WEBVTT 00:00:02.240 --> 00:00:04.799 Welcome to the AI Chat Podcast. I'm your host, 00:00:04.860 --> 00:00:07.179 Jaden Schaefer. Every day I cover cutting -edge 00:00:07.179 --> 00:00:09.320 AI news and talk with the leaders behind it, 00:00:09.339 --> 00:00:11.320 breaking down what it means for your life and 00:00:11.320 --> 00:00:18.960 business. Arena AI has just hit a $100 million 00:00:18.960 --> 00:00:21.239 run rate. This is only eight months after they 00:00:21.239 --> 00:00:23.440 launched paid evaluations. I want to break down 00:00:23.440 --> 00:00:25.359 what Arena does, why they're special, why they're 00:00:25.359 --> 00:00:28.239 unique, and why they're growing so fast. Palantir 00:00:28.239 --> 00:00:31.579 is tapping NVIDIA's Nemotron open models for 00:00:31.579 --> 00:00:34.659 US government AI. This is interesting and a lot 00:00:34.659 --> 00:00:36.990 of drama is behind this story as well. Elizabeth 00:00:36.990 --> 00:00:40.590 Warren and Scanlon are reviving a bill to ban 00:00:40.590 --> 00:00:43.250 AI firms from selling health data. We'll get 00:00:43.250 --> 00:00:46.270 into the details on that. Flexion Robotics is 00:00:46.270 --> 00:00:49.310 training hundreds of humanoids to run office 00:00:49.310 --> 00:00:53.630 errands autonomously. And China's CXMT is landing 00:00:53.630 --> 00:00:57.210 a $3 billion memory supply deal with Tencent. 00:00:57.210 --> 00:00:59.850 We know memory is one of the critical pieces 00:00:59.850 --> 00:01:02.420 of the AI. infrastructure build out. And there's 00:01:02.420 --> 00:01:04.980 a lot of issues going on with memory, increasing 00:01:04.980 --> 00:01:07.359 the costs of basically all electronics today. 00:01:07.480 --> 00:01:09.700 If you've ever been using an AI model like Claude 00:01:09.700 --> 00:01:11.379 and have been frustrated that it doesn't create 00:01:11.379 --> 00:01:14.359 images or audio or video, I'd love for you to 00:01:14.359 --> 00:01:17.060 try the MCP connector for AI box. That's my own 00:01:17.060 --> 00:01:19.420 startup. We essentially allow you with one link, 00:01:19.480 --> 00:01:21.700 give it to Claude, and it can bring any of the 00:01:21.700 --> 00:01:24.340 AI models that you use for everything else into 00:01:24.340 --> 00:01:26.680 Claude. So you can use chat GPT's image generation 00:01:26.680 --> 00:01:29.859 inside of Claude. You can use 11 labs audio generation. 00:01:30.219 --> 00:01:33.239 of Claude or Google VO3's video generation inside 00:01:33.239 --> 00:01:35.219 of Claude. So all of the capabilities of the 00:01:35.219 --> 00:01:37.040 different AI models, there's 80 different ones 00:01:37.040 --> 00:01:38.920 that we allow you to connect. You can bring them 00:01:38.920 --> 00:01:41.659 all inside of Claude or ChatGPT or Gemini or 00:01:41.659 --> 00:01:43.980 Cursor or any of your other places where you 00:01:43.980 --> 00:01:46.060 really do all of your work, all of your workspaces 00:01:46.060 --> 00:01:48.200 with AI. So if you want to check that out, it's 00:01:48.200 --> 00:01:52.260 AIbox .ai slash MCP. It's a super easy MCP connector. 00:01:52.379 --> 00:01:54.939 So you give Claude this link inside of the connectors 00:01:54.939 --> 00:01:57.319 and you log into your AIbox account and now you 00:01:57.319 --> 00:01:59.099 have all of these capabilities. I have a whole 00:01:59.099 --> 00:02:01.099 website. where I explain how this works, how 00:02:01.099 --> 00:02:03.719 you can do it. And it is $8 .99 a month to get 00:02:03.719 --> 00:02:05.680 started with it. So it's super cheap. And I hope 00:02:05.680 --> 00:02:07.700 that really unlocks a lot of creativity for you 00:02:07.700 --> 00:02:09.819 to be able to get 80 different AI models and 00:02:09.819 --> 00:02:12.360 capabilities inside of Claude or whatever else 00:02:12.360 --> 00:02:14.219 you are building with. Okay, let's talk about 00:02:14.219 --> 00:02:16.419 what's going on with Arena. This is a company 00:02:16.419 --> 00:02:20.259 that is now at $100 million. They were originally 00:02:20.259 --> 00:02:22.879 created in UC Berkeley. And it's basically an 00:02:22.879 --> 00:02:24.599 AI leaderboard, right? Like this is the company 00:02:24.599 --> 00:02:27.180 where you can go and test different AI models 00:02:27.180 --> 00:02:29.469 against each other. They've hit this $100 million 00:02:29.469 --> 00:02:33.030 annual revenue, and this is eight months after 00:02:33.030 --> 00:02:34.810 they launched their paid evaluations, just back 00:02:34.810 --> 00:02:37.050 in September. triple what they were doing in 00:02:37.050 --> 00:02:39.009 January, which was 30 million, which honestly, 00:02:39.090 --> 00:02:40.550 even in January, I thought this was really impressive. 00:02:40.689 --> 00:02:43.770 So the company is now directly competing with 00:02:43.770 --> 00:02:46.669 Scale .ai and Merkur for post -training dollars 00:02:46.669 --> 00:02:49.310 because they're selling labs structured analytics 00:02:49.310 --> 00:02:53.349 built on 10 million plus human model comparisons. 00:02:53.650 --> 00:02:55.789 So basically what's going on is they've just 00:02:55.789 --> 00:02:58.490 raised $250 million total across two different 00:02:58.490 --> 00:03:01.409 rounds. So they did $150 million Series A in 00:03:01.409 --> 00:03:04.129 January at a $1 .7 billion valuation. They did 00:03:04.129 --> 00:03:07.539 that from A16Z. So the way that they actually 00:03:07.539 --> 00:03:09.639 are making money is that the public leaderboard 00:03:09.639 --> 00:03:12.210 is still free. So essentially, if you haven't 00:03:12.210 --> 00:03:14.030 tried this before, you go to the site and it 00:03:14.030 --> 00:03:16.509 puts two different AI model responses side by 00:03:16.509 --> 00:03:18.789 side. A lot of people use this because you basically 00:03:18.789 --> 00:03:21.150 get free AI usage out of it. Instead of having 00:03:21.150 --> 00:03:23.229 to pay for ChaiGPT, you can go pay this and it 00:03:23.229 --> 00:03:24.830 will give you, you know, two responses side by 00:03:24.830 --> 00:03:26.669 side. You pick which one you like better and 00:03:26.669 --> 00:03:28.509 you're helping, you know, tell it which ones 00:03:28.509 --> 00:03:31.310 are the most popular. Those leaderboards are 00:03:31.310 --> 00:03:33.449 then, you know, created so you can see, you know, 00:03:33.469 --> 00:03:35.610 oh my gosh, ChaiGPT's new model is, you know, 00:03:35.610 --> 00:03:38.349 beating the open, you know, the model from Anthropic, 00:03:38.430 --> 00:03:39.729 whatever. So that's kind of where a lot of these 00:03:39.729 --> 00:03:42.250 leaderboard come from. companies, how they kind 00:03:42.250 --> 00:03:45.069 of run and why people use them, but how they're 00:03:45.069 --> 00:03:46.870 actually making money is that they have an AI 00:03:46.870 --> 00:03:49.729 evaluations product package, which is going to 00:03:49.729 --> 00:03:53.409 show data basically to different AI labs. So 00:03:53.409 --> 00:03:55.610 like OpenAI, Ananthropic, and Google, when they 00:03:55.610 --> 00:03:58.789 have their models in there getting voted on, 00:03:58.909 --> 00:04:01.289 they'll show them what areas their models are 00:04:01.289 --> 00:04:03.169 losing on to competitors, right? So it's like, 00:04:03.250 --> 00:04:05.530 hey, look, your model's good, but anytime a healthcare 00:04:05.530 --> 00:04:08.509 question gets asked, yours is doing poorly. Or 00:04:08.509 --> 00:04:10.159 anytime a finance question... or any type of 00:04:10.159 --> 00:04:11.840 question related to this. So they're telling 00:04:11.840 --> 00:04:14.060 them exactly where their models are lacking and 00:04:14.060 --> 00:04:16.060 they're helping them to guide reinforcement learning. 00:04:16.300 --> 00:04:18.740 So it's a, you know, it's a massive value for 00:04:18.740 --> 00:04:20.339 OpenAI and Google. They're going to pay a ton 00:04:20.339 --> 00:04:22.139 of money for that. It's a great value for people 00:04:22.139 --> 00:04:24.759 that get free AI usage. And also it's interesting 00:04:24.759 --> 00:04:26.800 for all of us to see what companies are doing 00:04:26.800 --> 00:04:29.899 the best in the leaderboards. So Merkur's annualized 00:04:29.899 --> 00:04:31.819 revenue, which is one of their competitors, hit 00:04:31.819 --> 00:04:35.240 $1 billion this year and Handshake AI's training 00:04:35.240 --> 00:04:38.160 army grew from $550 million to about a billion 00:04:38.160 --> 00:04:40.079 dollars in... three months. So I think this just 00:04:40.079 --> 00:04:42.180 kind of shows the scale of a lot of these post 00:04:42.180 --> 00:04:44.740 -training markets, how much money they're actually 00:04:44.740 --> 00:04:46.819 able to make. Let's talk about what's going on 00:04:46.819 --> 00:04:49.759 with Palantir. So they have just basically selected 00:04:49.759 --> 00:04:53.000 NVIDIA's Nemotron open models to use at the US 00:04:53.000 --> 00:04:55.000 government for all of their AI. Of course, there's 00:04:55.000 --> 00:04:57.300 a ton of drama with Anthropic and even OpenAI 00:04:57.300 --> 00:05:00.019 right now, and the US government having to, you 00:05:00.019 --> 00:05:01.699 know, pull their models or tell them like, hey, 00:05:01.740 --> 00:05:03.860 you can't release your model till we give it 00:05:03.860 --> 00:05:06.100 like an accurate assessment, whatever, right? 00:05:06.750 --> 00:05:09.009 So one thing that's interesting, I think in particular 00:05:09.009 --> 00:05:10.870 with the Department of War and a lot of the beef 00:05:10.870 --> 00:05:13.149 that they had with Anthropic and them, you know, 00:05:13.149 --> 00:05:15.670 classifying Anthropic as a supply chain risk 00:05:15.670 --> 00:05:16.930 because they wouldn't let them do all the things 00:05:16.930 --> 00:05:19.329 that they wanted to and stuff. Because of that, 00:05:19.430 --> 00:05:21.470 Palantir, who does a lot of the development for 00:05:21.470 --> 00:05:23.430 the US government, a lot of software development, 00:05:23.569 --> 00:05:26.050 has selected to use an open source model. So 00:05:26.050 --> 00:05:28.889 NVIDIA has their Nemotron model. They're using 00:05:28.889 --> 00:05:31.430 this open source model and they're using it to 00:05:31.430 --> 00:05:33.949 run a lot of stuff. Now, a lot of AI, you know, 00:05:33.949 --> 00:05:35.470 work inside of the US government. And the reason 00:05:35.470 --> 00:05:37.910 why an open source model is great for this particular 00:05:37.910 --> 00:05:39.790 use case, or why the US government would be so 00:05:39.790 --> 00:05:42.410 happy to use it is because there is no centralized, 00:05:42.490 --> 00:05:45.569 you know, servers that the models got to go back 00:05:45.569 --> 00:05:48.029 to, there is no one that can say, hey, you can't 00:05:48.029 --> 00:05:49.629 use it for x, y, z reasons, it's an open source 00:05:49.629 --> 00:05:51.069 model, you can do whatever you want with it. 00:05:51.129 --> 00:05:53.610 So in basically everything that they're doing, 00:05:53.750 --> 00:05:56.069 they will have, they won't have to go and get 00:05:56.069 --> 00:05:57.910 it signed off. I remember this kind of this famous 00:05:57.910 --> 00:06:02.250 conversation between the head of war and the 00:06:02.250 --> 00:06:04.949 CEO of Anthropic Dario, and they were saying 00:06:04.949 --> 00:06:08.170 like, hey, we need it to be able to do XYZ tasks. 00:06:08.370 --> 00:06:10.470 And Dario's like, oh, well, if you're going to 00:06:10.470 --> 00:06:12.129 do that, like technically that's against our 00:06:12.129 --> 00:06:14.089 terms of service. And then they're like, okay, 00:06:14.110 --> 00:06:15.649 well, what if like there was a moment when we 00:06:15.649 --> 00:06:17.129 really needed to make some sort of, you know, 00:06:17.149 --> 00:06:19.129 critical decision on this? And Dario's like, 00:06:19.250 --> 00:06:20.750 oh, well, you could just call me and like I could, 00:06:20.769 --> 00:06:22.329 if there's like an exception to like a rule, 00:06:22.410 --> 00:06:24.050 just call me and like I'll unblock something 00:06:24.050 --> 00:06:25.230 for you. And they're like, okay, well, we don't 00:06:25.230 --> 00:06:27.329 want to call you and unblock something if we're 00:06:27.329 --> 00:06:29.399 in the middle of like a confidential. you know, 00:06:29.399 --> 00:06:31.360 battle thing, blah, blah, blah. So I mean, a 00:06:31.360 --> 00:06:32.699 lot of people have a lot of opinions on this, 00:06:32.740 --> 00:06:34.319 but these are the two arguments that we hear 00:06:34.319 --> 00:06:36.480 on this. And so this is why it looks like Palantir 00:06:36.480 --> 00:06:39.319 is going to be using this to run a lot of US 00:06:39.319 --> 00:06:41.759 government agencies. And beyond all of the drama, 00:06:41.839 --> 00:06:44.240 I just think for keeping classified data without 00:06:44.240 --> 00:06:46.160 having to send that to an outside network, I 00:06:46.160 --> 00:06:47.680 think a lot of agencies will want to use that. 00:06:47.720 --> 00:06:50.160 So agencies also get to keep the customized model 00:06:50.160 --> 00:06:52.160 weights that they train, which I think solves 00:06:52.160 --> 00:06:54.600 basically the core problem that's kept a lot 00:06:54.600 --> 00:06:57.860 of secure AI adoption stuck. and being very slow 00:06:57.860 --> 00:07:00.459 in the past. And in particular with Nemotron, 00:07:00.620 --> 00:07:02.959 this is an open model and it's going to run inside 00:07:02.959 --> 00:07:05.519 of Palantir's Sovereign AI operating system. 00:07:05.899 --> 00:07:08.360 It's going to be all run on NVIDIA's accelerated 00:07:08.360 --> 00:07:11.259 hardware. There's going to be handling data authorization. 00:07:11.579 --> 00:07:13.899 There's going to be isolation and all of the 00:07:13.899 --> 00:07:15.879 auditing is all going to be in a closed loop. 00:07:16.060 --> 00:07:18.459 I'll also say this isn't too crazy. Like two 00:07:18.459 --> 00:07:20.480 thirds of companies right now already use open 00:07:20.480 --> 00:07:22.180 models. And I think a lot of them say they're 00:07:22.180 --> 00:07:25.519 doing this because of cost efficiency. NVIDIA 00:07:25.519 --> 00:07:27.329 for frames this as kind of the template for scaling 00:07:27.329 --> 00:07:30.509 AI in regulated environments, sensitive environments, 00:07:30.610 --> 00:07:32.509 so like finance and healthcare and government, 00:07:32.689 --> 00:07:35.879 obviously. In AI regulation, Elizabeth Warren 00:07:35.879 --> 00:07:39.360 and Scanlon are reviving their Health and Location 00:07:39.360 --> 00:07:41.839 Data Protection Act. They're adding a bunch of 00:07:41.839 --> 00:07:44.879 new things to it. It now explicitly bans AI companies 00:07:44.879 --> 00:07:47.399 like OpenAI, Anthropic, and XAI from selling 00:07:47.399 --> 00:07:51.279 health and location data users enter into ChatshipT, 00:07:51.360 --> 00:07:53.959 Claude, and Grok. The bill is going to allocate 00:07:53.959 --> 00:07:57.279 $1 billion to the FTC over 10 years to enforce 00:07:57.279 --> 00:07:59.779 this. And basically this is filling a regulatory 00:07:59.779 --> 00:08:02.620 gap because a lot of these AI labs are racing. 00:08:02.800 --> 00:08:05.360 to get users. And I mean, even myself, you have 00:08:05.360 --> 00:08:07.879 ChatGPT, they just added ChatGPT Finance, which 00:08:07.879 --> 00:08:11.339 of course I, my wife like hates this kind of 00:08:11.339 --> 00:08:13.699 thing. She hates giving data to AI models, and 00:08:13.699 --> 00:08:15.839 it's probably for good reason. Myself though, 00:08:15.899 --> 00:08:18.600 I went and added basically all of my credit cards 00:08:18.600 --> 00:08:21.000 and bank accounts to my ChatGPT. It does the 00:08:21.000 --> 00:08:24.500 integration through Plaid, but it's been super 00:08:24.500 --> 00:08:26.379 useful to be able to ask it questions about where 00:08:26.379 --> 00:08:28.399 we're spending, what our recurring expenses are, 00:08:28.500 --> 00:08:31.959 income from businesses. I mean, there's just 00:08:32.009 --> 00:08:34.549 so much interesting data you can get just chatting 00:08:34.549 --> 00:08:36.669 with this thing. So I love it. I know a lot of 00:08:36.669 --> 00:08:39.549 people use this for health things as well. And 00:08:39.549 --> 00:08:41.029 there's a lot of benefits, but a lot of people 00:08:41.029 --> 00:08:43.110 are concerned about, you know, giving all of 00:08:43.110 --> 00:08:46.210 these financial and health data to these companies. 00:08:46.309 --> 00:08:48.490 What are they going to do with it? So the FTC 00:08:48.490 --> 00:08:51.529 has to write implementation rules within 180 00:08:51.529 --> 00:08:54.149 days of this bill being passed. If it is passed, 00:08:54.289 --> 00:08:57.029 the agency state and attorney general and individuals 00:08:57.029 --> 00:09:00.169 can all sue for any violations on this. There's 00:09:00.169 --> 00:09:02.690 some co -sponsors on this. We have Senator Ron 00:09:02.690 --> 00:09:05.409 Wyndon, Bernie Sanders. The original bill is 00:09:05.409 --> 00:09:08.330 from June of 2022, but it only covered data brokers. 00:09:08.470 --> 00:09:10.909 It didn't have AI companies and kind of how they're 00:09:10.909 --> 00:09:14.529 collecting data upstream. Chatbot health transcripts 00:09:14.529 --> 00:09:17.110 are unusually rich, right? Because users are 00:09:17.110 --> 00:09:19.070 going to be pasting in their lab results, or 00:09:19.070 --> 00:09:20.230 they're going to upload scans, they're going 00:09:20.230 --> 00:09:22.470 to ask follow -ups, all of that. And all that 00:09:22.470 --> 00:09:25.350 can create a lot of really complete medical profiles 00:09:25.350 --> 00:09:28.159 that no other health... you know, traditional 00:09:28.159 --> 00:09:30.019 app would actually be able to capture it. People 00:09:30.019 --> 00:09:33.120 are uploading pictures of their x rays and so 00:09:33.120 --> 00:09:35.639 much more. So anyways, I do love all of this 00:09:35.639 --> 00:09:38.799 kind of data protection, not being able to sell 00:09:38.799 --> 00:09:41.360 that. That makes a lot of sense to me, even though 00:09:41.360 --> 00:09:43.740 I don't agree with everything that Bernie Sanders 00:09:43.740 --> 00:09:48.080 does or says when it comes to AI in particular, 00:09:48.299 --> 00:09:51.059 banning AI data centers from New York. I mean, 00:09:51.080 --> 00:09:53.399 I just think that that's kind of counterintuitive 00:09:53.399 --> 00:09:56.759 to a lot of what progress and the benefit that 00:09:56.759 --> 00:09:59.200 these data centers and well, the benefit that 00:09:59.200 --> 00:10:00.960 a lot of this AI can have on people. But anyways, 00:10:01.120 --> 00:10:03.519 all of that aside, this is one where I would 00:10:03.519 --> 00:10:05.679 definitely say I'm agreeing with Elizabeth Warren 00:10:05.679 --> 00:10:08.200 and Bernie. Sanders on the face of this bill. 00:10:08.279 --> 00:10:10.039 Now, I don't know if there's secret sneaky things 00:10:10.039 --> 00:10:12.799 stuck in there, but overall, don't sell my health 00:10:12.799 --> 00:10:15.039 data to outside companies that I give you. I 00:10:15.039 --> 00:10:18.500 am definitely in agreeance with that one altogether. 00:10:18.940 --> 00:10:21.600 Okay, let's talk about what's going on with Flexion 00:10:21.600 --> 00:10:24.639 Robotics. They are training humanoid robots to 00:10:24.639 --> 00:10:26.940 run office errands. They're helping them do this 00:10:26.940 --> 00:10:29.279 all autonomously. This is a Swiss startup. It 00:10:29.279 --> 00:10:32.340 was founded by ex -NVIDIA researchers that built 00:10:32.340 --> 00:10:35.620 a software stack essentially that's letting these 00:10:35.620 --> 00:10:39.100 humanoid robots do multi -step office tasks so 00:10:39.100 --> 00:10:41.659 things like and oh and by the way you can do 00:10:41.659 --> 00:10:44.000 this all by just talking to them so i think i'm 00:10:44.000 --> 00:10:46.419 really used to seeing those like dog robots from 00:10:46.419 --> 00:10:48.519 boston dynamics and there's like remote controls 00:10:48.519 --> 00:10:50.460 where you can kind of control them i've seen 00:10:50.460 --> 00:10:51.980 them at like conferences and stuff this will 00:10:51.980 --> 00:10:53.340 be different this is you're literally going to 00:10:53.340 --> 00:10:55.299 humanoid robot and like hey go get me a coffee 00:10:55.299 --> 00:10:57.120 although i'm sure there's you know better things 00:10:57.120 --> 00:10:59.460 than telling it to go get you a coffee it'd probably 00:10:59.460 --> 00:11:03.519 be like walking around and getting things faxed 00:11:03.519 --> 00:11:05.389 or scanned or dropping things off different people's 00:11:05.389 --> 00:11:07.309 desks. I don't know, right? This could actually 00:11:07.309 --> 00:11:09.789 be quite useful. I almost imagine this like if 00:11:09.789 --> 00:11:11.769 I was in an office setting, everyone would have 00:11:11.769 --> 00:11:13.950 like a little microphone button maybe that like 00:11:13.950 --> 00:11:16.789 summons this thing. And you could like think 00:11:16.789 --> 00:11:18.570 of how you kind of use like Claude or I guess 00:11:18.570 --> 00:11:20.789 how I use Claude is I hit like my voice to text 00:11:20.789 --> 00:11:23.110 button. I give it an instruction and it runs 00:11:23.110 --> 00:11:25.070 off doing it. Let's say there's one humanoid 00:11:25.070 --> 00:11:27.409 robot for the entire office. Not everyone needs 00:11:27.409 --> 00:11:29.289 their own, obviously. So you could have like 00:11:29.289 --> 00:11:31.070 one or two, maybe one per floor or something 00:11:31.070 --> 00:11:33.289 like that. You push a button, you give it its... 00:11:33.600 --> 00:11:35.580 command you tell what you want it to do it comes 00:11:35.580 --> 00:11:37.600 over to your desk it grabs the document grabs 00:11:37.600 --> 00:11:39.879 the paper grabs the you know whatever you need 00:11:39.879 --> 00:11:42.059 it to move or do and then it goes and executes 00:11:42.059 --> 00:11:44.899 it it's like a real world version of claude i 00:11:44.899 --> 00:11:47.899 love this if you can't tell basically this is 00:11:47.899 --> 00:11:50.820 a modified unitry humanoid robot and it's running 00:11:50.820 --> 00:11:53.559 all of their systems it's um you know they're 00:11:53.559 --> 00:11:55.840 getting it to like go retrieve like mail and 00:11:55.840 --> 00:11:58.220 stuff Apparently, it can go up and down stairs. 00:11:58.419 --> 00:12:01.000 It can go in the elevator. It's super capable. 00:12:01.179 --> 00:12:04.460 And they have it stocking a shelf in a recent 00:12:04.460 --> 00:12:06.340 demo that they released. So it's very capable. 00:12:06.860 --> 00:12:09.259 Right now, they're using reinforcement learning 00:12:09.259 --> 00:12:11.620 at basically every layer of what they've built. 00:12:11.679 --> 00:12:14.019 They have a master planning. They have simulation 00:12:14.019 --> 00:12:16.519 environments. They have motor controller. If 00:12:16.519 --> 00:12:18.519 you look at the overall market, ABI Research 00:12:18.519 --> 00:12:21.379 is projecting that robot foundational model markets 00:12:21.379 --> 00:12:24.919 are going to hit about $150 billion by 2036. 00:12:26.029 --> 00:12:27.990 Flexion is basically trying to position some 00:12:27.990 --> 00:12:30.710 of the software as the product, and they're saying, 00:12:30.789 --> 00:12:32.029 look, the hardware is a commodity. There's going 00:12:32.029 --> 00:12:33.289 to be a ton of these companies that are building 00:12:33.289 --> 00:12:35.269 these humanoid robots, maybe Tesla. We're not 00:12:35.269 --> 00:12:36.730 going to compete on there, but we're going to 00:12:36.730 --> 00:12:39.009 compete on the software that helps train these 00:12:39.009 --> 00:12:41.350 robots to do things in your environment, in your 00:12:41.350 --> 00:12:44.210 office, and how you guys actually control that. 00:12:44.330 --> 00:12:46.950 So honestly, I kind of love that. It's moving. 00:12:47.070 --> 00:12:49.990 It makes the humanoid robot feel more like an 00:12:49.990 --> 00:12:52.269 LLM, and there's all of the stuff getting built 00:12:52.269 --> 00:12:55.129 on top of it. Flexicon is hardware agnostics. 00:12:55.240 --> 00:12:56.500 So they're saying they work with a whole bunch 00:12:56.500 --> 00:12:58.860 of different humanoid robot makers. They're not 00:12:58.860 --> 00:13:00.159 going to build their own robot. They're going 00:13:00.159 --> 00:13:02.360 to be able to let you use Unitary, Figure, X1, 00:13:02.480 --> 00:13:05.940 and a bunch of other platforms. China's CXMT 00:13:05.940 --> 00:13:09.100 has landed a $3 billion memory supply deal with 00:13:09.100 --> 00:13:11.639 Tencent. This is the largest commercial contract, 00:13:11.860 --> 00:13:14.879 I think, in basically domestic DRAMs makers in 00:13:14.879 --> 00:13:17.620 like their entire history of their company. The 00:13:17.620 --> 00:13:19.700 deal right now, I think, is showing that China's 00:13:19.700 --> 00:13:22.639 largest cloud operators are now actually willing 00:13:22.639 --> 00:13:25.419 to source AI memory from local. suppliers, because 00:13:25.419 --> 00:13:27.279 a lot of the US export controls are tightening 00:13:27.279 --> 00:13:29.740 around basically all of the advanced chips, right? 00:13:29.860 --> 00:13:33.440 So Tencent is a huge customer, they need so much 00:13:33.440 --> 00:13:35.679 of this. And traditionally, they have bought 00:13:35.679 --> 00:13:37.980 a lot of this from American companies. But it 00:13:37.980 --> 00:13:39.960 seems like they're able to get this from Chinese 00:13:39.960 --> 00:13:41.720 companies now because of these export controls. 00:13:41.879 --> 00:13:43.559 Something interesting about this company in particular 00:13:43.559 --> 00:13:46.960 is that CXMT is kind of moving from like smartphones 00:13:46.960 --> 00:13:50.740 and PC OEMs to now hyperscale AI customers. And 00:13:50.740 --> 00:13:52.360 I think it's kind of the first time for them. 00:13:52.419 --> 00:13:55.679 It's very similar to what SK Hynix did, and they 00:13:55.679 --> 00:13:58.480 kind of hit this dominance path because they 00:13:58.480 --> 00:14:00.899 were getting customers like NVIDIA to buy from 00:14:00.899 --> 00:14:03.419 them. I'll be interested to see if CXMT is able 00:14:03.419 --> 00:14:06.159 to move outside of just China and really get 00:14:06.159 --> 00:14:08.299 more global dominance, or if it's something that's 00:14:08.299 --> 00:14:10.480 just going to be geographically landlocked because 00:14:10.480 --> 00:14:12.360 of kind of the sensitive nature of everything 00:14:12.360 --> 00:14:14.980 happening with cloud and AI and all of the data 00:14:14.980 --> 00:14:17.159 center buildouts. This podcast was a lot of fun 00:14:17.159 --> 00:14:19.220 today. I'm actually sitting here on a yoga ball 00:14:19.220 --> 00:14:22.519 holding my baby. I got a new baby as of a couple 00:14:22.519 --> 00:14:25.590 weeks. ago and hopefully you weren't able to 00:14:25.590 --> 00:14:27.409 hear him grunting too much. He's sleeping, so 00:14:27.409 --> 00:14:29.669 he should be pretty happy. But anyways, I'm going 00:14:29.669 --> 00:14:31.850 to keep getting these podcasts out for you. There's 00:14:31.850 --> 00:14:33.750 so much happening in AI news. I love this stuff. 00:14:33.809 --> 00:14:36.529 I'm super passionate about it, but lots of fun 00:14:36.529 --> 00:14:40.009 going over. But excuse any little sounds you 00:14:40.009 --> 00:14:43.629 might be hearing in the background as we're recording 00:14:43.629 --> 00:14:46.629 in the studio with Clay. So anyways, if you wouldn't 00:14:46.629 --> 00:14:48.710 mind leaving a review, it would help the show 00:14:48.710 --> 00:14:51.169 a ton. I appreciate them all. I read them all. 00:14:52.110 --> 00:14:54.009 But anyways, thanks so much. And make sure to 00:14:54.009 --> 00:14:57.710 check out AI box dot AI, the MCP in particular, 00:14:57.850 --> 00:14:59.610 if you want to get access to all of the different 00:14:59.610 --> 00:15:02.470 AI models, there's about 80 of them in one platform, 00:15:02.549 --> 00:15:04.649 or you can build them all into Claude. So you 00:15:04.649 --> 00:15:06.769 get image, audio, video all inside of Claude. 00:15:06.789 --> 00:15:08.049 All right, you guys already know the links in 00:15:08.049 --> 00:15:09.590 the description. I'll catch you in the next episode.