 remote servers. You essentially rented time on 00:06:11.480 --> 00:06:13.420 their massive corporate computers. You paid a 00:06:13.420 --> 00:06:16.930 constant subscription tax. But LTX 2 .3 runs 00:06:16.930 --> 00:06:18.990 directly on your personal hardware. It processes 00:06:18.990 --> 00:06:21.110 everything locally using your own machine. And 00:06:21.110 --> 00:06:23.189 the technical specs are absolutely incredible 00:06:23.189 --> 00:06:25.189 here. They really are pushing the boundaries. 00:06:25.389 --> 00:06:27.930 It generates video up to true 4K resolution. 00:06:28.209 --> 00:06:31.389 It runs at a buttery smooth 50 frames per second. 00:06:31.569 --> 00:06:34.110 You can generate complex clips up to 20 seconds 00:06:34.110 --> 00:06:37.290 long. That is serious runtime for heavy generative 00:06:37.290 --> 00:06:39.790 video. Yeah, and it includes full audio generation 00:06:39.790 --> 00:06:42.550 baked right in. Plus, it outputs native vertical 00:06:42.550 --> 00:06:45.870 video automatically. So that's 1080 by 1920 pixels, 00:06:46.050 --> 00:06:49.529 perfectly formatted for TikTok or Reels. Beat. 00:06:49.709 --> 00:06:53.319 Whoa. Imagine generating a Hollywood -grade 4K 00:06:53.319 --> 00:06:56.339 video entirely off the grid. It changes the entire 00:06:56.339 --> 00:06:58.620 game for independent creators. Look at their 00:06:58.620 --> 00:07:02.139 previous open -source model, LTX2. It hit 4 million 00:07:02.139 --> 00:07:05.300 downloads in just six weeks. That is staggering. 00:07:05.540 --> 00:07:08.160 The pent -up demand for local processing is clearly 00:07:08.160 --> 00:07:10.420 there. What does this mean for actual creative 00:07:10.420 --> 00:07:12.660 freedom? It means no subscription paywalls block 00:07:12.660 --> 00:07:14.720 your workflow. There are no external corporate 00:07:14.720 --> 00:07:17.420 content filters dictating your art. Your rough 00:07:17.420 --> 00:07:19.439 drafts remain completely private on your own 00:07:19.439 --> 00:07:22.410 hard drive. So we bypass the cloud tax and own 00:07:22.410 --> 00:07:24.790 our creations locally. Exactly. Whether running 00:07:24.790 --> 00:07:27.389 on massive cloud data centers or local laptops, 00:07:27.709 --> 00:07:30.389 this technology is already fundamentally rewiring 00:07:30.389 --> 00:07:33.649 human life. It's changing labor and love. And 00:07:33.649 --> 00:07:35.709 it's changing medicine. Look at the dating scene, 00:07:35.750 --> 00:07:38.269 for example. A Bumble recently launched their 00:07:38.269 --> 00:07:41.370 new Bee Assistant. It writes your profile and 00:07:41.370 --> 00:07:44.290 actually plans your physical dates. And Tinder 00:07:44.290 --> 00:07:48.360 just unveiled AI speed dating. It is a very strange 00:07:48.360 --> 00:07:51.420 concept to process emotionally. Soon, two AI 00:07:51.420 --> 00:07:54.060 agents might date each other first. They'll verify 00:07:54.060 --> 00:07:56.360 compatibility before the actual human beings 00:07:56.360 --> 00:07:59.360 ever meet. If our AI twins are going on dates 00:07:59.360 --> 00:08:01.579 with each other and applying for jobs with other 00:08:01.579 --> 00:08:04.399 AI recruiters, what is the human actually supposed 00:08:04.399 --> 00:08:07.160 to be doing? That's the defining philosophical 00:08:07.160 --> 00:08:09.899 question right now. Automation removes the friction, 00:08:10.079 --> 00:08:12.600 but humans shift to being the visionaries. We 00:08:12.600 --> 00:08:14.939 must still provide the intent. We stop doing 00:08:14.939 --> 00:08:16.660 the busy work and start directing the actual 00:08:16.660 --> 00:08:19.759 vision. And humans are still using AI for deeply 00:08:19.759 --> 00:08:22.139 personal things. Medicine is a perfect example. 00:08:22.459 --> 00:08:24.540 A tech founder's dog was diagnosed with severe 00:08:24.540 --> 00:08:26.920 cancer. Standard chemotherapy had completely 00:08:26.920 --> 00:08:29.220 failed to stop it. So he turned to the advanced 00:08:29.220 --> 00:08:32.320 tools he understood. He used ChatGPT and AlphaFold. 00:08:32.460 --> 00:08:35.019 And AlphaFold is an AI predicting the complex 00:08:35.019 --> 00:08:38.500 3D structure of microscopic proteins. He used 00:08:38.500 --> 00:08:40.820 those tools to design an experimental cancer 00:08:40.820 --> 00:08:43.639 vaccine. He custom built it specifically for 00:08:43.639 --> 00:08:46.100 his own dog. And it actually worked in real life. 00:08:46.279 --> 00:08:48.759 One of the aggressive tumors shrank after the 00:08:48.759 --> 00:08:51.980 treatment. That is incredible. The power of a 00:08:51.980 --> 00:08:54.440 massive research lab moved into the hands of 00:08:54.440 --> 00:08:57.129 one individual. But while individuals are empowered, 00:08:57.389 --> 00:09:00.110 the massive corporate scale continues expanding, 00:09:00.210 --> 00:09:03.830 too. Elon Musk just teased Tesla's TeraFab AI 00:09:03.830 --> 00:09:07.590 chip factory launch on March 21st. That's a $20 00:09:07.590 --> 00:09:10.470 billion project to control their own massive 00:09:10.470 --> 00:09:13.740 AI stack. Axiom also just raised $200 million 00:09:13.740 --> 00:09:17.460 at a $1 .6 billion valuation. Their mission is 00:09:17.460 --> 00:09:20.059 to build AI that formally proves math. Meaning 00:09:20.059 --> 00:09:22.200 mathematically proving that a computer algorithm 00:09:22.200 --> 00:09:24.639 is completely correct. Right. We're also seeing 00:09:24.639 --> 00:09:27.679 an explosion of daily rapid tools. Eleven Creative 00:09:27.679 --> 00:09:30.159 lets you generate and edit premium audio, video, 00:09:30.200 --> 00:09:32.600 and images all in one place. Then you have the 00:09:32.600 --> 00:09:35.639 Google Workspace CLI. It gives you over 100 dedicated 00:09:35.639 --> 00:09:37.909 agents for Drive and Gmail. The Banana app is 00:09:37.909 --> 00:09:40.149 another fascinating breakthrough. It offers real 00:09:40.149 --> 00:09:42.190 -time voice translation in over 80 different 00:09:42.190 --> 00:09:44.450 languages. And it preserves your own voice without 00:09:44.450 --> 00:09:47.230 needing subtitles. Yeah, and Motion is a new 00:09:47.230 --> 00:09:49.429 tool that upgrades screen capture. It records 00:09:49.429 --> 00:09:51.950 in high -res, smooths out your frantic mouse 00:09:51.950 --> 00:09:55.250 movements, and adds cinematic zoom. These tools 00:09:55.250 --> 00:09:57.590 are making everyone a one -person production 00:09:57.590 --> 00:10:00.149 studio. But let's look at the actual workplace. 00:10:00.720 --> 00:10:04.700 The site jobloss .ai tracks corporate layoffs 00:10:04.700 --> 00:10:08.039 directly blamed on AI. It logs companies replacing 00:10:08.039 --> 00:10:11.779 human workers with automated systems. 75 % of 00:10:11.779 --> 00:10:13.879 resumes never reach a human eye anymore. And 00:10:13.879 --> 00:10:16.840 then there's the Amazon workload paradox. An 00:10:16.840 --> 00:10:18.519 independent study confirmed that introducing 00:10:18.519 --> 00:10:21.679 automation actually increases daily human workloads. 00:10:22.460 --> 00:10:24.519 Why does automation actually increase our daily 00:10:24.519 --> 00:10:26.899 workload instead of reducing it? Because capitalism 00:10:26.899 --> 00:10:29.879 hates a vacuum. The moment a complex task gets 00:10:29.879 --> 00:10:32.379 faster, your baseline expectations simply shifts 00:10:32.379 --> 00:10:34.860 upward. You don't gain free time. You just process 00:10:34.860 --> 00:10:36.960 a much higher volume of work. Automation just 00:10:36.960 --> 00:10:39.019 creates an entirely new category of busy work. 00:10:39.159 --> 00:10:42.259 We're navigating a truly bizarre historical paradox 00:10:42.259 --> 00:10:44.759 right now. On one hand, we are building a $900 00:10:44.759 --> 00:10:47.639 billion house of cards. The entire system relies 00:10:47.639 --> 00:10:50.120 heavily on a few tech giants battling for blue 00:10:50.120 --> 00:10:52.960 bubble status. Right. Yet simultaneously, this 00:10:52.960 --> 00:10:55.340 exact same underlying technology is becoming 00:10:55.340 --> 00:10:58.559 intensely personal. It's capable of running locally 00:10:58.559 --> 00:11:01.659 on our laptops to render 4K video off the grid. 00:11:01.860 --> 00:11:04.299 It's seamlessly translating our actual voices 00:11:04.299 --> 00:11:06.440 in 80 different languages. It's scheduling our 00:11:06.440 --> 00:11:08.799 romantic dates for us. And it's even stepping 00:11:08.799 --> 00:11:11.019 in to save our pets when traditional medicine 00:11:11.019 --> 00:11:13.669 completely fails. Thank you for joining us for 00:11:13.669 --> 00:11:15.450 this deep dive. I want to leave you with one 00:11:15.450 --> 00:11:18.830 final thought to mull over. If AI platforms are 00:11:18.830 --> 00:11:21.009 becoming identity markers like smartphone ecosystems, 00:11:21.629 --> 00:11:24.549 how will your choice of AI today define your 00:11:24.549 --> 00:11:27.009 personal or professional brand in five years?