WEBVTT - AI Builds Itself as the Spending Boom Accelerates 0:00:02.520 --> 0:00:04.210 Bloomberg Audio Studios. 0:00:04.690 --> 0:00:05.370 Podcasts. 0:00:05.690 --> 0:00:07.050 Radio. News. 0:00:13.530 --> 0:00:17.080 Bloomberg Tech is live from the heart of Silicon Valley 0:00:17.620 --> 0:00:19.900 with Ed Ludlow in San Francisco. 0:00:23.489 --> 0:00:26.050 This is Bloomberg Tech coming up as the global debate 0:00:26.070 --> 0:00:29.520 around AI heats up and Fropic finds its claw chatbots 0:00:29.560 --> 0:00:33.980 leading 26% of its AI research and development, plus SoftBank's 0:00:34.020 --> 0:00:36.979 closing out the week with nearly $ 21 billion in potential 0:00:37.020 --> 0:00:40.460 fresh borrowing as it builds out its AI financing capacity. 0:00:40.820 --> 0:00:43.839 and Crusoe CEO Chase Lockmiller joins us on the race 0:00:43.900 --> 0:00:47.729 to build the infrastructure powering AI. The company closes nearly 0:00:47.729 --> 0:00:50.690 $ 4 billion in fresh funding. We're going to end the 0:00:50.930 --> 0:00:54.170 week much as we started it, talking about Anthropic. Anthropic 0:00:54.230 --> 0:00:57.490 says its Claude Chatbot drives more than a quarter of 0:00:57.510 --> 0:01:01.590 the company's research and development work for AI and collaborates 0:01:01.610 --> 0:01:04.690 with staffers for about 90% of their work. It's the 0:01:04.730 --> 0:01:08.610 clearest indication yet that AI can help speed up development 0:01:09.090 --> 0:01:12.720 of future models. Bloomberg's Shireen Ghaffari has the data. And Shireen, 0:01:12.740 --> 0:01:15.880 just start by explaining the data that Anthropic disclosed. 0:01:17.440 --> 0:01:17.880 That's right. 0:01:17.940 --> 0:01:23.370 So Anthropic is basically measuring how much its own researchers, 0:01:23.430 --> 0:01:26.069 its own R & D department is using AI in 0:01:26.130 --> 0:01:29.370 its work to create new AI models. And what they 0:01:29.430 --> 0:01:33.480 found is that right now, Claude is leading, Claude meaning 0:01:33.620 --> 0:01:38.940 Anthropic's AI, 26% of that AI. 0:01:38.840 --> 0:01:39.540 R & D work. 0:01:40.160 --> 0:01:43.840 Now, that's not the same as AI doing AI. Everything 0:01:43.880 --> 0:01:47.220 by itself, but that is significant. It's one of the 0:01:47.400 --> 0:01:51.210 few kind of numbers that we have quantifying what many 0:01:51.270 --> 0:01:54.910 labs like Anthropic Open AI are now moving toward, which 0:01:55.030 --> 0:01:59.300 is the AI essentially feeding the AI to keep improving itself. 0:02:00.400 --> 0:02:02.340 We're at the end of a seven-day week, even though 0:02:02.380 --> 0:02:06.970 it's Friday, because on Saturday... seven days ago, Dario Amodei 0:02:07.030 --> 0:02:09.970 published his essay. And one of the things he highlighted 0:02:10.010 --> 0:02:12.340 at the top for the rationale for doing it now 0:02:12.960 --> 0:02:17.800 is recursive self-improvement, the idea that AI is making the 0:02:17.840 --> 0:02:20.540 training of future models better. But do you have an 0:02:20.639 --> 0:02:23.000 answer to that, Shireen, why the data that was published 0:02:23.020 --> 0:02:24.480 yesterday came out when it did? 0:02:26.770 --> 0:02:28.630 You know, I think we don't know if there was 0:02:28.690 --> 0:02:30.970 any one specific reason, but I will say the backdrop 0:02:30.990 --> 0:02:34.510 of all of this is while there is obviously huge 0:02:35.310 --> 0:02:39.649 business opportunity in reaching AI that can improve itself with RSI, 0:02:39.669 --> 0:02:44.310 there's also huge concern, anxiety, risk. And we're seeing AI 0:02:44.370 --> 0:02:48.710 lab employees at Anthropic, at OpenAI, starting to raise concerns 0:02:48.770 --> 0:02:52.960 publicly about this. We saw the viral note of Jacob Coxon, 0:02:52.970 --> 0:02:56.620 a former Anthropic and employee researcher saying, saying he doesn't 0:02:56.660 --> 0:03:00.040 think any of the labs are handling this responsibly right now. 0:03:01.120 --> 0:03:04.100 And I think that there's a real call for these 0:03:04.360 --> 0:03:09.890 companies to start quantifying and potentially slowing down by Dario 0:03:09.950 --> 0:03:14.290 Amadei's own essays recognition the pace of this, even if 0:03:14.350 --> 0:03:17.720 that means losing some business momentum. 0:03:18.850 --> 0:03:21.100 Author of the Bloomberg Q & AI newsletter, to which 0:03:21.139 --> 0:03:24.060 you should subscribe, Bloomberg's Shireen Ghaffari. Thank you very much indeed. 0:03:24.280 --> 0:03:27.260 As AI starts helping build the next generation of AI, 0:03:27.660 --> 0:03:30.780 warnings about losing control are growing louder. But critics say 0:03:31.040 --> 0:03:35.250 the focus on existential risk is overshadowing problems that are 0:03:35.310 --> 0:03:38.550 already here. It's the subject of Bloomberg, Rachel Metz and 0:03:38.590 --> 0:03:44.510 Davey Alba, latest reporting and, frankly, great body of work 0:03:44.730 --> 0:03:47.450 trying to cut through the noise of the week. Devi, 0:03:47.490 --> 0:03:50.110 I'm going to start with you. Just summarize what our 0:03:50.170 --> 0:03:54.130 story outlines and take us to the origin, which is 0:03:54.150 --> 0:03:56.950 the Hugging Face incident. Yeah, absolutely. 0:03:57.050 --> 0:04:00.460 So we did a deep dive into sort of the 0:04:00.560 --> 0:04:06.920 Hugging Face incident as an example of how a lot 0:04:06.960 --> 0:04:09.480 of the rhetoric that's been going around right now is 0:04:09.520 --> 0:04:14.200 talking about how a failure in safety and in the 0:04:14.260 --> 0:04:20.990 sandbox not being properly cordoned off, that there wasn't warnings. 0:04:21.270 --> 0:04:25.950 There were warnings that the human employees themselves did not 0:04:27.270 --> 0:04:31.909 take seriously. That is being framed as, look, we can't 0:04:31.930 --> 0:04:32.750 contain this AI. 0:04:32.810 --> 0:04:33.669 It's so dangerous. 0:04:33.730 --> 0:04:34.610 It's going to escape. 0:04:34.670 --> 0:04:38.440 And that being an inevitability and that just, you know, 0:04:38.460 --> 0:04:42.270 sort of works to sidestep the responsibility that these companies 0:04:42.350 --> 0:04:44.890 have to avoid safety failures. 0:04:46.170 --> 0:04:49.830 Hey, Rachel, you know what's weird about AI? What's weird 0:04:49.890 --> 0:04:56.560 about AI? What's weird? Humans, peeps, design the experiment and 0:04:56.600 --> 0:04:59.940 design the safeguards. Like that's the takeaway I got reading 0:04:59.980 --> 0:05:04.140 your piece. Remind everyone that AI came from humans. Run 0:05:04.170 --> 0:05:04.380 with it. 0:05:05.930 --> 0:05:06.170 Yes. 0:05:07.010 --> 0:05:09.659 Yeah, I think that's something that I've been thinking about 0:05:09.860 --> 0:05:11.700 for a while. Since we heard about the hugging face 0:05:12.120 --> 0:05:15.800 inadvertent hack and talking to people about it, I mean, 0:05:15.940 --> 0:05:18.400 this comes up quite a lot, that humans were very 0:05:18.440 --> 0:05:19.500 much involved with this. 0:05:19.560 --> 0:05:22.810 I mean, humans set goals for AI. 0:05:22.910 --> 0:05:27.570 Humans set up the situation here in the sandbox where 0:05:27.670 --> 0:05:30.070 the AI is working on things individually. 0:05:30.330 --> 0:05:32.600 And humans are training. I mean, there are humans up 0:05:32.640 --> 0:05:32.920 and down. 0:05:33.100 --> 0:05:36.820 Humans also noticed some situations where they could have stepped 0:05:36.920 --> 0:05:39.570 in during evaluations when the AI wasn't doing quite what 0:05:39.589 --> 0:05:40.290 it should have been doing. 0:05:40.770 --> 0:05:41.590 And did not. 0:05:41.710 --> 0:05:44.210 And OpenAI has acknowledged all of these things. And to 0:05:44.250 --> 0:05:47.070 its credit, it said quite a lot about this incident 0:05:47.120 --> 0:05:49.600 and also what it's doing to change things over there. 0:05:49.980 --> 0:05:52.200 But I think that that came up a lot with 0:05:52.240 --> 0:05:58.500 this story, especially as we're talking to people ongoing, that 0:05:58.640 --> 0:06:00.900 people have a lot to do with this situation. 0:06:01.000 --> 0:06:03.650 It's not as simple as saying, oh, well, AI went 0:06:03.710 --> 0:06:04.349 and did something. 0:06:05.520 --> 0:06:08.200 Rachel, Davey, we have some breaking news. Gavin Newsom, the 0:06:08.260 --> 0:06:11.680 governor of California, has issued an executive order on the 0:06:11.760 --> 0:06:17.630 creation of an AI kill switch. The order accelerates what 0:06:17.690 --> 0:06:21.210 he says will be independent oversight of AI. That story 0:06:21.250 --> 0:06:24.870 just hitting the terminal right now. Davey, this is difficult 0:06:24.890 --> 0:06:29.210 because it's fluid, right? It's just hitting the newswire. You 0:06:29.290 --> 0:06:32.549 talk as well in your parallel newsletter about this idea 0:06:32.589 --> 0:06:36.650 that there is... plenty of elements of human control. Hugging 0:06:36.710 --> 0:06:39.720 Face was the case study for that. Take us there. 0:06:40.640 --> 0:06:41.300 Yeah, absolutely. 0:06:41.380 --> 0:06:43.859 And, you know, with regard to this AI kill switch, 0:06:43.920 --> 0:06:48.200 we also cover in our story the language that federal 0:06:48.240 --> 0:06:53.990 lawmakers have started to use to create policy that would affect, 0:06:54.270 --> 0:06:58.550 you And one thing that we point out is a 0:06:58.730 --> 0:07:01.410 lot of this sort of rhetoric of doom is migrating 0:07:01.510 --> 0:07:06.550 from the AI industry into the rhetoric of these lawmakers. 0:07:06.610 --> 0:07:09.150 So I would want to take a look at how 0:07:09.630 --> 0:07:14.160 Newsom is framing this kill switch. One thing that the 0:07:14.200 --> 0:07:19.140 researchers that we spoke to said was that when you 0:07:19.200 --> 0:07:23.240 start from a place of doom, sort of this assumption 0:07:23.300 --> 0:07:27.980 that AI is going to threaten all humanity, the policies 0:07:28.020 --> 0:07:31.940 may not match the actual harms that AI is causing 0:07:31.980 --> 0:07:37.930 right now. So when we're talking about misinformation, surveillance, these 0:07:37.990 --> 0:07:42.870 things that are happening right now, that is what is 0:07:42.910 --> 0:07:45.450 sort of getting lost in the focus on doom. 0:07:47.100 --> 0:07:49.420 Rachel, on the show in the last 24 hours, there 0:07:49.440 --> 0:07:52.540 have been plenty of voices. Let's take researcher Andrew Ng 0:07:52.940 --> 0:07:57.060 or Databricksio Aligotzi saying this is way overstated. The risk 0:07:57.320 --> 0:08:00.470 to human life, to their mind, is near zero. You're 0:08:00.490 --> 0:08:02.970 up in Montreal right now, right? Who have you been 0:08:02.990 --> 0:08:06.510 speaking to? What are you hearing on the kind of counter-argument, 0:08:06.750 --> 0:08:09.800 the anti-doomerism school of thought? 0:08:11.200 --> 0:08:13.550 Yeah, I mean, I think up here I'm hearing a 0:08:13.690 --> 0:08:17.450 little bit of both. I mean, I spoke to Yoshua Bengio, 0:08:18.250 --> 0:08:20.590 who is a legendary figure in AI and has long 0:08:21.010 --> 0:08:26.100 talked about what he sees as very real harms that 0:08:26.120 --> 0:08:29.800 may come from AI. And so he sees this situation, 0:08:29.820 --> 0:08:32.290 the Hugging Face incident in particular, as real. When I 0:08:32.330 --> 0:08:34.030 spoke to him yesterday, it was sort of a mix 0:08:34.070 --> 0:08:36.809 of a human and machine problem, and he's very, very 0:08:36.850 --> 0:08:39.710 concerned about what may happen going forward. On the other hand, 0:08:39.750 --> 0:08:43.790 I've spoken to other people, such as the CEO of Cohere, 0:08:43.929 --> 0:08:47.660 Aidan Gomez, also a very key figure in AI, and 0:08:47.679 --> 0:08:48.000 he had. 0:08:49.860 --> 0:08:51.620 A thought that was a little bit more like, look. 0:08:51.860 --> 0:08:54.700 I don't think that this is going to kill people 0:08:54.880 --> 0:08:57.719 or kill us en masse, but there are some really 0:08:58.200 --> 0:09:01.780 real issues here, and those do need to be worked on. 0:09:01.900 --> 0:09:05.380 And maybe slowing down certain things could be helpful, but 0:09:05.520 --> 0:09:10.980 it's not a situation where we're at imminent risk of 0:09:11.280 --> 0:09:12.400 all being extinct. 0:09:14.250 --> 0:09:17.709 Bloomberg's Rachel Metz and Davey Alba, authors of the report 0:09:17.770 --> 0:09:18.849 Anthropics Warning. 0:09:20.460 --> 0:09:24.120 Of existential risk hijacks large AI debate. It's a must 0:09:24.179 --> 0:09:26.580 read on Bloomberg. Thank you both very much. We are 0:09:26.600 --> 0:09:28.900 going to keep this conversation going. Sara Arahi is the 0:09:28.940 --> 0:09:33.420 Franklin Templeton portfolio manager and partner with Franklin Ventures and 0:09:33.540 --> 0:09:39.070 joins us now. And Franklin Ventures, also an investor in Anthropic. Okay, Friday, 0:09:39.470 --> 0:09:41.189 net net, where have we ended this week? 0:09:42.470 --> 0:09:46.030 Nice ending, what Shireen just talked about, giving us data 0:09:46.090 --> 0:09:51.770 about the 26%. 0:09:49.860 --> 0:09:51.200 Agents in the R & D work. 0:09:51.260 --> 0:09:54.340 I think that's a good ending versus the doomerism started 0:09:54.400 --> 0:09:56.650 at the beginning of the week. I'm definitely in the 0:09:56.750 --> 0:10:03.320 anti-doomerism camp. I think that I'm appreciative that these founders 0:10:03.380 --> 0:10:06.740 are talking about guardrails. They're talking about making sure that 0:10:06.880 --> 0:10:11.940 everything is safely done. But I also think we. 0:10:11.880 --> 0:10:13.640 Shouldn't be slowing down our pace here. 0:10:13.980 --> 0:10:16.470 And even if we did on the pre-training side or 0:10:16.490 --> 0:10:18.810 the training side where there's a lot of expenses on 0:10:18.850 --> 0:10:19.290 that front. 0:10:19.929 --> 0:10:23.160 Inference is going to continue. And that means the spend 0:10:23.179 --> 0:10:23.820 will continue. 0:10:23.940 --> 0:10:26.140 And as a function of that, you're going to see, 0:10:26.620 --> 0:10:29.100 you know, a number of companies that pulled back, you know, 0:10:29.140 --> 0:10:31.400 stocks in the equity market pulled back on Monday with 0:10:31.520 --> 0:10:33.260 fears of, oh, slow down spending. 0:10:33.620 --> 0:10:35.520 That spending can't slow, especially. 0:10:35.660 --> 0:10:37.900 This is why I'm grateful to have you here. You 0:10:37.960 --> 0:10:40.040 are a public market investor and you are a private 0:10:40.059 --> 0:10:42.260 market investor. And so, like, let's go to public markets 0:10:42.300 --> 0:10:45.650 for a second. The idea was if the labs are 0:10:45.730 --> 0:10:48.870 pacing at the frontier, and that pre-training or training work 0:10:48.929 --> 0:10:51.530 is impacted, the question posed, does that mean a pullback 0:10:51.590 --> 0:10:55.110 on the spend for the compute? The net conclusion on 0:10:55.170 --> 0:10:55.770 that is. 0:10:56.010 --> 0:11:00.130 No, because inference requires a lot of compute. You've seen 0:11:00.200 --> 0:11:02.340 a number of deals that these companies have done the 0:11:02.380 --> 0:11:05.080 last few months to secure compute deals. Do you think 0:11:05.120 --> 0:11:06.580 they would have done that if they thought they were 0:11:06.600 --> 0:11:08.880 going to pull back, if pacing was going to pull 0:11:08.920 --> 0:11:11.620 back spend? We don't think that would have happened. And 0:11:11.660 --> 0:11:15.660 then it's interesting because, yes, that narrative was being talked 0:11:15.700 --> 0:11:19.670 about on Monday. I was at an industrials conference on Tuesday, 0:11:19.950 --> 0:11:22.790 and at that conference, a number of these infrastructure companies 0:11:22.830 --> 0:11:25.329 that are exposed to data center were saying, our businesses 0:11:25.370 --> 0:11:26.170 are diversified. 0:11:26.429 --> 0:11:27.849 We have non-data center. 0:11:27.610 --> 0:11:30.220 Exposure as well because they saw the reaction in the market. 0:11:30.550 --> 0:11:32.620 But on Wednesday, some of that had already cleared up. 0:11:33.000 --> 0:11:35.580 Both businesses are going to continue to grow. 0:11:36.220 --> 0:11:37.900 It's really hard to frame. There are lots of different 0:11:37.980 --> 0:11:41.840 camps on the right course of action. All of them 0:11:41.880 --> 0:11:46.040 accept that the capabilities of the models are improving and 0:11:46.220 --> 0:11:49.199 outpacing the rate at which alignment and safety keep up. 0:11:49.240 --> 0:11:53.060 That seems to be a point of agreement. What Dario 0:11:53.100 --> 0:11:57.380 Amodei advocated for in his essay on Saturday was independent evaluation, 0:11:57.980 --> 0:12:01.800 but also the idea that antitrust waivers are appropriate so 0:12:01.870 --> 0:12:05.270 that that collection of frontier labs can work together. How 0:12:05.290 --> 0:12:06.120 do you feel about that? 0:12:06.260 --> 0:12:09.240 Yeah, look, I don't know if they really want widespread 0:12:09.340 --> 0:12:11.080 regulation and federal regulation. 0:12:11.120 --> 0:12:12.439 You don't think that they want that? 0:12:12.500 --> 0:12:13.559 I don't think they want that. 0:12:13.740 --> 0:12:17.590 I do think that they want good actors here and 0:12:18.250 --> 0:12:21.089 every company to put guardrails in place and make sure 0:12:21.150 --> 0:12:24.369 that everything is done in a safe way, in a sandbox. 0:12:24.490 --> 0:12:27.630 And we've seen examples of that, but that hasn't happened, unfortunately. 0:12:27.650 --> 0:12:29.780 Which is the school of thought, right? You're the frontier labs, 0:12:29.800 --> 0:12:31.980 you're the experts, you should have a handle on this. 0:12:32.000 --> 0:12:32.070 Yes. 0:12:32.620 --> 0:12:34.700 And don't release the model if it's not safe. 0:12:34.880 --> 0:12:35.360 Exactly. 0:12:35.780 --> 0:12:38.460 On top of that, though, some are talking about the 0:12:38.500 --> 0:12:41.760 fact that there are a number of security products that 0:12:41.800 --> 0:12:44.089 are going to be launched very soon. And so we 0:12:44.309 --> 0:12:47.849 can also protect you. You're going to see some of 0:12:47.890 --> 0:12:50.170 those product launches come again. A couple of weeks ago 0:12:50.190 --> 0:12:52.730 at a tech conference, the entire theme at that tech 0:12:52.770 --> 0:12:56.060 conference was cybersecurity. Right. Whether it's the public companies exposed 0:12:56.100 --> 0:13:00.059 to this or the venture community was talking about what 0:13:00.080 --> 0:13:01.660 some of the frontier labs are going to start to 0:13:01.740 --> 0:13:02.800 offer and a number. 0:13:02.620 --> 0:13:04.360 Of private companies. So can you help us understand this? 0:13:04.380 --> 0:13:06.439 I think you heard Ali Ghodsi, the Dedebrick CEO, on 0:13:06.460 --> 0:13:09.870 the show yesterday. He went cybersecurity straight away. Andrew Ng 0:13:09.929 --> 0:13:11.370 also brought it up. Loads of people brought it up. 0:13:11.390 --> 0:13:15.070 But the idea is that it means very strong cybersecurity 0:13:15.090 --> 0:13:17.330 platforms are more important than ever. 0:13:17.610 --> 0:13:18.790 They're incredibly important. 0:13:18.830 --> 0:13:22.000 Not only are they important, but also just pacing these models. 0:13:22.030 --> 0:13:26.040 You haven't seen widespread adoption of all these models right 0:13:26.080 --> 0:13:30.040 now and outside of Anthropic talking about it right here, 0:13:30.080 --> 0:13:30.640 right now, right? 0:13:30.679 --> 0:13:31.040 Yes. 0:13:31.380 --> 0:13:32.400 So corporations are. 0:13:32.320 --> 0:13:35.939 Going to start, you know, you have GPT-5, GPT-6, you know, 0:13:36.000 --> 0:13:39.560 Fable coming to market. Companies haven't, knowledge workers are not 0:13:39.600 --> 0:13:42.370 even used to using these right now. Just with the 0:13:42.410 --> 0:13:45.650 launch of Muse, Instinct, some of these agents that consumers 0:13:45.690 --> 0:13:48.550 are starting to work with, I think that will bring 0:13:48.650 --> 0:13:51.810 some adoption on the consumer side, and consumers are going 0:13:51.830 --> 0:13:53.970 to start asking for that in the workplace, just like 0:13:54.010 --> 0:13:57.730 you saw with the iPhone. It took a while for 0:13:57.770 --> 0:14:01.000 consumers to move from BlackBerrys to iPhones, right? But at 0:14:01.030 --> 0:14:03.600 the end of the day, it was they started seeing 0:14:03.660 --> 0:14:05.620 what kind of benefit they get from this, and then 0:14:05.640 --> 0:14:08.520 they started asking for it in their workplace. 0:14:08.600 --> 0:14:11.079 Sarah, I appreciate there's a limit to what you can 0:14:11.160 --> 0:14:14.780 say as an existing investor in Anthropic, but a question 0:14:14.820 --> 0:14:18.820 posed by the market is, Are these safety concerns the 0:14:18.860 --> 0:14:24.420 real reason why OpenAI will not do an IPO until 2027, 0:14:24.420 --> 0:14:27.160 which Sam Altman said in an interview? Actually, there was 0:14:27.220 --> 0:14:30.900 plenty of reporting that the IPO would be 2027 anyway. 0:14:31.280 --> 0:14:33.640 But are the safety concerns a real reason for the 0:14:33.760 --> 0:14:36.700 market to slow down and say, like, we need to 0:14:36.760 --> 0:14:39.280 stop funding this in private or public markets? 0:14:39.790 --> 0:14:41.310 Again, I don't think so. 0:14:42.170 --> 0:14:45.860 We're starting to see some of these benefits, the productivity 0:14:45.920 --> 0:14:49.430 gains from these models, the benefits that society will see, 0:14:49.570 --> 0:14:53.000 not the... 10% or wherever that number comes from in 0:14:53.060 --> 0:14:54.840 terms of doomerism. 0:14:54.860 --> 0:14:56.420 The economic upside. 0:14:56.540 --> 0:14:57.040 The upside. 0:14:57.640 --> 0:15:00.780 We're starting to see that diffuse across a variety of industries. 0:15:00.800 --> 0:15:03.240 You've seen the market start to broaden as a result 0:15:03.260 --> 0:15:05.800 of that. So we shouldn't stop funding. And I think 0:15:06.490 --> 0:15:09.450 IPO timelines, I think some of them were already being 0:15:09.510 --> 0:15:11.830 pushed out to 2027. And I think we're going to 0:15:11.870 --> 0:15:13.510 start to see some of these come through. 0:15:13.770 --> 0:15:15.890 Sarah O'Rourke here, Franklin Templeton. It's really great to have 0:15:15.910 --> 0:15:16.630 you back on the show. 0:15:16.650 --> 0:15:16.880 Thank you. 0:15:16.910 --> 0:15:18.130 Happy Friday, by the way. 0:15:18.150 --> 0:15:18.850 Happy Friday. 0:15:18.890 --> 0:15:21.450 Coming up, SoftBank is closing out the week with fresh 0:15:21.490 --> 0:15:24.270 firepower to back its AI bets and maybe on the 0:15:24.320 --> 0:15:26.300 road to more. We're going to go talk about the 0:15:26.320 --> 0:15:29.380 Japanese conglomerate's massive new debt hole. That's next. This is 0:15:29.400 --> 0:15:41.330 Bloomberg Tech. Let's take a look at today's big number, 0:15:41.330 --> 0:15:45.430 $ 21 billion. That's how much fresh debt SoftBank's potentially closing 0:15:45.470 --> 0:15:48.070 the week out with. Bloomberg's Peter Elstrom, who leads our 0:15:48.090 --> 0:15:51.850 coverage of Asian technology companies, joins us now. It is 0:15:51.890 --> 0:15:54.780 an ongoing story. It's an almost daily story. What's new 0:15:54.840 --> 0:15:56.460 about this SoftBank borrowing? 0:15:57.800 --> 0:16:01.020 Yeah, it's worth slowing down here to just pause and 0:16:01.080 --> 0:16:04.280 understand exactly what Masayoshi Son is doing with SoftBank at 0:16:04.300 --> 0:16:06.780 this point. As you pointed out, he's got more than 0:16:06.780 --> 0:16:10.880 $ 20 billion in new loans that he's securing in order 0:16:10.900 --> 0:16:13.490 to make this investment in OpenAI. He said he's going 0:16:13.510 --> 0:16:16.590 to invest about $ 65 billion. So this is three different 0:16:16.630 --> 0:16:19.260 slices of money that he's coming up with. One, he's 0:16:19.720 --> 0:16:22.460 increasing the amount of money that he's borrowing against his 0:16:22.660 --> 0:16:25.990 arm holding. He holds about 90%. of Arm, the chip maker. 0:16:26.090 --> 0:16:29.450 He's also going to borrow money from Apollo. He's increasing 0:16:29.490 --> 0:16:32.170 the size there. And he's also getting a loan from 0:16:32.170 --> 0:16:35.040 20 banks that he's securing all together. Add it up, 0:16:35.400 --> 0:16:38.020 and he's borrowing about 45 to 50 billion dollars out 0:16:38.040 --> 0:16:40.920 of the 65 billion dollars that he's gonna put into OpenAI. 0:16:41.200 --> 0:16:43.120 And that's not the end of the leverage. He's actually 0:16:43.180 --> 0:16:46.840 borrowing money against that OpenAI stake too. That's another 10 0:16:46.840 --> 0:16:49.140 billion dollars or so. So this is sort of Masa 0:16:49.160 --> 0:16:52.080 doing Masa things, building a lot of leverage on top 0:16:52.120 --> 0:16:54.250 of these startup bets that he's making. 0:16:55.240 --> 0:16:58.550 So I think a lot of people might pose the 0:16:58.570 --> 0:17:02.190 reasonable question, what is SoftBank? Like, okay, we know that 0:17:02.210 --> 0:17:06.160 there is Masayoshi San, there's SoftBank, they have exposure to OpenAI, 0:17:06.180 --> 0:17:07.379 just explain the basics. 0:17:08.880 --> 0:17:12.730 Well, SoftBank has been through many iterations since Masayoshi Son 0:17:12.830 --> 0:17:16.710 founded the company. It's done everything from distribute software in 0:17:16.730 --> 0:17:19.230 the past. It ran tech trade shows. It had a 0:17:19.250 --> 0:17:21.390 bunch of tech magazines in the past. But right now, 0:17:21.800 --> 0:17:24.680 it's primarily an investment vehicle sitting on top of a 0:17:24.720 --> 0:17:27.260 telecom operation that has thrown off a bunch of cash. 0:17:27.700 --> 0:17:28.200 In the past. 0:17:28.220 --> 0:17:30.500 SoftBank Corp, he spun off a few years ago. So 0:17:30.520 --> 0:17:33.140 it's an independent company, but it still generates a lot 0:17:33.180 --> 0:17:35.100 of the profit. And he's been using that money to 0:17:35.140 --> 0:17:37.369 make a bunch of different bets. He, of course, started 0:17:37.430 --> 0:17:41.310 up the Vision Fund that fatally put money into WeWork. 0:17:41.630 --> 0:17:44.169 But it's also now making other investments. And really, the 0:17:44.210 --> 0:17:47.210 big bet that he's making is OpenAI. He's talked about 0:17:47.570 --> 0:17:49.470 how important he thinks AI is going to be in 0:17:49.490 --> 0:17:51.609 the future. And he believes that Sam Altman is going 0:17:51.630 --> 0:17:53.680 to help lead that. So he's Put so much money 0:17:53.700 --> 0:17:56.280 into the company, even people inside a soft bank have 0:17:56.340 --> 0:17:58.920 raised questions about how risky that is and if he's 0:17:59.000 --> 0:18:01.710 concentrated his bets too much on a company that now, 0:18:01.730 --> 0:18:03.810 of course, as we've talked about on this show, they're 0:18:03.850 --> 0:18:06.560 pushing back their IPO. So you've got him borrowing more 0:18:06.600 --> 0:18:10.020 money as interest rates are rising. for an investment in 0:18:10.060 --> 0:18:13.200 a company who's pushing back its IPO, pushing back that 0:18:13.270 --> 0:18:14.649 day when he's going to be able to have a 0:18:14.710 --> 0:18:18.190 liquid publicly traded stock to use as collateral. 0:18:18.270 --> 0:18:18.490 Right. 0:18:19.050 --> 0:18:21.490 Bloomberg's Peter Elstrom, happy Friday to you. Thank you very 0:18:21.530 --> 0:18:23.970 much indeed. Now coming up, one of the year's biggest 0:18:24.010 --> 0:18:28.710 AI trades is reversing as software stocks stage a comeback 0:18:29.060 --> 0:18:32.899 and chips slide. Breakdown what's driving the turnaround. This is 0:18:32.920 --> 0:18:46.939 Bloomberg Tech. Wall Street Week goes to India to see 0:18:46.960 --> 0:18:50.140 what the fastest growing large economy offers to investors from 0:18:50.160 --> 0:18:53.800 the tech center of Bangalore to Genome Valley in Hyderabad. 0:18:54.220 --> 0:18:56.820 The economy is growing on the back of its tech sector, 0:18:56.840 --> 0:18:59.840 but its future hinges on its ability to keep up 0:18:59.880 --> 0:19:01.040 in the global AI race. 0:19:02.260 --> 0:19:05.140 So I think it's both a challenge and an opportunity. 0:19:05.280 --> 0:19:08.600 I think the challenge lies in the fact that India 0:19:08.680 --> 0:19:14.310 has had great strength in IT services and providing a 0:19:14.369 --> 0:19:19.790 lot of functions, white collar services at very attractive costs. 0:19:20.910 --> 0:19:24.670 Some of that will get replaced by AI. And so 0:19:24.910 --> 0:19:28.540 I think the India bears say, ah, that's going to happen. 0:19:29.119 --> 0:19:33.440 I think what they fail to recognize is that These 0:19:33.520 --> 0:19:36.200 companies will evolve. We own a large stake in a 0:19:36.240 --> 0:19:39.770 business called Emphasis, an IT services company, and they have 0:19:39.850 --> 0:19:45.270 pivoted their business to focus on AI deployment because changing 0:19:45.350 --> 0:19:49.090 companies utilizing AI is also going to require an enormous 0:19:49.130 --> 0:19:54.040 amount of manpower. I also think they underestimate the willingness 0:19:54.400 --> 0:19:57.340 of people in India to utilize the AI and what 0:19:57.380 --> 0:20:00.300 it will do there. The creation of new companies there, 0:20:00.660 --> 0:20:03.090 the creation of AI infrastructure. 0:20:04.060 --> 0:20:05.880 If I look back two decades ago, we were like 0:20:05.920 --> 0:20:08.600 a call center. That call center has gone to a 0:20:08.740 --> 0:20:11.710 back office. The back office went to a mid office. 0:20:12.270 --> 0:20:15.010 And the mid office now is literally a front office 0:20:15.770 --> 0:20:19.010 because tech is really part of the business, right? And 0:20:19.109 --> 0:20:23.410 that's why I feel, despite AI, we're doing very well. 0:20:23.470 --> 0:20:26.460 The demand's very robust. And we look forward to a 0:20:26.540 --> 0:20:28.780 really strong rest of the decade. 0:20:29.770 --> 0:20:31.490 You can watch the full episode of Wall Street Week 0:20:31.550 --> 0:20:35.689 today at 6 p.m. Eastern, 3 p.m. Pacific. Okay, software 0:20:35.710 --> 0:20:39.330 stocks are making a comeback. Strong earnings and warnings about 0:20:39.350 --> 0:20:42.260 the pace of AI development are reversing one of this 0:20:42.280 --> 0:20:46.280 year's biggest trades, long chips, short software. Okay, this Friday's 0:20:46.320 --> 0:20:49.540 session might not reflect that necessarily, but Bloomberg's Ryan Veselica 0:20:49.820 --> 0:20:51.780 has one of the best read stories on the Bloomberg. 0:20:52.280 --> 0:20:56.840 This happened in June and... Today, it's the same, right? 0:20:56.880 --> 0:20:59.790 The software doomerism is also saspocalypse fading. 0:21:01.470 --> 0:21:04.369 Yeah, there's been a real reversal of sentiment. The first 0:21:04.410 --> 0:21:07.270 half of the year was really marked by this concern 0:21:07.310 --> 0:21:10.040 that AI was going to come in and decimate the 0:21:10.100 --> 0:21:12.940 legacy software industry, which is why we saw the chip 0:21:12.980 --> 0:21:16.399 sector do so well and the software sector really just collapsed. 0:21:16.420 --> 0:21:19.040 Since then, we have seen a bit of a reversal. 0:21:19.080 --> 0:21:23.110 We've seen some narrowing of this performance gap. Most recently, 0:21:23.170 --> 0:21:26.650 we had a very strong earnings season for the software group, 0:21:26.750 --> 0:21:29.689