WEBVTT 00:00:02.239 --> 00:00:04.780 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 --> 00:00:11.300 breaking down what it means for your life and 00:00:11.300 --> 00:00:17.429 business. Anthropic has just signed a $10 billion 00:00:17.429 --> 00:00:20.929 cloud deal with Volta for a Norway data center. 00:00:21.089 --> 00:00:23.910 Elon Musk spent almost half of the recent Tesla 00:00:23.910 --> 00:00:25.989 earnings call talking about AI and robotics. 00:00:26.510 --> 00:00:30.250 Mistral is looking at doing a $23 billion valuation 00:00:30.250 --> 00:00:34.229 as US export limits are making Europe really 00:00:34.229 --> 00:00:37.789 think about open weight AI. AMD has just committed 00:00:37.789 --> 00:00:41.539 up to $5 billion to Anthropic in a two - gigawatt 00:00:41.539 --> 00:00:45.100 chip deal and google's frozen v2 chip is looking 00:00:45.100 --> 00:00:49.340 at a 6 to 10x efficiency gain for gemini before 00:00:49.340 --> 00:00:51.500 the news i wanted to highlight one of my favorite 00:00:51.500 --> 00:00:54.140 clod use cases that i have been up to over the 00:00:54.140 --> 00:00:56.619 last two months and that is vibe coding my very 00:00:56.619 --> 00:01:00.399 own ios and android app it is a bible study guide 00:01:00.399 --> 00:01:03.359 app that has cost me millions of clod tokens 00:01:03.359 --> 00:01:06.480 i actually had to get two clod max $200 a month 00:01:06.480 --> 00:01:08.180 subscriptions to keep up with it because there 00:01:08.180 --> 00:01:10.340 was so much content to write i ended up writing 00:01:10.379 --> 00:01:12.879 writing over 1200 chapter guides. So every single 00:01:12.879 --> 00:01:14.939 chapter in the Bible, there's a study guide that 00:01:14.939 --> 00:01:16.980 breaks down the history, the context, the backgrounds 00:01:16.980 --> 00:01:19.420 of the people in it. I went and scraped all of 00:01:19.420 --> 00:01:21.459 the images from the Met and a bunch of other 00:01:21.459 --> 00:01:24.040 famous museums to get the last thousand years 00:01:24.040 --> 00:01:26.700 of incredible historical Christian artwork and 00:01:26.700 --> 00:01:29.099 add it in there. And then I indexed it. So every 00:01:29.099 --> 00:01:31.519 single image went to the chapter that it was 00:01:31.519 --> 00:01:33.280 painted about. So when you read the study guide, 00:01:33.379 --> 00:01:35.280 there's paintings of all of this stuff in there. 00:01:35.319 --> 00:01:37.019 I really tried to put a lot of the human element 00:01:37.019 --> 00:01:38.819 and the human touch in there. Well, it's also 00:01:38.819 --> 00:01:41.200 pretty upfront. This is, you know, AI generated 00:01:41.200 --> 00:01:44.019 and really helps people understand things with 00:01:44.019 --> 00:01:46.180 AI. You also have the ability to click on any 00:01:46.180 --> 00:01:48.439 word, get an explanation of it and any verse 00:01:48.439 --> 00:01:51.420 and get a background explanation on what is happening 00:01:51.420 --> 00:01:53.480 in the verse and the context there. This is something 00:01:53.480 --> 00:01:55.480 I've been working on for many months, but I spent 00:01:55.480 --> 00:01:58.099 a ton of time making this look beautiful and 00:01:58.099 --> 00:02:00.659 super aesthetic. So if you're interested in studying 00:02:00.659 --> 00:02:02.519 the Bible, this is completely free. There's no 00:02:02.519 --> 00:02:05.000 ads. This is just a labor of love for me. Go 00:02:05.000 --> 00:02:06.640 check it out. I'll leave a link in the description. 00:02:06.739 --> 00:02:08.580 And if you're not interested in the app per se. 00:02:08.810 --> 00:02:10.689 still go look at just the page if you want to 00:02:10.689 --> 00:02:12.610 see the quality of what you can build with claude 00:02:12.610 --> 00:02:15.110 everything from the screenshots on the app store 00:02:15.110 --> 00:02:17.990 listing to all of the graphics and images were 00:02:17.990 --> 00:02:20.569 all generated by ai other than of course all 00:02:20.569 --> 00:02:22.849 of the historical artwork so go check it out 00:02:22.849 --> 00:02:25.389 it's learn of christ .com is the website but 00:02:25.389 --> 00:02:27.030 i'll leave a link in the description to the actual 00:02:27.030 --> 00:02:29.409 app to download all of it was built with claude 00:02:29.409 --> 00:02:32.349 anthropic just locked in 10 billion dollars worth 00:02:32.349 --> 00:02:34.770 of computing power from volta this is a startup 00:02:34.770 --> 00:02:37.270 that was founded just this year they purchased 00:02:37.270 --> 00:02:39.629 six years worth of compute for them. But for 00:02:39.629 --> 00:02:41.250 me, it's pretty wild to see a company starting 00:02:41.250 --> 00:02:43.430 this year and Anthropic is going and buying the 00:02:43.430 --> 00:02:45.229 compute from them versus Anthropic trying to 00:02:45.229 --> 00:02:46.689 stand this up themselves, like opening up. There's 00:02:46.689 --> 00:02:48.330 two different strategies that we're seeing right 00:02:48.330 --> 00:02:50.569 now. I think this is just showing how hard it 00:02:50.569 --> 00:02:53.569 is for basically all of these AI labs to get 00:02:53.569 --> 00:02:56.629 chip capacity. I think Texas also just banned 00:02:56.629 --> 00:02:58.650 all new data centers while they go and audit 00:02:58.650 --> 00:03:02.030 all of their grids. So there is a lot of headwinds 00:03:02.030 --> 00:03:03.949 for the data centers. And if you're able to get 00:03:03.949 --> 00:03:06.650 access, if you're able to get approval and regulatory 00:03:06.650 --> 00:03:08.960 approval, in different locations, I think companies 00:03:08.960 --> 00:03:10.639 like Anthropic OpenAI, they're just going to 00:03:10.639 --> 00:03:13.680 swoop up and get as much bandwidth as they can, 00:03:13.800 --> 00:03:16.280 meta included. This compute deal in particular 00:03:16.280 --> 00:03:19.819 is going to be about 133 megawatts. It's the 00:03:19.819 --> 00:03:22.539 data center in Norway. It was originally built 00:03:22.539 --> 00:03:26.580 with crypto miner BidDeer, and it's now powered 00:03:26.580 --> 00:03:29.919 by NVIDIA's newest AI chip, the Vera Rubin. Volta 00:03:29.919 --> 00:03:33.280 raised $2 .4 billion for their valuation. They're 00:03:33.280 --> 00:03:36.259 based largely on this one Anthropic contract. 00:03:36.460 --> 00:03:38.180 So like Anthropic gives them a contract, they're 00:03:38.180 --> 00:03:39.860 able to raise a ton of money to go and build 00:03:39.860 --> 00:03:41.860 it for them. This is, I think, something very 00:03:41.860 --> 00:03:44.740 typical for some of these new GPU cloud startups. 00:03:44.939 --> 00:03:47.180 They're all competing for AI lab customers. And 00:03:47.180 --> 00:03:49.139 if they can get the deal, then they can raise 00:03:49.139 --> 00:03:51.560 more money to build what they need. Anthropic 00:03:51.560 --> 00:03:53.719 right now is spreading their compute across a 00:03:53.719 --> 00:03:55.139 bunch of different platforms. So you've probably 00:03:55.139 --> 00:03:57.419 heard a lot of the deals, but AWS is one of the 00:03:57.419 --> 00:03:59.939 OGs. Google Cloud is actually the OG. They were 00:03:59.939 --> 00:04:02.580 the first ones to give them $300 million, which 00:04:02.580 --> 00:04:04.319 seemed like an insane amount of money when they 00:04:04.319 --> 00:04:07.379 were much, much smaller. company. SpaceX has 00:04:07.379 --> 00:04:09.719 signed a big deal with them where they pay them 00:04:09.719 --> 00:04:12.259 almost a billion dollars a month, and now Volta. 00:04:12.360 --> 00:04:14.259 All of this, I think, is reducing the risk that 00:04:14.259 --> 00:04:17.220 any one single provider is going to run short 00:04:17.220 --> 00:04:19.879 if there's a major model launch or their usage 00:04:19.879 --> 00:04:21.959 is really ramping up. They have Amazon, they 00:04:21.959 --> 00:04:23.579 have Google, they have SpaceX, they have Volta, 00:04:23.620 --> 00:04:25.500 and I'm sure they'll roll out more. It's interesting 00:04:25.500 --> 00:04:28.240 because OpenAI and some others feels like they 00:04:28.240 --> 00:04:30.800 might be a little bit more restricted or they 00:04:30.800 --> 00:04:33.720 might have less providers and they're trying 00:04:33.720 --> 00:04:35.350 to build it out on their own. So it'll be interesting 00:04:35.350 --> 00:04:37.230 to see which of these two strategies works the 00:04:37.230 --> 00:04:40.790 best. AI labs right now are going way past just 00:04:40.790 --> 00:04:43.490 relying on a handful of cloud giants, and they're 00:04:43.490 --> 00:04:46.930 willing to pay premium prices about 12 .5 million 00:04:46.930 --> 00:04:50.930 per megawatt per year if they can lock in dedicated 00:04:50.930 --> 00:04:53.870 capacity years ahead of time. So it's not just 00:04:53.870 --> 00:04:55.430 like, hey, well, you know, we'll pay this price 00:04:55.430 --> 00:04:56.829 today. It's like we'll pay this price today and 00:04:56.829 --> 00:04:58.250 for six years. And that's what they're really 00:04:58.250 --> 00:05:01.509 looking for is this extended capacity. Elon Musk 00:05:01.509 --> 00:05:04.589 is now spending about half of his Tesla earning 00:05:04.589 --> 00:05:06.709 calls talking about AI and robotics. This is 00:05:06.709 --> 00:05:08.509 according to the last earnings call he was just 00:05:08.509 --> 00:05:10.589 on. He was talking about the new Optimus robot. 00:05:10.810 --> 00:05:14.170 This is up from 15 to 20 percent in 2022. So 00:05:14.170 --> 00:05:16.430 the amount of time he spends on the earnings 00:05:16.430 --> 00:05:18.290 call talking about robots and AI is definitely 00:05:18.290 --> 00:05:21.870 accelerating. The cars are obviously the main 00:05:21.870 --> 00:05:24.209 driver of revenue, about 70 percent of Tesla's 00:05:24.209 --> 00:05:26.560 revenue. But I think this really just goes to 00:05:26.560 --> 00:05:29.199 show you where Elon Musk believes the future 00:05:29.199 --> 00:05:32.379 of Tesla is, where the value is going to grow, 00:05:32.560 --> 00:05:36.279 how the stock is being priced in the market. 00:05:36.740 --> 00:05:39.379 Optimus, which is Tesla's humanoid robot, consumed 00:05:39.379 --> 00:05:43.379 nearly a third of all of the comments and things 00:05:43.379 --> 00:05:46.100 that he was talking about on the Q3 2025 call 00:05:46.100 --> 00:05:50.319 versus about 2 % or less just three years ago. 00:05:50.439 --> 00:05:52.360 So this is something that is just being talked 00:05:52.360 --> 00:05:55.790 about all the time. executives like the CFO and 00:05:55.790 --> 00:05:58.089 the VP of engineering still spend about 30 % 00:05:58.089 --> 00:06:01.029 of their call on cars and manufacturing. That's 00:06:01.029 --> 00:06:03.930 down from 50 % in 2024. And of course, I know 00:06:03.930 --> 00:06:06.129 this is ridiculous. I know people are like laughing 00:06:06.129 --> 00:06:08.529 at me for talking about the percentage of time 00:06:08.529 --> 00:06:10.009 that the people on the earnings call talking 00:06:10.009 --> 00:06:11.810 about different topics. But guys, this really 00:06:11.810 --> 00:06:13.689 goes to show where the priorities are. And I 00:06:13.689 --> 00:06:16.310 think you can get a pretty good understanding 00:06:16.310 --> 00:06:18.649 of the direction the company is pushing in and 00:06:18.649 --> 00:06:21.689 the objectives and the things that they're going 00:06:21.689 --> 00:06:24.050 to be working on. So Tesla shipped about 500 00:06:24.050 --> 00:06:26.589 ,000 vehicles last quarter. They continue to 00:06:26.589 --> 00:06:29.889 rely on car sales, obviously, for most of the 00:06:29.889 --> 00:06:31.670 revenue. A lot of people are saying this is making 00:06:31.670 --> 00:06:34.490 them bearish on the car segment. Well, I think 00:06:34.490 --> 00:06:36.250 overall, most people are saying this is just 00:06:36.250 --> 00:06:38.430 making them very bullish on robotics, or at least 00:06:38.430 --> 00:06:40.629 the amount of focus and time that Tesla is going 00:06:40.629 --> 00:06:42.810 to put into that. Mistral, which is a French 00:06:42.810 --> 00:06:46.529 AI lab, is raising funds at a $23 billion valuation. 00:06:46.709 --> 00:06:47.930 I've talked about this on the podcast before 00:06:47.930 --> 00:06:49.410 that they were looking at it. It looks like it's 00:06:49.410 --> 00:06:51.389 officially in the books and happening. This is 00:06:51.389 --> 00:06:53.680 about double their last year. valuation, which 00:06:53.680 --> 00:06:56.459 was $13 .5 billion. There's of course, all of 00:06:56.459 --> 00:06:58.560 the US export control limits, there is the whole 00:06:58.560 --> 00:07:00.519 safety incidents that are all happening in all 00:07:00.519 --> 00:07:02.459 of these American companies, opening eye and 00:07:02.459 --> 00:07:04.420 anthropic are both saying like, look, our models 00:07:04.420 --> 00:07:06.779 are breaking out of our sandboxes and hacking 00:07:06.779 --> 00:07:09.120 companies and all this wildness is going on. 00:07:09.279 --> 00:07:12.600 Because of that. companies and countries are 00:07:12.600 --> 00:07:15.339 pushing Europe towards open weight models. This 00:07:15.339 --> 00:07:17.779 is, of course, the AI system where the code is 00:07:17.779 --> 00:07:19.819 publicly available, not necessarily how you, 00:07:19.879 --> 00:07:22.459 so open source would be when the whole model 00:07:22.459 --> 00:07:24.500 and how you built it is publicly available, but 00:07:24.500 --> 00:07:26.680 open weight means you don't tell people necessarily 00:07:26.680 --> 00:07:28.819 how you train the model, the code to train it, 00:07:28.879 --> 00:07:30.819 but you let people download the model, they can 00:07:30.819 --> 00:07:33.379 run the model, you give them the weights, and 00:07:33.379 --> 00:07:35.500 so it all works on your own computer. They're 00:07:35.500 --> 00:07:37.699 making this all publicly available because of 00:07:37.699 --> 00:07:40.160 some of their worries and some of the... risks, 00:07:40.220 --> 00:07:42.899 alleged risks with the American companies. The 00:07:42.899 --> 00:07:44.980 company's revenue at Mistral has jumped about 00:07:44.980 --> 00:07:48.100 20x in the last year, because they've done a 00:07:48.100 --> 00:07:49.579 bunch of deals with the French government with 00:07:49.579 --> 00:07:53.040 Microsoft with HSBC. And because of all of these 00:07:53.040 --> 00:07:56.379 kind of export restrictions on opening iron anthropics 00:07:56.379 --> 00:07:58.379 models, like Trump came out and was like, hey, 00:07:58.399 --> 00:08:00.339 you know, like anthropic can't push out their 00:08:00.339 --> 00:08:02.360 Fable 5 model is too dangerous, yada, yada. So 00:08:02.360 --> 00:08:04.480 he basically banned it for all the other countries. 00:08:05.310 --> 00:08:07.490 even if they might have wanted to use it. And 00:08:07.490 --> 00:08:09.649 so I think that put a bad taste in a lot of people's 00:08:09.649 --> 00:08:12.050 mouth. And they're like, look, we need like a 00:08:12.050 --> 00:08:17.319 European. or European model that isn't, the Americans 00:08:17.319 --> 00:08:19.819 can't just shut off for us. So, I mean, I think 00:08:19.819 --> 00:08:21.579 from a geopolitical perspective, this is great. 00:08:21.680 --> 00:08:23.699 And I love all the competition. I love that it's 00:08:23.699 --> 00:08:26.060 not just going to be OpenAI and Anthropic leading 00:08:26.060 --> 00:08:28.519 the AI charge. I want to see as many competitors 00:08:28.519 --> 00:08:30.680 in this as possible. So I, for one, I'm actually 00:08:30.680 --> 00:08:32.720 pretty happy with this. Mistral shifted their 00:08:32.720 --> 00:08:34.759 strategy basically from competing on just the 00:08:34.759 --> 00:08:37.460 raw performance to building smaller custom models 00:08:37.460 --> 00:08:39.179 for specific industries. So they're going after 00:08:39.179 --> 00:08:41.580 things like manufacturing, finance. They're also 00:08:41.580 --> 00:08:44.379 selling hosted cloud services. and embedded engineering 00:08:44.379 --> 00:08:46.480 teams. But OpenAI and Anthropic are also kind 00:08:46.480 --> 00:08:48.279 of doing that forward deployed engineering teams. 00:08:48.559 --> 00:08:51.059 The open weight model adoption is getting really 00:08:51.059 --> 00:08:54.740 popular in Europe and in all of, I think, the 00:08:54.740 --> 00:08:56.539 world. But you also have a lot of these cheaper 00:08:56.539 --> 00:08:58.700 Chinese alternatives. You have DeepSeek and you 00:08:58.700 --> 00:09:02.220 have Quen that recently came out and you have, 00:09:02.220 --> 00:09:04.559 you know, a lot of these other players coming 00:09:04.559 --> 00:09:07.279 out of China, which just make this a very cheap 00:09:07.279 --> 00:09:09.059 way to run these models. You can run them on 00:09:09.059 --> 00:09:11.080 your own devices and Mistral is trying to get 00:09:11.080 --> 00:09:14.830 into that mix. AMD is going to invest $5 billion 00:09:14.830 --> 00:09:16.769 in Anthropic, and they're going to supply two 00:09:16.769 --> 00:09:20.909 gigawatts of its Instinct MI450 chips. This is 00:09:20.909 --> 00:09:24.590 starting in 2027, but it's basically AMD's biggest 00:09:24.590 --> 00:09:27.809 foothold into AI training alongside NVIDIA. This, 00:09:27.909 --> 00:09:29.710 I think, is important because it's basically 00:09:29.710 --> 00:09:32.129 going to give Anthropic a second major chip supplier. 00:09:32.450 --> 00:09:35.950 They're trying to build a computing power and 00:09:35.950 --> 00:09:38.269 everything that's needed for Claude, but they're 00:09:38.269 --> 00:09:41.190 also being able to establish AMD as a credible 00:09:41.190 --> 00:09:43.419 alternative to NVIDIA. nvidia for frontier ai 00:09:43.419 --> 00:09:45.779 lab so this is a great deal for amd obviously 00:09:45.779 --> 00:09:47.559 right they're like look anthropic uses us or 00:09:47.559 --> 00:09:50.240 the big player we've seen kind of the same logo 00:09:50.240 --> 00:09:53.940 name brand um show off for basically all of the 00:09:53.940 --> 00:09:56.860 other cloud providers amazon aws was kind of 00:09:56.860 --> 00:09:58.500 the most famous they gave anthropic a lot of 00:09:58.500 --> 00:09:59.919 money and we're like look anthropics running 00:09:59.919 --> 00:10:03.399 on aws we had a google cloud that did that and 00:10:03.399 --> 00:10:06.820 even to this day I use AWS for my startup, AIbox 00:10:06.820 --> 00:10:09.080 .ai. You can get like 80 different AI models, 00:10:09.340 --> 00:10:11.639 right? Like Anthropic and OpenAI and Gemini and 00:10:11.639 --> 00:10:14.200 Cloud, like all of them in one place for $9 a 00:10:14.200 --> 00:10:15.860 month. But what's interesting is we run our whole 00:10:15.860 --> 00:10:18.600 backend on AWS. And to this day, we still get 00:10:18.600 --> 00:10:21.039 AWS reps emailing me all the time being like, 00:10:21.120 --> 00:10:24.840 hey, man, if you want to move your usage of Cloud 00:10:24.840 --> 00:10:27.700 off of, because we have to go buy like API credits 00:10:27.700 --> 00:10:30.440 from Anthropic and pull them into our platform 00:10:30.440 --> 00:10:32.059 and from all the different players, right? They're 00:10:32.059 --> 00:10:33.179 like, hey, if you want to move it off of there 00:10:33.179 --> 00:10:35.330 and just buy it directly, from us aws we can 00:10:35.330 --> 00:10:36.730 get you all these deals we can get you some free 00:10:36.730 --> 00:10:39.149 credits like they have all of these kind of perks 00:10:39.149 --> 00:10:41.389 that they've been these deals that they've been 00:10:41.389 --> 00:10:43.669 making for a very long time. And so it's interesting 00:10:43.669 --> 00:10:45.970 because all of them want to kind of show off 00:10:45.970 --> 00:10:48.309 the big player and then try to get some of the 00:10:48.309 --> 00:10:51.049 usage from that from out from under them. So 00:10:51.049 --> 00:10:53.929 in this case, in particular with AMD, the first 00:10:53.929 --> 00:10:56.029 gigawatt is going to deploy in the first half 00:10:56.029 --> 00:11:00.090 of 2027. Inside of AMD's Helio rack scale systems, 00:11:00.289 --> 00:11:02.509 the second gigawatt is going to come after that. 00:11:02.710 --> 00:11:06.230 Anthropix yearly revenue hit $47 billion as of 00:11:06.230 --> 00:11:10.460 May, that's up 10 billion. from all of last year. 00:11:10.600 --> 00:11:12.539 This, of course, is because Cloud Code is getting 00:11:12.539 --> 00:11:15.179 super popular, adoption with developers. I'm 00:11:15.179 --> 00:11:17.240 not even a developer, but I use it all day, every 00:11:17.240 --> 00:11:19.019 day. Although I've recently switched to OpenAI 00:11:19.019 --> 00:11:21.419 and I'll keep people updated on what I'm doing 00:11:21.419 --> 00:11:23.799 there, but I'm testing it out and seeing some 00:11:23.799 --> 00:11:25.960 really incredible results. So I do feel like 00:11:25.960 --> 00:11:27.740 there's some competition there. Anthropic is 00:11:27.740 --> 00:11:31.440 spending about $1 .25 billion a month through 00:11:31.440 --> 00:11:35.080 May of 2029 on compute from SpaceX's Colossus 00:11:35.080 --> 00:11:37.259 One data center. They're spending... That's about 00:11:37.259 --> 00:11:40.059 $15 billion annually. And they also have deals 00:11:40.059 --> 00:11:42.340 with Amazon, Google, and a bunch of other players. 00:11:42.399 --> 00:11:45.299 So tons is being spent here. Google is building 00:11:45.299 --> 00:11:49.340 a new AI server chip called Frozen V2. It's going 00:11:49.340 --> 00:11:51.500 to run the Gemini models. And apparently it's 00:11:51.500 --> 00:11:54.659 going to be six to 10 times more efficient when 00:11:54.659 --> 00:11:56.779 it comes to what they have for their current 00:11:56.779 --> 00:11:59.179 chips. So this is going to come in 2028. All 00:11:59.179 --> 00:12:01.000 of these chip announcements, like it's kind of 00:12:01.000 --> 00:12:02.559 exciting when you're like, oh my gosh, 10 times 00:12:02.559 --> 00:12:04.379 more efficient. And it's like, yeah, and it'll 00:12:04.379 --> 00:12:06.500 be here in a couple of years. So I think the 00:12:06.500 --> 00:12:08.059 model, like there's just, and I know that's like 00:12:08.059 --> 00:12:10.159 not that much time, but especially for hardware, 00:12:10.220 --> 00:12:12.100 but like there's just so much can change in the 00:12:12.100 --> 00:12:13.759 market for how efficient these models can run. 00:12:13.929 --> 00:12:15.470 Like, I wouldn't be surprised if they figured 00:12:15.470 --> 00:12:17.149 out how to make the current Gemini model 10 times 00:12:17.149 --> 00:12:19.889 more efficient just because of software updates 00:12:19.889 --> 00:12:22.009 in the next two years, right? But in any case, 00:12:22.070 --> 00:12:25.690 Alphabet is going to spend $180 to $190 billion 00:12:25.690 --> 00:12:28.610 on AI infrastructure this year alone. And so 00:12:28.610 --> 00:12:31.789 that chip is going to cut their power costs significantly. 00:12:31.909 --> 00:12:34.009 If it really is 10 times more efficient, that 00:12:34.009 --> 00:12:36.269 could save them billions of dollars. It'll also 00:12:36.269 --> 00:12:39.129 let Google serve. more AI requests and kind of 00:12:39.129 --> 00:12:42.029 make that cheaper and onboard more users. Frozen 00:12:42.029 --> 00:12:45.009 V2 is going to measure efficiency by tokens generated 00:12:45.009 --> 00:12:47.789 per watt of power. That is basically the standard 00:12:47.789 --> 00:12:51.070 metric for comparing how much AI output a chip 00:12:51.070 --> 00:12:53.950 producer or how much it produces per unit of 00:12:53.950 --> 00:12:56.409 electricity. OpenAI released their very own inference 00:12:56.409 --> 00:12:59.830 chip called Jalapeno in June and Anthropic is 00:12:59.830 --> 00:13:02.509 negotiating with Samsung to build custom chips. 00:13:02.649 --> 00:13:05.269 I think every major AI lab right now is treating 00:13:05.269 --> 00:13:08.159 in -house silicone is essential. They all want 00:13:08.159 --> 00:13:09.580 to build their own chips. They know they can't 00:13:09.580 --> 00:13:12.019 just rely on other people. Google has designated 00:13:12.019 --> 00:13:15.539 custom AI chips as it's kind of a big priority. 00:13:15.620 --> 00:13:17.059 It's actually something that they've been designing 00:13:17.059 --> 00:13:20.000 internally since 2015. And they have used them 00:13:20.000 --> 00:13:23.080 alongside NVIDIA GPUs. But they are also owning 00:13:23.080 --> 00:13:25.320 the chip design and doing that basically gives 00:13:25.320 --> 00:13:26.879 them a little bit earlier and cheaper access 00:13:26.879 --> 00:13:29.620 to manufacturing and some of the capacity and 00:13:29.620 --> 00:13:31.679 all of the stuff that they need instead of just 00:13:31.679 --> 00:13:34.659 going to NVIDIA to buy everything. That was the 00:13:34.659 --> 00:13:36.320 podcast for today, guys. Thank you. so much for 00:13:36.320 --> 00:13:38.299 tuning into the show. Like I mentioned earlier, 00:13:38.500 --> 00:13:40.720 if you're interested in checking out my 100 % 00:13:40.720 --> 00:13:44.559 vibe coded new app, it is called Learn of Christ. 00:13:44.700 --> 00:13:46.899 It is a Bible study app. It's free. There's no 00:13:46.899 --> 00:13:50.240 ads on it. And I hope that it is inspiring, even 00:13:50.240 --> 00:13:52.500 if you haven't done a lot of Bible study in your 00:13:52.500 --> 00:13:55.539 life, just for the design and what you can build 00:13:55.539 --> 00:13:59.240 with AI as far as like a huge content library, 00:13:59.600 --> 00:14:02.320 pulling in real world data, building something 00:14:02.320 --> 00:14:05.320 that looks and acts great. And I mean, There's 00:14:05.320 --> 00:14:06.860 a lot of companies that have spent hundreds of 00:14:06.860 --> 00:14:08.840 thousands of dollars building things with less 00:14:08.840 --> 00:14:11.259 features that are less smooth. So I'd love for 00:14:11.259 --> 00:14:13.559 you to go check it out. It is linked in the description 00:14:13.559 --> 00:14:16.019 and is also learnofchrist .com. Thank you so 00:14:16.019 --> 00:14:17.600 much for tuning into the podcast. I will catch 00:14:17.600 --> 00:14:18.899 you all in the next episode.