00:00I want to go down to some sound that we got from our colleague Ed Ludlow, who's in San Francisco,
00:03of course.
00:03He's the host of Bloomberg Tech. He was at an AI summit that took place there yesterday.
00:07And Jensen Huang, the CEO of NVIDIA, was there as well.
00:09That's right. He spoke exclusively with Bloomberg Tech host Ed Ludlow about the massive $500 billion investment in South Korea's
00:16SK Group,
00:16as well as recent developments in Chinese AI.
00:19The semiconductor industry has really changed.
00:22And the reason for that, because we used to build computers for people to use,
00:27is then we're going to still continue to build incredible computers.
00:30These are now processing AIs for humans to collaborate with.
00:35But in the future, we also have AI agents and robots, and they're going to be using computers.
00:41So instead of just a billion people using computers, we're going to have 100 billion agents and billions of robots
00:47all using computers.
00:48The computer industry that's built on top of the chip industry surely is not big enough.
00:54And so this is one of the realizations of the semiconductor industry, that now computers are built not just for
01:00people to use,
01:01but computers are being built for computers to use.
01:04My guess is that the semiconductor industry is probably going to have to be 10 times larger than it is
01:11today over the next decade or so.
01:15And so working with our partners in Korea and around the world to scale up the supply chain of semiconductors
01:21so that we're prepared for this AI future is really important.
01:23I've had the opportunity to ask you about this more than once this year,
01:26but how much do you need the Korean economy to kind of get going to increase the supply of HBM
01:31bits for NVIDIA-based systems wherever they are?
01:34Well, we don't have enough bits.
01:37We're constrained in HBM memories, LPDDR memories.
01:40We're constrained in just about every part of the supply chain.
01:43We're even constrained now with land and power and construction workers to set up the data centers.
01:49I think this is one of the areas that is going to make sure that we continue to build out
01:55in a throttled way for a decade.
01:57And the reason for that is because these infrastructure, unlike electronic devices like PCs and phones and things like that,
02:04it's really, really hard to scale up land, power, and shell.
02:07And so all of the supply chain just really needs to get built out over the years.
02:12I think we have the ability as an industry to double each year, but we're going to have a hard
02:17time growing much faster than that.
02:18The $500 billion number is large.
02:21Could you just talk a little bit more about what it encompasses?
02:23We've gone over a lot on your commitment to the U.S. in terms of spending.
02:26Is that NVIDIA spending in the Korean economy, or it's SK fronting capital expenditures just a little bit more detail?
02:35We're going to be purchasing memories from them for many years to come.
02:39And as you know, we build a lot of computers.
02:42In order to build a trillion dollars' worth of their Rubin systems, you're going to have to buy a lot
02:47of system memories to go with it.
02:48And so we have large purchase agreements and large purchase intentions with SK Hynix.
02:56Meanwhile, SK Telecom is going to become an AI cloud.
03:00We're starting to build already.
03:02They're intending to build up to two gigawatts in the near future.
03:06And in that agreement, we will be selling AI supercomputers to them.
03:11So between us, we're going to do half a trillion dollars' worth of business over half a trillion dollars' worth
03:17of business.
03:17I was able to sit down with SK Group Chairman Chey Taewon very recently for about 40 minutes.
03:22And at the end of the conversation, we got to what is the difference in approach, the academic difference in
03:27approach on AI between the United States and China?
03:30And his view on it was that China is very focused on lowering the dollar per token.
03:35In America, we're still focused on the quality of tokens.
03:39I wonder what you think of that.
03:42The goal of AI is to produce an intelligent and smart answer.
03:48Now, you could approach it in a couple of different ways.
03:50You could, of course, make all of the tokens smarter and smarter.
03:52And as a result, result in using less tokens to do so.
03:56You could also produce AIs that are much more efficient.
04:01And maybe you can think longer, explore more options.
04:03And as a result, produce a smart answer.
04:07There are many different ways to reach intelligence and deliver smart answers.
04:13In the end, really, I think you have to take a step back and just realize that both countries have
04:19extraordinary AI researchers.
04:21And whatever conditions and whatever resources that they have, amazing people will find great answers.
04:28And so you're going to find, you're going to, you know, my expectation is that China and the United States
04:33will continue to advance AI.
04:35The conditions are different.
04:36Their resources are different.
04:37Their constraints are different.
04:38But they're all, they're, you know, these amazing researchers will find answers.
04:42And I think that in the case of China, they're producing more AI researchers than probably all of the world
04:51has, you know, in any given year.
04:53And so they're producing, if manufacturing intelligence is important, they manufacture the most important version of it, which is the
05:02researchers.
05:03And so this is an area, a country that's going to produce excellent AI technology.
05:09We ought to keep, continue to learn from them, work with them.
05:13As you know, you're here in Silicon Valley, right here in San Francisco.
05:16The number of AI researchers here that came from China that are Chinese is really quite significant.
05:21And so, you know, we're really fortunate to have them here and, you know, we just got to keep on
05:27racing.
05:27You made your first post on X.
05:30I did.
05:31And you did so by sharing a letter signed by many of your peers, American companies, to talk about the
05:39importance of open models to America, to the industry, to the development of AI.
05:44And in the letter, it's pretty well explained, you know, your rationale.
05:47But what was the catalyst for now?
05:50Why did you and Satya Nadella and others need to do that in this moment?
05:55Well, we sense that there's a growing sentiment and the wrong sentiment for open models.
06:10It's really important to realize that open models is essential for safety.
06:15Open models is essential for security, for cybersecurity.
06:18Open models are essential for innovation.
06:20It's necessary for startups.
06:21It's necessary for sovereignty, company sovereignty.
06:26I see a future where the world uses tons of closed models.
06:31And I encourage everybody, including my company, to use OpenAI and Cloud and Cursor and Cognition and Perplexity.
06:40Use everything that you can out of the cloud because it's just easier.
06:47And you build only what you must.
06:50And so in order to build what you must, you need to have open models to do that with.
06:54And the areas where we must, maybe it's because we have expertise that we simply cannot afford to share.
07:00This is our company's alpha, our company's intelligence, and we have to make sure we keep that proprietary.
07:04Maybe it's because our company works in an industry that's regulated, and therefore we simply can't pass along the service
07:12-level agreement.
07:13And we have to make sure that we can deliver fully on the service and the promise that we sign
07:18up for.
07:19Maybe it's something to do with sovereignty that you simply, in a particular country, you have to have your own
07:26AI.
07:27You have to control your own AI.
07:28Whatever those reasons are, they could be cost reasons.
07:31But I think that largely I would recommend people build their own AIs, especially when they need to control it
07:38for whatever reason.
07:39And so I think the future is going to have lots and lots of use of AI that's closed and
07:44AI that's open that you can there build your own AI.
07:48Now, one of the things that people misunderstand about these open models is, yes, you can host it yourself.
07:57But you can build your own computer, but most people use computers in the cloud.
08:01Frankly, I think closed models are cheaper.
08:04If you don't have to build it yourself, if you don't have to train it yourself, it costs a lot
08:07of expertise to fine-tune and maintain and guardrail and keep it safe and evaluate it.
08:14And, of course, even build computers to host it.
08:16So there's nothing cheap about doing that.
08:18The reason why you need open models is because you need to have control, because you need to adapt something
08:25for your own very specialized use cases.
08:28And so I think there's a lot of misunderstanding about closed versus open.
08:32We felt that it was important for people to understand that there's a world for both.
08:36Open-weighted versus open-source as well.
08:39There is a distinction.
08:42Open-weighted, as much as open as you can.
08:46The more open it is, in the way that we work, we put the weights out there.
08:51We also teach people how to train the model from the data that we also open-source.
08:56And the reason for that is we want to enable you to completely reproduce the AI model that we've open
09:03-weighted.
09:03And so that ability, by us teaching you how to do that, you can then do it for yourself.
09:07You know, I think the idea that the world is going to be one or the other is just completely
09:14wrong.
09:16And the idea that open models is somehow unsafe is also fundamentally wrong.
09:21And so we just want to make sure that people understand.
09:23Amen.
Comments