1 00:00:00,280 --> 00:00:07,240 Speaker 1: Bloomberg Audio Studios, podcasts, radio News. 2 00:00:08,080 --> 00:00:11,000 Speaker 2: Hey, I got to say that there are moments in 3 00:00:11,039 --> 00:00:13,640 Speaker 2: time like today, with all the spend and build out 4 00:00:13,680 --> 00:00:18,120 Speaker 2: and companies doing things related to AI and questions around 5 00:00:18,160 --> 00:00:20,520 Speaker 2: the return on investment, the ROI and the impact of 6 00:00:20,680 --> 00:00:23,720 Speaker 2: artificial intelligence, that we kind of wish we could pull 7 00:00:23,760 --> 00:00:25,480 Speaker 2: out that crystal ball and see a little bit into 8 00:00:25,480 --> 00:00:25,880 Speaker 2: the future. 9 00:00:26,000 --> 00:00:27,880 Speaker 1: I think it's fair to say Chris Miller saw something 10 00:00:28,080 --> 00:00:31,320 Speaker 1: years ago. He's the author of the twenty twenty two 11 00:00:31,480 --> 00:00:34,640 Speaker 1: bestseller Chip War, The Fight for the World's Most Critical Technology. 12 00:00:34,840 --> 00:00:38,320 Speaker 1: He joins us from Pittsburgh this afternoon. Chris, it's good 13 00:00:38,320 --> 00:00:39,760 Speaker 1: to have you back with us. Carol and I were 14 00:00:39,760 --> 00:00:43,040 Speaker 1: just talking, and you know, you wrote this book close 15 00:00:43,080 --> 00:00:47,400 Speaker 1: to four years ago. At this point, how different is 16 00:00:47,440 --> 00:00:52,280 Speaker 1: the world of semiconductors and the world's reliance on semiconductors 17 00:00:52,640 --> 00:00:54,720 Speaker 1: now than when you wrote the book. 18 00:00:57,480 --> 00:00:59,680 Speaker 3: Well, I think the key difference is just the scale 19 00:00:59,720 --> 00:01:02,640 Speaker 3: of spending that we've got right now on building data 20 00:01:02,680 --> 00:01:06,199 Speaker 3: centers and buying all of the chips that are required 21 00:01:06,400 --> 00:01:09,160 Speaker 3: inside of them, and Nvidia first and foremost, but not only, 22 00:01:09,200 --> 00:01:12,440 Speaker 3: we've seen companies transformed by the data center build out 23 00:01:12,600 --> 00:01:14,440 Speaker 3: but in a lot of ways, not much just changed 24 00:01:14,480 --> 00:01:18,360 Speaker 3: other than that we're still sourcing our key semi conductors 25 00:01:18,400 --> 00:01:22,160 Speaker 3: from Asia and above all from Taiwan, and so the 26 00:01:22,200 --> 00:01:26,560 Speaker 3: industry is supercharged, its size, growing faster than ever, but 27 00:01:26,600 --> 00:01:29,000 Speaker 3: dealing with some of the same supply chain choke points 28 00:01:29,000 --> 00:01:30,280 Speaker 3: that it had a decade ago. 29 00:01:30,680 --> 00:01:34,600 Speaker 2: You know, we talk a lot with one of our 30 00:01:34,600 --> 00:01:37,600 Speaker 2: brilliant voices on the chip world, and that is our 31 00:01:37,640 --> 00:01:39,800 Speaker 2: own Ian King, who has seen a lot of cycles, 32 00:01:39,840 --> 00:01:41,559 Speaker 2: the ups, the downs, and we talk a lot about 33 00:01:41,560 --> 00:01:44,720 Speaker 2: that supply demand imbalance and what happens when there's a 34 00:01:44,760 --> 00:01:47,760 Speaker 2: lot of demand, there's not enough supply. The buildout continues, 35 00:01:47,800 --> 00:01:50,600 Speaker 2: the investment supply goes up, and then there's a glut. 36 00:01:50,960 --> 00:01:54,000 Speaker 2: So I'm just curious what your view is when it 37 00:01:54,040 --> 00:01:58,160 Speaker 2: comes to the cyclical nature of the semiconductor world. Has 38 00:01:58,240 --> 00:02:02,080 Speaker 2: AI changed it or we we'll see also and over 39 00:02:02,160 --> 00:02:04,600 Speaker 2: build and oversupply and then a glut. 40 00:02:06,480 --> 00:02:09,400 Speaker 3: I think if you look historically, certainly you see cycles up, 41 00:02:09,440 --> 00:02:12,840 Speaker 3: cycles down, but there are moments when you see step 42 00:02:12,960 --> 00:02:16,680 Speaker 3: changes in terms of demand for certain types of chips. 43 00:02:16,720 --> 00:02:19,320 Speaker 3: We saw that with smartphones, for example, there was no 44 00:02:19,360 --> 00:02:22,120 Speaker 3: demand for smartphone chips and then now everyone needs a 45 00:02:22,120 --> 00:02:24,800 Speaker 3: new smartphone every couple of years, and what we're seeing 46 00:02:24,800 --> 00:02:26,840 Speaker 3: in AI right now is that type of step change, 47 00:02:26,880 --> 00:02:29,840 Speaker 3: a huge increase in just the baseline amount of chips 48 00:02:29,840 --> 00:02:32,680 Speaker 3: that we're going to need for data centers driven by AI, 49 00:02:32,840 --> 00:02:34,720 Speaker 3: and so I think we shouldn't expect cycles to be 50 00:02:34,760 --> 00:02:37,560 Speaker 3: over Certainly there'll be ups and downs in the future, 51 00:02:37,680 --> 00:02:40,359 Speaker 3: but it's now very clear. I think that we're just 52 00:02:40,400 --> 00:02:43,359 Speaker 3: going to need a lot more compute for AI purposes 53 00:02:43,400 --> 00:02:45,040 Speaker 3: in the future, and as a result, will need a 54 00:02:45,040 --> 00:02:46,880 Speaker 3: lot more of the AI chips that go inside of 55 00:02:46,919 --> 00:02:47,480 Speaker 3: data centers. 56 00:02:47,919 --> 00:02:51,640 Speaker 1: In your view, is the promise that many people think 57 00:02:52,080 --> 00:02:57,800 Speaker 1: AI is supposed to deliver, Will it actually be delivered? Like, 58 00:02:57,880 --> 00:03:02,000 Speaker 1: what's the future that you envision after all this capex 59 00:03:02,080 --> 00:03:02,440 Speaker 1: is spent? 60 00:03:04,600 --> 00:03:06,880 Speaker 3: Why? I think it's in some ways funny when we 61 00:03:06,919 --> 00:03:09,480 Speaker 3: ask will AI deliver? Because if you look just three 62 00:03:09,560 --> 00:03:12,400 Speaker 3: years ago, even after chat schipt was released, there are 63 00:03:12,440 --> 00:03:16,960 Speaker 3: so many things that were not possible that are possible today, 64 00:03:17,280 --> 00:03:20,920 Speaker 3: whether it's the number or the scale of hallucinations and 65 00:03:21,400 --> 00:03:24,360 Speaker 3: answers that chatbots give you, or the scale of work 66 00:03:24,400 --> 00:03:27,520 Speaker 3: that you can put together today that just wasn't possible 67 00:03:28,200 --> 00:03:30,800 Speaker 3: in the past. We've already had so much new capability 68 00:03:30,840 --> 00:03:34,200 Speaker 3: generated just in the couple of years since Chat schipt 69 00:03:34,360 --> 00:03:37,000 Speaker 3: that in some ways, I think it's an absurd question 70 00:03:37,080 --> 00:03:38,920 Speaker 3: to ask will AI deliver it already has in a 71 00:03:38,960 --> 00:03:41,720 Speaker 3: lot of ways. But I think I understand why there's 72 00:03:41,760 --> 00:03:44,280 Speaker 3: plenty of questions about what about the investment is happening 73 00:03:44,400 --> 00:03:48,040 Speaker 3: right now, will that investment pay off? And I think 74 00:03:48,040 --> 00:03:52,600 Speaker 3: there are reasons to think carefully about how much each 75 00:03:52,640 --> 00:03:56,080 Speaker 3: company is putting in over what time horizon. But I 76 00:03:56,080 --> 00:03:58,840 Speaker 3: guess when I zoom out, I ask myself, do I 77 00:03:58,840 --> 00:04:00,360 Speaker 3: want a world in which there is more access to 78 00:04:00,400 --> 00:04:04,440 Speaker 3: compute or last? And it seems to me that we should, 79 00:04:04,720 --> 00:04:07,080 Speaker 3: rather than being concerned that there's too much compute being 80 00:04:07,520 --> 00:04:10,720 Speaker 3: put in the ground, be excited that there's actually companies 81 00:04:10,760 --> 00:04:12,960 Speaker 3: willing to invest in the infrastructure that's going to deliver 82 00:04:13,400 --> 00:04:14,360 Speaker 3: all these capabilities. 83 00:04:14,480 --> 00:04:16,120 Speaker 1: Well, I think it's an important question to ask for 84 00:04:16,160 --> 00:04:18,360 Speaker 1: a few reasons, and one of those is because if 85 00:04:18,400 --> 00:04:20,839 Speaker 1: you look at just the capex that these companies are spending, 86 00:04:20,880 --> 00:04:23,800 Speaker 1: you know, one company two hundred billion dollars in a 87 00:04:23,839 --> 00:04:27,440 Speaker 1: single year. That's a big commitment, and that's something that 88 00:04:27,440 --> 00:04:30,479 Speaker 1: they have to convince investors that is money that is 89 00:04:30,520 --> 00:04:34,160 Speaker 1: going to the right place, that's money that has to 90 00:04:34,200 --> 00:04:38,120 Speaker 1: be earned back. Plus on companies coming to a hyper 91 00:04:38,120 --> 00:04:41,599 Speaker 1: scaler and saying, yes, we think that this money is 92 00:04:41,640 --> 00:04:43,640 Speaker 1: not only being well spent, but then we can use 93 00:04:43,680 --> 00:04:47,200 Speaker 1: this compute to actually create a product that will provide 94 00:04:47,200 --> 00:04:49,599 Speaker 1: a return on investments. So, yes, we've seen a lot 95 00:04:49,640 --> 00:04:52,320 Speaker 1: of the hyperscalers benefit the you know, the anthropics and 96 00:04:52,360 --> 00:04:55,960 Speaker 1: the open ais and the mistraws like that's amazing stuff. 97 00:04:56,000 --> 00:04:57,920 Speaker 1: But at the end of the day, there are a 98 00:04:58,040 --> 00:05:01,760 Speaker 1: lot of companies that are not necessarily technology companies that 99 00:05:02,080 --> 00:05:06,920 Speaker 1: maybe aren't yet necessarily seeing an increase in productivity as 100 00:05:06,960 --> 00:05:10,400 Speaker 1: a result of these tools. So my question is do 101 00:05:10,520 --> 00:05:12,800 Speaker 1: those companies start to see that do we live in 102 00:05:12,839 --> 00:05:15,160 Speaker 1: a world where this is a layer, just like the 103 00:05:15,160 --> 00:05:18,839 Speaker 1: Internet was a layer of technology. 104 00:05:19,000 --> 00:05:20,640 Speaker 3: I think there's no doubt that we're going to have 105 00:05:20,720 --> 00:05:24,320 Speaker 3: AI as a layer that's embedded into all technology that 106 00:05:24,480 --> 00:05:27,039 Speaker 3: we use. And when I look at the economic question, 107 00:05:27,240 --> 00:05:30,039 Speaker 3: I say, first off, is it profitable to serve AI 108 00:05:30,120 --> 00:05:32,880 Speaker 3: systems today, not train the next generation, but serve today. 109 00:05:33,040 --> 00:05:36,520 Speaker 3: And if you listen to what's publicly reported about OpenAI 110 00:05:36,720 --> 00:05:40,480 Speaker 3: or anthropic, their margins on their inference business are not 111 00:05:40,560 --> 00:05:43,440 Speaker 3: just positive but quite good, and so that I think 112 00:05:43,520 --> 00:05:46,360 Speaker 3: is pretty strong evidence that the delivery of already existing 113 00:05:46,400 --> 00:05:50,479 Speaker 3: AI services is a pretty profitable business. The next question 114 00:05:50,560 --> 00:05:52,320 Speaker 3: is should we be investing in R and D in 115 00:05:52,360 --> 00:05:56,200 Speaker 3: the next generation, which is of course very very expensive. 116 00:05:56,320 --> 00:05:59,160 Speaker 3: But I think it's hard to argue we should dramatically 117 00:05:59,200 --> 00:06:01,160 Speaker 3: slow down R and D and AI just given all 118 00:06:01,200 --> 00:06:04,080 Speaker 3: of the extraordinary improvements that we've seen over the past 119 00:06:04,160 --> 00:06:06,680 Speaker 3: couple of years in terms of capability. And so when 120 00:06:06,720 --> 00:06:08,760 Speaker 3: you start breaking it down and ask, well, which specific 121 00:06:08,800 --> 00:06:12,400 Speaker 3: investment do we think is excessive? Which specific investment would 122 00:06:12,480 --> 00:06:14,680 Speaker 3: rather not be doing? Would you really like to be 123 00:06:14,760 --> 00:06:17,400 Speaker 3: the CEO of the only big tech company that's not 124 00:06:17,480 --> 00:06:19,640 Speaker 3: investing in AI. That doesn't seem like a very comfortable 125 00:06:20,160 --> 00:06:22,159 Speaker 3: place to be. And I'd rather have a situation in 126 00:06:22,200 --> 00:06:25,520 Speaker 3: which the big technology companies are investing more rather than 127 00:06:26,160 --> 00:06:28,480 Speaker 3: putting money in the bank or buying back their stock 128 00:06:28,480 --> 00:06:31,040 Speaker 3: because they didn't have any comfitable by investment opportunity. 129 00:06:31,120 --> 00:06:33,159 Speaker 2: Well, you know, I would say Apple's doing it differently, 130 00:06:33,200 --> 00:06:35,200 Speaker 2: and I guess time will tell whether or not their 131 00:06:35,240 --> 00:06:39,280 Speaker 2: approach works out. Chris, I am curious to you know, 132 00:06:40,160 --> 00:06:42,200 Speaker 2: who do you talk to, what research do you follow, 133 00:06:42,240 --> 00:06:44,480 Speaker 2: what are the leading voices, what are the leading companies 134 00:06:44,480 --> 00:06:49,200 Speaker 2: that you watch to figure out kind of how the 135 00:06:49,200 --> 00:06:52,920 Speaker 2: semiconductor space is evolving. I mean, we still know TSMC 136 00:06:53,320 --> 00:06:56,120 Speaker 2: is still the big manufacturer of all chips in the world, 137 00:06:56,160 --> 00:06:58,880 Speaker 2: and there's geopolitical attentions to that. But give us an 138 00:06:58,920 --> 00:07:02,839 Speaker 2: idea of where do you think kind of investors and 139 00:07:02,920 --> 00:07:05,960 Speaker 2: just the world at large need to be focusing their 140 00:07:06,040 --> 00:07:09,520 Speaker 2: attention on. When it comes to the semiconductor world. Is 141 00:07:09,520 --> 00:07:12,000 Speaker 2: it Nvidia, is it China? Is it something else? 142 00:07:13,240 --> 00:07:15,360 Speaker 3: I think the hard problem is that it's all of 143 00:07:15,400 --> 00:07:18,520 Speaker 3: the above, and we've seen this play out in the 144 00:07:19,000 --> 00:07:21,760 Speaker 3: GPU supply chain over the past couple of years, where 145 00:07:21,760 --> 00:07:26,040 Speaker 3: we've had shortages and different types of components, different materials, 146 00:07:26,120 --> 00:07:29,720 Speaker 3: and the memory chips that go next to GPUs and 147 00:07:29,760 --> 00:07:33,560 Speaker 3: AI servers. Each part of the supply chain has had 148 00:07:33,600 --> 00:07:37,960 Speaker 3: to dramatically ramp up its production capacity to respond to 149 00:07:38,000 --> 00:07:39,680 Speaker 3: the surge and demand. And so if you only look 150 00:07:39,680 --> 00:07:43,000 Speaker 3: at one part of the supply chain, you miss the 151 00:07:43,120 --> 00:07:47,160 Speaker 3: challenges that other parts are often facing. And so it's 152 00:07:47,240 --> 00:07:49,600 Speaker 3: the chip designers. It's the chip makers, but it's also 153 00:07:49,640 --> 00:07:53,400 Speaker 3: the materials suppliers, which are often not even thought of 154 00:07:53,440 --> 00:07:56,800 Speaker 3: as being semiconductor firms, but produce many of the capabilities 155 00:07:56,800 --> 00:07:59,240 Speaker 3: that are critical to actually manufacture the AI chips and 156 00:07:59,280 --> 00:08:00,840 Speaker 3: servers that we need. 157 00:08:01,040 --> 00:08:02,680 Speaker 2: Do you think the world do you think the US 158 00:08:02,760 --> 00:08:05,280 Speaker 2: specifically does need to be restricting sales of its most 159 00:08:05,280 --> 00:08:09,200 Speaker 2: sophisticated chips to China or on others? I mean, we 160 00:08:09,200 --> 00:08:12,600 Speaker 2: did see Nvidia. They still face the uncertainty in China, 161 00:08:12,720 --> 00:08:16,120 Speaker 2: that's their largest market, or which it is the largest 162 00:08:16,120 --> 00:08:20,000 Speaker 2: market for chips. The government granting some licenses to ship 163 00:08:20,000 --> 00:08:22,720 Speaker 2: a small amount of some of its processors to customers there, 164 00:08:22,760 --> 00:08:25,080 Speaker 2: but and Vidia is not sure if the Chinese government 165 00:08:25,120 --> 00:08:27,120 Speaker 2: will give its approval. So there's still some back and 166 00:08:27,160 --> 00:08:32,320 Speaker 2: forth here. But is it still an arms race of source. 167 00:08:34,040 --> 00:08:35,240 Speaker 3: Arms race of sorts? 168 00:08:35,240 --> 00:08:35,720 Speaker 2: Excuse me? 169 00:08:35,840 --> 00:08:39,199 Speaker 3: Yeah, I think armed races is not a bad analogy. 170 00:08:39,280 --> 00:08:42,200 Speaker 3: When you listen to tech CEOs, they'll speak in the 171 00:08:42,200 --> 00:08:44,840 Speaker 3: same language. They're struggling to get access to all the 172 00:08:44,840 --> 00:08:48,400 Speaker 3: computing power that they need, that they envision needing more 173 00:08:48,440 --> 00:08:50,960 Speaker 3: tomorrow than they've got today. And if you listen to 174 00:08:51,040 --> 00:08:54,160 Speaker 3: Chinese technology leaders. What you hear from them is challenges 175 00:08:54,200 --> 00:08:58,520 Speaker 3: and getting access to computing power. And the primary reason 176 00:08:58,559 --> 00:09:03,080 Speaker 3: they've struggled to deploy AI products at scale is because 177 00:09:03,080 --> 00:09:05,720 Speaker 3: they've struggled to get access to all of the computing capabilities. 178 00:09:05,760 --> 00:09:07,960 Speaker 3: And it's on a regular basis we see new Chinese 179 00:09:08,000 --> 00:09:12,120 Speaker 3: models launched that can actually be deployed at scale because 180 00:09:12,160 --> 00:09:15,000 Speaker 3: they don't have access to the advanced chips that they need. 181 00:09:15,120 --> 00:09:17,080 Speaker 3: So this does, I think, seem to me that this 182 00:09:17,160 --> 00:09:20,360 Speaker 3: is still a very powerful card that the US has 183 00:09:20,440 --> 00:09:21,959 Speaker 3: to play, and so I think we should be very 184 00:09:21,960 --> 00:09:26,000 Speaker 3: careful or on any decisions to give China more access 185 00:09:26,080 --> 00:09:28,160 Speaker 3: or at least make sure that we're getting something in 186 00:09:28,200 --> 00:09:28,840 Speaker 3: exchange for that. 187 00:09:29,280 --> 00:09:34,120 Speaker 1: On that, Chris, the US support of TSMC and the 188 00:09:34,280 --> 00:09:38,520 Speaker 1: US support of Intel different types of support, And I 189 00:09:38,559 --> 00:09:41,520 Speaker 1: guess you could call the TSMC one encouragement to build 190 00:09:41,600 --> 00:09:44,280 Speaker 1: here in the United States. What is the right industrial 191 00:09:44,320 --> 00:09:48,960 Speaker 1: policy to reduce reliance on companies outside of the US. 192 00:09:51,240 --> 00:09:54,120 Speaker 3: It's a really hard problem because TSMC is such a 193 00:09:54,160 --> 00:09:58,800 Speaker 3: capable and efficient manufacturer in their home base in Taiwan. 194 00:09:58,880 --> 00:10:01,400 Speaker 3: But I think the US government is right to say 195 00:10:01,400 --> 00:10:04,680 Speaker 3: that we need a more diversified manufacturing base for the 196 00:10:04,679 --> 00:10:07,800 Speaker 3: world's most important semiconductors. I think we've learned over the 197 00:10:07,840 --> 00:10:10,679 Speaker 3: last couple of years there's no silver bullet. President Trump's 198 00:10:10,679 --> 00:10:14,640 Speaker 3: tried tariffs that comes with obvious downside. President Biden tried 199 00:10:15,240 --> 00:10:19,600 Speaker 3: subsidies that worked to a degree, but only to a degree. 200 00:10:19,800 --> 00:10:23,400 Speaker 3: Here's the reality. The chip industry has involved hundreds of 201 00:10:23,400 --> 00:10:27,120 Speaker 3: billions of dollars of cappax over the last several decades, 202 00:10:27,120 --> 00:10:29,320 Speaker 3: and so it's just not going to move fast. And 203 00:10:29,360 --> 00:10:32,240 Speaker 3: so if you want to slowly change the structure of 204 00:10:32,240 --> 00:10:35,840 Speaker 3: where chips are produced, you've got a plan for years 205 00:10:35,880 --> 00:10:39,959 Speaker 3: of years of implementation of measures designed at the shift 206 00:10:39,960 --> 00:10:41,200 Speaker 3: economics of the GIP industry. 207 00:10:41,440 --> 00:10:43,439 Speaker 2: All right, Kenly with there, Chris Miller, thank you so much. 208 00:10:43,480 --> 00:10:45,800 Speaker 2: Good to check in with you. Professor of International History 209 00:10:45,800 --> 00:10:49,600 Speaker 2: at the Fletcher School at Tufts University, author of Chipward 210 00:10:49,720 --> 00:10:51,839 Speaker 2: joining us on this Thursday.