Big Tech Firms' Demand for AI Chips

Show Transcript
Bloomberg May 1 23:24

Chris Miller, "Chip War" author and Tufts University Professor of International History, tells us about the growing demand for AI chips and explains Nvidia's extraordinary lead in the industry. He joins David Westin on "Wall Street Week" daily.

Disclaimer: This content is for informational and educational purposes only and does not constitute a recommendation or endorsement of any specific investment or investment strategy. Read more

Transcript

  • 00:00 We had earnings out of AMD.
  • 00:01 They're trying to make something to be with NVIDIA.
  • 00:03 I guess my basic question is,
  • 00:05 is there anybody that can narrow the gap with NVIDIA and man, you're actually in these high end chips.
  • 00:11 Well, the evidence right now is that Nvidia's got a extraordinary lead and looks like it's maintaining it.
  • 00:16 We see from all the big tech firms, they're
  • 00:18 pouring 10 or so billion dollars a year each quarter into building out their data center capacity.
  • 00:23 And
  • 00:24 the lion's share that money is going to NVIDIA, which is producing almost all of the key chips that are necessary both for training AI systems but also for deploying them at scale.
  • 00:34 Well, can
  • 00:35 you talk to us a little bit about the pie?
  • 00:37 I mean, when it comes to taking on NVIDIA
  • 00:40 or growing in this industry, is this about taking share from NVIDIA or is the overall pie growing as well?
  • 00:47 Well, the PI's going to grow dramatically.
  • 00:49 No matter what
  • 00:50 estimate you look at, there's going to be much more spend
  • 00:53 on AI infrastructure over the next five years than there is
  • 00:57 today.
  • 00:58 And there's different types of AI workloads.
  • 01:00 There's training big systems like the GPT systems that
  • 01:04 Open AI has released, but there's also the deployment of them running AI across many different types of devices.
  • 01:10 And it's possible that
  • 01:11 as we deploy more AI, we'll want different types of chips that are optimized to different domains.
  • 01:16 But
  • 01:17 right now it looks like both in the training and in the deployment, Nvidia's got a very strong position.
  • 01:22 So give us an overall sense of the supply and demand for these high end
  • 01:25 parallel processing chips.
  • 01:27 What is the supply?
  • 01:28 What is the demand?
  • 01:29 Because every time we turn around there's some big
  • 01:32 outfit like Google, like Microsoft saying we're going to invest a lot more in AI.
  • 01:38 Well, last year there was a really severe constraint in terms of supply for AI chips.
  • 01:42 It's it's lessened up somewhat this year.
  • 01:45 But one of the challenges that remains is that next to every NVIDIA GPU processor you need an ultra high-powered memory chip produced by SK Hynix or Samsung or Micron.
  • 01:56 And these memory chips are still in pretty tight supply.
  • 01:59 And so the entire system remains hard to get access to even for the biggest tech companies because these, the AI memory chips
  • 02:07 are still in deficit relative to where demand is.
  • 02:10 And so how does that write itself?
  • 02:11 I mean, is this a question of just increasing supply or
  • 02:15 are we going to see even bigger CapEx spends to try and get some more chips?
  • 02:21 Well, all of the world's big chip makers are trying to increase their capacity to produce these chips.
  • 02:26 The memory chip makers
  • 02:28 have said they're going to ramp up as fast as they can.
  • 02:31 TSMC in Taiwan, which produces most of Nvidia's chips
  • 02:35 has itself been bringing online new capacity over the past couple of years.
  • 02:39 So we're seeing the chip makers invest a lot more
  • 02:42 in bringing this ability online.
  • 02:45 But the the key factor on the demand side is that whether it's Meta or Microsoft or Amazon,
  • 02:50 their own CapEx RX keeps ramping up and it's
  • 02:53 being driven higher because they want to buy more and more of these chips relative to their prior
  • 02:58 expectations.
  • 02:59 So unless
  • 03:00 AI demand slows, and right now there's no evidence of it,
  • 03:03 we're going to see very tight markets for chips for the foreseeable future.
  • 03:07 Our government's making a difference in this area right now.
  • 03:10 We certainly have, for example, the Chips and Science Act here in the United States.
  • 03:13 They've got a fair amount of money, although given the size of the pie, I'm not sure it's material.
  • 03:18 Is it making a difference?
  • 03:19 Will it make a difference
  • 03:21 by the CHIPS and Science Act is going to impact the market over the medium term.
  • 03:25 But you got to remember the money is being allocated this year and it takes several years to build a chip making facility.
  • 03:31 So it's not going to be until
  • 03:33 towards the end of the decade when we start to see meaningful shifts in terms of what type of supply is online and also where that supply is being built.
  • 03:41 We are certainly seeing a lot more investment in chip making facilities in the United States.
  • 03:46 That's the point of the CHIPS Act.
  • 03:48 But those facilities aren't going to be up and running for at least a couple of years.
  • 03:51 And it's good perspective.
  • 03:52 I mean, these are long sort of conversations we're going to be having about when you see the US manufacturing industry when it comes to chips, really revive.
  • 04:00 But just to put some numbers on the CHIPS Act, I mean, it's going to divvy up about $39 billion in direct grants.
  • 04:07 We're talking about $75 billion in loans and that's between several different companies.
  • 04:13 Is that enough money to actually turn the tide here?
  • 04:18 Well, I think what you find is that those government funds have catalyzed a much larger increase in private sector investment.
  • 04:25 And so if you look for example
  • 04:26 at the data on the amount of manufacturing investment in computing and electronic systems, which is where
  • 04:33 chip fabs fall, you find a 15 times increase over the past couple of years relative to the prior decade.
  • 04:38 So there is a real meaningful
  • 04:40 uptick in investment in chip making facilities
  • 04:43 on its own, it's not going to
  • 04:45 dramatically transform how the industry is structured, but it is leading to
  • 04:50 substantial facilities being built in the United States.
  • 04:52 And that matters because today
  • 04:54 almost all AI chips have produced in Taiwan.
  • 04:57 Well, I want to pick about exactly that, the geopolitics.
  • 05:00 Would you cover in your book Chip war
  • 05:02 and you referred earlier to TSMC in Taiwan.
  • 05:05 We talked with Neil Ferguson recently, that economic historian and he emphasized how big a problem that really choke point in Taiwan might be.
  • 05:12 This is part of what Neil had to say.
  • 05:14 The current
  • 05:16 AI
  • 05:17 investment boom, the kind of mania that we have seen
  • 05:21 since Open AI revealed ChatGPT
  • 05:25 assumes that TSMC,
  • 05:29 the most sophisticated semiconductor manufacturer, will continue to be able to make those things for NVIDIA, and NVIDIA will be able to ship them
  • 05:38 to the people doing AI.
  • 05:40 Now, if there were a war over Taiwan
  • 05:42 that would immediately be
  • 05:44 disrupted, I think the economic implications of what would be the the Taiwan Semiconductor Crisis would be much, much larger
  • 05:53 than the economic implications of the Cuban Missile Crisis in 1962.
  • 05:57 So, Chris, what about how big a choice point is Taiwan right now
  • 06:02 and is it getting any better?
  • 06:05 Well, today almost all AI chips are made in Taiwan, so it's a huge choke point today.
  • 06:09 But I think if you look down the road as new factories are brought online, what you see is that
  • 06:15 the almost monopolistic position of Taiwanese manufacturers is going to be reduced somewhat.
  • 06:20 We see Intel, for example,
  • 06:22 building a series of new facilities in the United States and also catching up in terms of its manufacturing capabilities.
  • 06:28 TSMC itself is building
  • 06:30 3 new plants in Arizona which will be capable of producing AI processors.
  • 06:34 Samsung, the third of the
  • 06:36 three most advanced chip makers when it comes to processor chips, is building a vast new facility in Texas, which will also be capable of making
  • 06:44 AI processors.
  • 06:45 So we are going to end up in five years time with a much more diversified landscape when it comes to manufacturing the most advanced chips, including the chips that are critical for AI applications.