WEBVTT - ASML's Record Surge, Microsoft Investigates DeepSeek Code 0:00:02.759 --> 0:00:09.920 Bloomberg Audio Studios, podcasts, radio news from the heart of 0:00:10.000 --> 0:00:15.720 where innovation, money and power collide in Silicon Valley and beyond. 0:00:16.120 --> 0:00:40.360 This is Bloomberg Technology with Caroline Hyde and Ed Lovelow. 0:00:34.280 --> 0:00:36.360 Live from New York. I'm Caroline Hyde. 0:00:36.360 --> 0:00:40.520 And I'm Mike Sheppard in San Francisco. This is Bloomberg Technology. 0:00:40.000 --> 0:00:43.880 Coming up asml fights back Deep Seak concerns and surges 0:00:43.920 --> 0:00:46.840 the most since twenty twenty as stronger chip equipment orders, 0:00:46.880 --> 0:00:51.400 Impress plus, did Deep Seek improperly use open ai data? 0:00:51.880 --> 0:00:54.960 Microsoft and open Ai investigate? We have the latest and 0:00:55.000 --> 0:00:57.080 we sit down with T Mobile CEO on the heels 0:00:57.080 --> 0:00:59.160 of the company's standout fourth quarter. 0:00:59.240 --> 0:01:01.400 But first we check in on these markets. 0:01:01.520 --> 0:01:03.880 Remember, folks, it is fed day where we get a 0:01:03.920 --> 0:01:05.160 pause that comes later today. 0:01:05.160 --> 0:01:06.520 But we're off by four tens percent. 0:01:06.560 --> 0:01:08.760 Now's that on the downside, we're being dragged down by 0:01:08.760 --> 0:01:11.160 the like some video Microsoft in the red. Let's move 0:01:11.160 --> 0:01:13.680 on and see what earnings we have coming off after 0:01:13.720 --> 0:01:15.560 the bell because it is thick and fast. Not only 0:01:15.600 --> 0:01:17.759 are we reading from deep Seek, but we've got to. 0:01:17.680 --> 0:01:18.760 Digest the real numbers. 0:01:18.760 --> 0:01:22.240 Microsoft, Meta, Tesla, Apple a little bit later, but tonight, 0:01:22.480 --> 0:01:24.600 we keep an eye on these front three and then 0:01:24.680 --> 0:01:26.760 we think about what we've already had out of the 0:01:26.800 --> 0:01:30.160 bell in Europe and is ASML the chip equipment maker 0:01:30.200 --> 0:01:34.199 managing to fight back some of those anxieties around deep 0:01:34.240 --> 0:01:38.800 seek cheap AA models? Ultimately does it matter for chips? 0:01:38.880 --> 0:01:42.280 For chip equipment ASML surging currently we see up five 0:01:42.319 --> 0:01:44.360 percent in US trading. It was up the most that's 0:01:44.360 --> 0:01:46.600 twenty twenty earlier, Mike, and this was on the back 0:01:46.640 --> 0:01:47.960 of very strong orders. 0:01:49.480 --> 0:01:52.160 And we're staying on ASML and their big earnings. Be 0:01:52.360 --> 0:01:55.960 joining us to discuss this is Bloomberg's Chan coach in Amsterdam. 0:01:56.320 --> 0:01:56.559 Chan. 0:01:56.920 --> 0:02:01.120 There were a surgeon orders in the last quarter for ASML, 0:02:01.520 --> 0:02:04.240 far more than expected. But all of that was before 0:02:04.280 --> 0:02:07.919 the deep Seek revelations. How did the CEO of ASML 0:02:08.320 --> 0:02:10.040 message this to investors today? 0:02:11.919 --> 0:02:12.280 All right? 0:02:12.440 --> 0:02:17.840 CHRISTOSML CEO doubled down on AI. Basically, he thinks Deep 0:02:17.880 --> 0:02:22.160 Seek will be good news for ASML, and his thinking 0:02:22.240 --> 0:02:26.680 is that anyone that lowers cost lowers barriers to entry, 0:02:26.960 --> 0:02:31.200 so lower cost means AI can be used in more applications. 0:02:31.400 --> 0:02:35.480 More applications mean more chips, and more chips would mean 0:02:35.600 --> 0:02:39.880 ASML can sell more machines to chip makers, and they 0:02:39.919 --> 0:02:44.200 think deep sea can lead to democratization of AI, which 0:02:44.240 --> 0:02:47.440 will then create more customers for them charm. 0:02:47.480 --> 0:02:51.040 When we talk about chips, we can't get away from geopolitics. 0:02:51.320 --> 0:02:55.000 How did some of the US restrictions on sales of 0:02:55.120 --> 0:02:58.480 sensitive equipment to China play in the results that we 0:02:58.520 --> 0:02:59.120 saw today? 0:03:00.480 --> 0:03:04.200 Will China accounted for twenty seven percent of ASML sales 0:03:04.240 --> 0:03:07.520 in the fourth quarter, which was down from nearly half 0:03:08.639 --> 0:03:12.760 of its sales in the third quarter. The company expects 0:03:12.919 --> 0:03:16.519 China sales to follow up one fifth of its total 0:03:16.560 --> 0:03:19.560 revenue this year, and they think that their sales in 0:03:19.639 --> 0:03:22.720 China are taking a hit from the prospect of more 0:03:22.840 --> 0:03:26.920 US restrictions under Donald Trump's administration, as well as a 0:03:26.960 --> 0:03:29.160 broader weakness in the chip sector. 0:03:30.520 --> 0:03:33.240 Bloomberg's John Coach, thank you so much, Caro. 0:03:33.680 --> 0:03:36.120 We booke a plenty more earnings and other socks we're 0:03:36.160 --> 0:03:36.960 watching this morning. 0:03:37.360 --> 0:03:39.080 Apple t Mobile. 0:03:38.800 --> 0:03:41.360 In particular, two companies that have been working secretly together. 0:03:41.360 --> 0:03:44.480 We understand ad support for SpaceX is Starling network in 0:03:44.520 --> 0:03:46.440 a surprise release. The tech is now being supported with 0:03:46.480 --> 0:03:49.640 an installation of the iPhone's latest software update on Monday, 0:03:49.840 --> 0:03:51.520 but more on T Mobile because. 0:03:51.360 --> 0:03:53.200 The company, of course is out with this fourth quarter. 0:03:53.040 --> 0:03:57.840 Results beating analyst expectations. The telecom providers saw continued growth 0:03:57.840 --> 0:04:00.880 in wireless subscribers and home internet customers. Get this, adding 0:04:00.920 --> 0:04:03.720 nine hundred and three thousand monthly phone customers and guiding 0:04:03.720 --> 0:04:06.760 from continued strong growth. Joining us now might see that 0:04:06.800 --> 0:04:09.920 his Tmobile CEO. And what's interesting is descend. But you said, 0:04:10.240 --> 0:04:12.280 be cautious about our fourth quarter. 0:04:12.680 --> 0:04:13.360 What changed? 0:04:13.880 --> 0:04:15.640 Well, the only thing I was pointing out back then, 0:04:15.680 --> 0:04:18.320 which maybe was misunderstood, is that it's a seasonal quarter, 0:04:18.400 --> 0:04:20.440 and you know, the last couple of weeks are important. 0:04:20.680 --> 0:04:22.840 But you know, we just are firing on all cylinders. 0:04:22.880 --> 0:04:26.599 We just finished our strongest growth year ever, and we 0:04:26.680 --> 0:04:29.760 finished it strong with a monster Q four that exceeded 0:04:29.760 --> 0:04:31.160 a lot of people's expectations. 0:04:31.240 --> 0:04:33.760 And you're guiding for postpaid net customers to be adding 0:04:33.839 --> 0:04:35.960 up to six million in twenty twenty five. That's a 0:04:35.960 --> 0:04:38.400 big ramp. Where are those customers coming to? 0:04:38.440 --> 0:04:41.000 You know, this is the strongest beginning of year guide 0:04:41.040 --> 0:04:43.039 on new subscribers we've ever given, and. 0:04:43.040 --> 0:04:43.800 It's really simple. 0:04:43.839 --> 0:04:47.000 Our consumer business is firing on all cylinders in the 0:04:47.040 --> 0:04:50.120 top one hundred markets, also smaller markets in rural areas. 0:04:50.279 --> 0:04:53.720 But now our business sector is starting to rise as well. 0:04:53.760 --> 0:04:56.640 And so you look across all the segments. People are 0:04:56.640 --> 0:04:57.120 flocking to. 0:04:57.160 --> 0:04:59.400 T Mobile, governments to New York one of them that 0:04:59.440 --> 0:05:04.000 you've added. But I'm interested that the other competitors out there, 0:05:04.640 --> 0:05:07.320 at and T Verizon they added to where are these 0:05:07.360 --> 0:05:08.320 customers coming from? 0:05:08.520 --> 0:05:10.320 Look this is I like to think about this. 0:05:10.680 --> 0:05:12.880 You know, we're the best house in a great neighborhood. 0:05:13.600 --> 0:05:16.479 I don't need for our competitors to fail in order 0:05:16.480 --> 0:05:19.839 for us to succeed. We are taking share. But also 0:05:19.960 --> 0:05:22.800 the market is vibrant and growing, and people are, you know, 0:05:22.839 --> 0:05:25.400 getting more value out of rate plans. They're buying up 0:05:25.440 --> 0:05:28.720 our rate card. You know, our average revenues per customer 0:05:28.800 --> 0:05:31.680 per month are rising, not because of rising prices, but 0:05:31.720 --> 0:05:34.800 because they're self selecting up our rate card, and that 0:05:34.839 --> 0:05:38.040 allows revenues to rise where nobody else is the loser. 0:05:38.160 --> 0:05:40.400 People need more data. They're going to need more data 0:05:40.680 --> 0:05:44.080 for Generator of AI. Is that become a choke point for. 0:05:44.040 --> 0:05:44.680 You in some way? 0:05:45.040 --> 0:05:47.080 It's going to be a way for us to showcase 0:05:47.120 --> 0:05:50.880 our differentiation you know, both Oukla and Open Signal have 0:05:51.000 --> 0:05:54.160 once again said we ran the gamut in terms of 0:05:54.200 --> 0:05:56.160 network awards and we have the. 0:05:56.080 --> 0:05:57.680 Most capacity by far. 0:05:57.960 --> 0:06:00.599 And so as the demand for our network on AI 0:06:00.640 --> 0:06:05.320 based applications starts to grow, it'll only showcase t mobiles differentiation. 0:06:05.400 --> 0:06:07.560 I think consumers will really be able to see the difference. 0:06:08.279 --> 0:06:12.080 Will consumers care about when they're out without any signal 0:06:12.200 --> 0:06:15.159 and now they're able potentially to tap in to Starlin 0:06:15.240 --> 0:06:16.120 easing team Mobile. 0:06:16.160 --> 0:06:19.040 When we first announced this partnership in twenty twenty two, 0:06:19.120 --> 0:06:22.000 it was the single most covered thing we've ever talked about. 0:06:22.040 --> 0:06:25.480 And that's saying something because our uncarrier moves are big deals. 0:06:25.800 --> 0:06:28.000 This is the biggest one because it speaks to a 0:06:28.480 --> 0:06:31.279 universal pain point in this industry. This is a big 0:06:31.279 --> 0:06:36.240 country and there are five hundred thousand square miles that 0:06:36.480 --> 0:06:39.640 none of the carriers cover. And so now there's our 0:06:39.720 --> 0:06:42.239 vision is so simple. If you can see the sky, 0:06:42.760 --> 0:06:44.279 you're connected and you don't. 0:06:44.120 --> 0:06:45.880 Have to point your phone at it anymore, which is 0:06:45.920 --> 0:06:48.440 what you had to do with an Apple previously. Why 0:06:48.480 --> 0:06:51.000 did you sort of roll it out so secretly? Why 0:06:51.000 --> 0:06:53.159 did it feel like suddenly we woke up and quietly 0:06:53.160 --> 0:06:55.360 you could get onto your T mobile iPhone device when 0:06:55.360 --> 0:06:56.520 we wouldn't have thought it already. 0:06:56.560 --> 0:06:58.600 Honestly, we've been a little self conscious than that. We 0:06:58.760 --> 0:07:01.080 announced this a long time ago, and so it's been 0:07:01.120 --> 0:07:02.760 a couple of years and we didn't want to keep 0:07:02.800 --> 0:07:04.680 crowing about something until it was ready. 0:07:04.720 --> 0:07:05.640 So we made a big. 0:07:05.480 --> 0:07:07.359 Deal out of it when we struck the partnership and 0:07:07.400 --> 0:07:10.440 we said, look, let's just stay quiet until it's time 0:07:10.480 --> 0:07:13.040 to hand people this service. About two weeks ago we 0:07:13.120 --> 0:07:16.160 quietly started letting people into our beta program. We now 0:07:16.200 --> 0:07:19.000 have thousands of users and we're about to throw the 0:07:19.040 --> 0:07:21.000 doors wide open on our beta. It's going to be 0:07:21.040 --> 0:07:23.720 a big moment and then commercial service this year. 0:07:23.800 --> 0:07:25.720 So it's performing well in beta. 0:07:25.440 --> 0:07:28.800 Absolutely, and right now the launches are happening really rapidly. 0:07:28.840 --> 0:07:31.760 So there's well over four hundred satellites. They're low Earth 0:07:31.880 --> 0:07:34.600 orbit satellites, so they're very performant. You mentioned some of 0:07:34.600 --> 0:07:38.679 the differences versus higher level satellites, and people are getting 0:07:38.680 --> 0:07:39.520 a great experience. 0:07:39.560 --> 0:07:39.720 You know. 0:07:39.760 --> 0:07:40.760 Our goal is to. 0:07:40.760 --> 0:07:44.080 Start with text messaging so that you can get your 0:07:44.080 --> 0:07:47.000 texts out regardless of where you are, but it'll very 0:07:47.080 --> 0:07:50.960 quickly move to picture messaging, MMS, data and voice. 0:07:51.120 --> 0:07:52.880 I hate to try and find any clouds when you're 0:07:52.920 --> 0:07:56.280 pointing to silver linings throughout, But when you signal that 0:07:56.440 --> 0:07:59.240 caution in December, Yes, about sort of the cyclicality, but 0:07:59.320 --> 0:08:03.080 ultimately you've had stellar growth after sellar growth. How much 0:08:03.120 --> 0:08:06.120 can you keep on ramping, keep on bringing on more consumers? 0:08:06.240 --> 0:08:08.440 How strong is the consumer sentiment out there? 0:08:08.640 --> 0:08:10.880 It's steady and consistent. I think that's what you want 0:08:10.920 --> 0:08:12.920 from us. You know, one of the things that differentiates 0:08:13.000 --> 0:08:15.840 us is we set out a set of outlooks and 0:08:15.880 --> 0:08:18.680 then we consistently meet or beat them. And what we 0:08:18.760 --> 0:08:21.600 put out today was a strong guide, but it wasn't 0:08:21.640 --> 0:08:23.720 expecting much from our business that we haven't done in 0:08:23.720 --> 0:08:25.920 the past. And so you can rely on the things 0:08:25.920 --> 0:08:28.200 that we say because we look at the trends underneath it. 0:08:28.480 --> 0:08:30.280 And you know, the bottom line is we offer a 0:08:30.280 --> 0:08:33.280 stronger network value proposition, and we offer it at a 0:08:33.320 --> 0:08:35.319 lower price, and at the end of the day, that's 0:08:35.360 --> 0:08:38.680 going to cause people across segments to continue choosing T Mobile. 0:08:39.760 --> 0:08:42.960 What is it that investors in particular a needing at 0:08:42.960 --> 0:08:44.880 this moment to hear from you? I mean you've got 0:08:44.920 --> 0:08:48.040 twenty two buys, only nine holds and only two cells. 0:08:48.440 --> 0:08:51.120 But where are you hearing the questions of caution coming 0:08:51.120 --> 0:08:52.080 from an investor base right? 0:08:52.080 --> 0:08:53.760 Well, they want to know what our next act is, 0:08:53.960 --> 0:08:55.880 you know, And so that's a big piece of what 0:08:55.920 --> 0:08:58.320 we focused on in our Capital Markets day last fall. 0:08:58.640 --> 0:09:01.360 So talk about what comes be on the next couple 0:09:01.440 --> 0:09:03.200 of years, and you know, AI is a big piece 0:09:03.240 --> 0:09:06.800 of it. We're transforming our business operations to be able 0:09:06.800 --> 0:09:09.480 to think about how do we serve customers in the 0:09:09.559 --> 0:09:12.080 era of AI in new ways? And this industry has 0:09:12.120 --> 0:09:15.479 always treated everybody exactly the same, and it's so analogue. 0:09:15.679 --> 0:09:17.960 You go in and spend a Saturday of your life 0:09:18.040 --> 0:09:20.480 in one of our retail stores trying to upgrade your 0:09:20.480 --> 0:09:21.560 family to new phones. 0:09:21.800 --> 0:09:22.679 We know we can do. 0:09:22.720 --> 0:09:25.080 Better, and our Tea Life app that's now in the 0:09:25.080 --> 0:09:29.120 hands of fifty million people can give you an experience 0:09:29.160 --> 0:09:31.840 of T Mobile that's different than somebody else in terms 0:09:31.880 --> 0:09:35.480 of the offers. You may see, the prioritization, the upgrades, 0:09:35.559 --> 0:09:38.400 et cetera. And so it's an exciting moment for us 0:09:38.440 --> 0:09:41.800 to rethink how we serve customers and give each customer 0:09:41.800 --> 0:09:43.480 a fantastic journey with T Mobile. 0:09:44.320 --> 0:09:46.280 What could the phone make has been doing better at 0:09:46.320 --> 0:09:49.119 the moment because ultimately we see Apple getting another downgrade. 0:09:49.440 --> 0:09:52.400 Many worries about China, but we aren't upgrading our phones 0:09:52.440 --> 0:09:54.160 as much as we used to do. You have to 0:09:54.200 --> 0:09:56.880 capitalize on giving the add ons in this moment because 0:09:56.880 --> 0:09:59.720 Samsung and Apple just they're not presenting as new a 0:10:00.320 --> 0:10:01.480 use cases as they used to. 0:10:01.880 --> 0:10:03.559 Well, it works for our business model. I mean, I 0:10:03.559 --> 0:10:04.280 can't speak for them. 0:10:04.600 --> 0:10:07.320 One thing that's happening is upgrade rates have slowed down, 0:10:07.360 --> 0:10:10.480 but they're pretty steady. Prices have risen, so the OEMs 0:10:10.480 --> 0:10:13.160 have made up for it. That way, customers are keeping 0:10:13.200 --> 0:10:15.600 their phones longer. But what you saw with T Mobile 0:10:15.600 --> 0:10:17.640 this time is our upgrade rates did tick up a 0:10:17.640 --> 0:10:19.880 little bit. With the lowest upgrade rates in the industry. 0:10:19.920 --> 0:10:22.720 That's good for our financials. But they started to normalize, 0:10:22.720 --> 0:10:24.960 so they ticked up. I signaled earlier in the quarter 0:10:25.000 --> 0:10:27.760 that our iPhone sales were very strong, and they were, 0:10:27.760 --> 0:10:28.960 and you saw it in our numbers. 0:10:29.920 --> 0:10:30.960 Mike, so great to have you. 0:10:31.000 --> 0:10:32.559 Great to see you again in the back of your earnings. 0:10:32.600 --> 0:10:41.240 Come back soon. Mike's even is a T Mobile CEO, 0:10:42.520 --> 0:10:45.760 Microsoft and open Ai are investigating where the deep seek 0:10:46.040 --> 0:10:48.880 used data output from open AI's technology if it was 0:10:48.920 --> 0:10:51.880 obtained in an unauthorized manner by a group linked to 0:10:51.920 --> 0:10:55.200 the Chinese AI startups, all according to sources. Now, Deep 0:10:55.240 --> 0:10:57.920 Seek and the High Flyer Hedge fund where deep Seak 0:10:58.000 --> 0:11:01.199 was started, have not commented on there's so far, so 0:11:01.320 --> 0:11:03.800 let's discuss it and the data privacy concerns around deep 0:11:03.800 --> 0:11:07.319 seat with Alex Staemosi's chief mentioned security officer at Centinel 0:11:07.320 --> 0:11:11.240 One and former chief security officer at Facebook. Look Alex 0:11:11.440 --> 0:11:14.280 is open ai and indeed other AI labs at the moment. 0:11:14.320 --> 0:11:16.360 Are they right to be concerned about their ip here? 0:11:17.840 --> 0:11:18.640 I think they are. 0:11:19.160 --> 0:11:22.760 There's a lot of people who doubt the claim that 0:11:23.080 --> 0:11:27.959 deepsek was able to build such a incredibly capable model 0:11:28.559 --> 0:11:31.439 with such a little compute and such as little cost 0:11:31.559 --> 0:11:34.280 as they claim. Now there are some great breakthroughs here. 0:11:34.400 --> 0:11:38.640 Deep Seek published at paper that demonstrated great breakthroughs in 0:11:38.679 --> 0:11:42.439 both training cost and inference costs. So inference is what 0:11:42.480 --> 0:11:44.679 you do when you're actually running the model for people 0:11:44.720 --> 0:11:47.920 to use later, and they have some legitimate breakthroughs. So those 0:11:47.920 --> 0:11:50.760 are breakthroughs that to their credit, they have documentary that 0:11:50.840 --> 0:11:52.480 other people are going to be able to use now. 0:11:52.880 --> 0:11:55.360 But there's a lot of people who believe that they 0:11:55.520 --> 0:12:00.000 utilized both open AI's model probably running up in Microsoft's 0:12:00.040 --> 0:12:03.680 Azure service, as well as metas open source models to 0:12:03.840 --> 0:12:07.800 build upon to build Deepseek, And that's what's being investigated. 0:12:07.320 --> 0:12:11.640 Right now, Alex Is there anything that Microsoft and other 0:12:12.480 --> 0:12:16.080 open and open ai and other providers can do to 0:12:16.160 --> 0:12:19.840 perhaps prevent this from happening again? If not deep Seek, 0:12:19.880 --> 0:12:22.040 then perhaps another competitor from China. 0:12:23.240 --> 0:12:25.440 Well, this is the challenge that they're going to have 0:12:25.520 --> 0:12:29.000 that Microsoft's business model is that you can go to 0:12:29.200 --> 0:12:31.920 Microsoft Azure and you can throw down a credit card, 0:12:31.920 --> 0:12:34.400 you can open up an account, and you can ask 0:12:34.480 --> 0:12:38.079 them to create a private copy of open AI's models 0:12:38.080 --> 0:12:40.200 for you that you can use for your own purposes. 0:12:40.920 --> 0:12:45.559 This is how Microsoft plans on monetizing their multi billion 0:12:45.600 --> 0:12:50.079 dollar investment in open Ai. That is a great thing 0:12:50.120 --> 0:12:52.640 for them, that's a great thing in theory for both 0:12:52.720 --> 0:12:55.240 open Ai and the customers that you can have a 0:12:55.320 --> 0:12:58.719 version of open Ai that in theory does not expose 0:12:59.080 --> 0:13:01.959 their model weights to the customer and does not expose 0:13:02.040 --> 0:13:04.960 the customer's private enterprise data to open AI. 0:13:05.520 --> 0:13:07.920 But what has happened is that people have figured. 0:13:07.640 --> 0:13:11.360 Out ways that you can ask a model many many 0:13:11.480 --> 0:13:14.599 millions or billions of questions to suck all of the 0:13:14.679 --> 0:13:17.840 data out of it. So imagine you have an ancient 0:13:18.760 --> 0:13:22.040 wise professor, and you're able to ask that wise professor many, many, 0:13:22.040 --> 0:13:24.360 many questions and then effectively become a. 0:13:24.360 --> 0:13:25.880 Copy of that professor. 0:13:26.400 --> 0:13:29.280 That's exactly what can happen here, except at machine speed. 0:13:29.720 --> 0:13:33.440 You can do that over days, weeks, or months, and 0:13:33.480 --> 0:13:36.600 that is the business model that Microsoft is operating here. 0:13:36.840 --> 0:13:40.360 So detecting that kind of behavior might be theoretically possible, 0:13:40.640 --> 0:13:45.000 but that's also exactly what they're selling. And so I think, yes, 0:13:45.040 --> 0:13:47.480 they're going to look into what Deepseat did here, but 0:13:47.559 --> 0:13:49.720 it will only be a cat and mouse game to 0:13:49.800 --> 0:13:53.080 try to determine whether somebody is actually utilizing the models 0:13:53.080 --> 0:13:55.160 in the way Microsoft wants them to and wants to 0:13:55.160 --> 0:13:57.600 get paid for, and whether or not they're trying to 0:13:57.640 --> 0:13:58.959 actually copy those models. 0:14:00.240 --> 0:14:01.719 Is we want to ask you a little bit more 0:14:01.760 --> 0:14:05.920 about the detail of concerns anxieties privacy. Teresa Peyton joined US. 0:14:05.920 --> 0:14:08.680 She's former CEO for the White House. Here's what she 0:14:08.720 --> 0:14:10.680 had to say about worries on using deep Seat. 0:14:11.840 --> 0:14:16.720 This is an untested app. I would caution people not 0:14:16.920 --> 0:14:21.640 to put too much proprietary company information into it, and 0:14:22.080 --> 0:14:25.320 until there's been an opportunity to actually do something called 0:14:25.400 --> 0:14:29.200 ethical hacking or red teaming pen testing of the actual 0:14:29.280 --> 0:14:32.040 app and learning more about how is your data treated, 0:14:32.080 --> 0:14:35.240 where is it stored, how do the algorithms work. It 0:14:35.360 --> 0:14:38.080 is open source, but things do still need to be 0:14:38.080 --> 0:14:39.200 put through their paces. 0:14:39.440 --> 0:14:41.720 Of course, deep Seat was the number one app downloaded 0:14:41.760 --> 0:14:45.240 on Apple. We've also now got Ali Baba announcing its 0:14:45.320 --> 0:14:49.320 latest powerful AI model. Should similar reticence be used by 0:14:49.440 --> 0:14:50.200 US companies? 0:14:50.200 --> 0:14:51.360 By US people? Adds? 0:14:52.240 --> 0:14:53.960 Okay, So this is where it gets complicated and we 0:14:54.080 --> 0:14:56.720 got to be really careful here to separate out two 0:14:56.760 --> 0:14:59.680 different things. There is the deep Seak app that people 0:14:59.680 --> 0:15:03.600 are downloading on their phones. That app, that data goes 0:15:03.600 --> 0:15:05.840 straight to China. Right, So if you're downloading deep Seek 0:15:05.880 --> 0:15:07.640 on your phone, or if you're going to deep Seak's 0:15:07.640 --> 0:15:10.800 website and you're typing in query that app, that data 0:15:10.800 --> 0:15:13.760 goes straight to China. That's available to that Chinese company, 0:15:13.920 --> 0:15:17.160 that is probably available to the Chinese government without any problem. 0:15:17.320 --> 0:15:21.600 So yes, absolutely, companies should not be allowing their employees 0:15:21.640 --> 0:15:24.200 to download the deep Seap app onto their phones. If 0:15:24.240 --> 0:15:27.120 you're a company and you download the open source model weights, 0:15:27.600 --> 0:15:30.080 that data does not go back to China. Now there 0:15:30.120 --> 0:15:33.119 are some potential risks right now. 0:15:32.880 --> 0:15:35.720 Those risks are pretty much theoretical. The deep seek model 0:15:35.720 --> 0:15:39.000 weights do have an ideological. 0:15:39.720 --> 0:15:42.440 Twist towards the CCP, so if you asked you about 0:15:42.440 --> 0:15:46.280 Tineman square, it won't answer. It will not create, for example, 0:15:46.680 --> 0:15:49.840 a Winnie the Pooh story about Shishan Pin. 0:15:50.280 --> 0:15:53.000 So it corresponds to the. 0:15:52.920 --> 0:15:57.080 Rules the Chinese government has created about the ideological consistency 0:15:57.080 --> 0:16:01.240 of AI, but that is not a security risk. That 0:16:01.360 --> 0:16:04.520 being said, there are some interesting theoretical ideas people have 0:16:04.600 --> 0:16:07.800 come up with about backdooring AI when it's used for 0:16:07.840 --> 0:16:11.680 security sensitive uses, but right now that is not as 0:16:11.760 --> 0:16:14.240 big of a deal. I think a more interesting question 0:16:14.320 --> 0:16:17.040 for enterprises that are looking to use the deep Seak 0:16:17.080 --> 0:16:19.600 model WAKES is the legal issues. If it turns out 0:16:19.640 --> 0:16:23.240 that deep seek was trained off of metas and open 0:16:23.280 --> 0:16:27.600 AIS models, what does that mean for the legality of 0:16:27.680 --> 0:16:30.560 using deep seeks models. Deep Seak has open source theirs, 0:16:30.720 --> 0:16:33.000 but they do not have the ability to get rid 0:16:33.040 --> 0:16:35.600 of any legal issues that are caused by open AI. 0:16:35.680 --> 0:16:38.320 Of course, open ai is being sued for their use 0:16:38.360 --> 0:16:41.200 of training data that's open that. This is one of 0:16:41.200 --> 0:16:44.920 the issues with AI. The intellectual property issues have not 0:16:45.040 --> 0:16:48.480 been solved. We're dealing with eighteen, nineteenth and early twentieth 0:16:48.480 --> 0:16:52.160 century copyright ideas and we're trying to apply it to 0:16:52.240 --> 0:16:55.800 these twenty first century problems. And it's a real legal 0:16:55.880 --> 0:16:59.520 mess that the legal system is years behind what the 0:16:59.600 --> 0:17:01.320 cutting edges in technology. 0:17:02.400 --> 0:17:06.359 Alex Stamos, Sentinel One Chief Information Security Officer, Thank you. 0:17:07.040 --> 0:17:09.480 Coming up, we'll cover deep Seek's impact on the chip 0:17:09.520 --> 0:17:11.840 world with Andrew Feldman from Sarah Bras. 0:17:12.080 --> 0:17:13.159 This is Bloomberg. 0:17:15.440 --> 0:17:19.320 Not all hardware providers are worried about deep seek advantas. 0:17:19.400 --> 0:17:22.679 The partner of Nvidia that sells chip testing equipment just 0:17:22.840 --> 0:17:26.600 lifted its full year forecast almost forty percent in anticipation 0:17:26.680 --> 0:17:31.159 of elevated spending around AI, and it downplayed the impact 0:17:31.240 --> 0:17:34.560 of deep Seek's big debut. Now let's bring in someone 0:17:34.560 --> 0:17:38.199 else in the field. Andrew Feldman, CEO Sarah Brus, a 0:17:38.240 --> 0:17:43.920 company building computer systems for complex AI deep learning applications. Andrew, 0:17:44.359 --> 0:17:46.840 your company is trying to go toe to toe with 0:17:47.200 --> 0:17:51.040 a big competitor and that is Nvidia. What do you 0:17:51.240 --> 0:17:54.679 draw in terms of an underdog story from what we 0:17:54.720 --> 0:17:56.320 saw the other day with deep Seek? 0:17:57.480 --> 0:18:00.280 Okay, I think this is an enormously exciting moment for 0:18:00.320 --> 0:18:04.680 the community. I think it sends multiple powerful messages. Clearly, 0:18:04.920 --> 0:18:10.320 they used less compute, they used fewer people and produced 0:18:10.359 --> 0:18:14.960 something quite impressive. Now, I think that's a reminder that 0:18:15.119 --> 0:18:17.480 those of us who are underdogs and are competing against 0:18:17.520 --> 0:18:22.879 the eight hundred pound gorillas in the industry, small teams 0:18:22.880 --> 0:18:27.560 of hardworking people with big ideas can produce industry changing results. 0:18:28.040 --> 0:18:29.640 And that's what we do at Cerebras. 0:18:30.280 --> 0:18:35.280 Yeah, Andrew, they used you, they used Cerebras inference offering. 0:18:35.680 --> 0:18:36.480 What did you see? 0:18:36.560 --> 0:18:39.639 What have you taken on in terms of the amount 0:18:39.720 --> 0:18:42.320 of share necessity they need in this moment? 0:18:43.680 --> 0:18:48.160 Sure, So we are now in beta with our deep 0:18:48.200 --> 0:18:51.359 Seak offering. And as your last guest said, there are 0:18:51.359 --> 0:18:53.240 two ways to use deep Seek. You can use it 0:18:53.280 --> 0:18:57.479 through their app, and I think he's absolutely correct that 0:18:57.480 --> 0:18:59.639 that that data you can assume goes right to China. 0:19:00.320 --> 0:19:03.200 Or you can use it through a company like us 0:19:03.240 --> 0:19:06.919 or Perplexity or collection of others that are offering the model, 0:19:07.840 --> 0:19:11.200 and we of course have all sorts of protections for 0:19:11.280 --> 0:19:13.320 your data such that you can be assured it will 0:19:13.359 --> 0:19:18.119 go nowhere. There is tremendous demand for this type of model. 0:19:18.640 --> 0:19:21.600 One of the things that separated this model from previous 0:19:21.640 --> 0:19:26.000 models was that its ability to reason and to do logic, 0:19:26.920 --> 0:19:31.200 and those type of models in particular outperform the other 0:19:31.320 --> 0:19:34.960 forms of models, and they do that by using more 0:19:35.000 --> 0:19:40.360 compute during inference. And this is something that the whole 0:19:40.359 --> 0:19:43.520 industry has been grappling with and we are today the 0:19:43.600 --> 0:19:46.119 fastest of this and what that means for you, the 0:19:46.240 --> 0:19:50.040 user is a better user experience, quicker answers, less waiting, 0:19:51.200 --> 0:19:53.720 and that's what's really moving the customers right now. 0:19:54.840 --> 0:19:57.960 Andrew Cerebras had been planning an IPO, do you have 0:19:58.000 --> 0:20:00.280 an update on that? And we also understand and that 0:20:00.400 --> 0:20:04.360 an interagency panel in Washington known as SIPIUS has been 0:20:04.480 --> 0:20:05.480 reviewing all of this. 0:20:05.600 --> 0:20:10.280 Do you have an update there? I don't, But I. 0:20:10.160 --> 0:20:15.840 Think the deep Seek model was put out right around 0:20:16.119 --> 0:20:17.040 Trump's inauguration. 0:20:17.280 --> 0:20:18.600 I don't think that was an accident. 0:20:19.320 --> 0:20:22.879 I think this was an effort to communicate at the 0:20:22.920 --> 0:20:31.200 geopolitical level to the US that the previous administration's regulations 0:20:31.320 --> 0:20:36.480