- 7 weeks ago
Category
🗞
NewsTranscript
00:00These companies investing trillions of dollars in CapEx, trillions of dollars in the race to build artificial intelligence support systems, trillions of dollars of our tech companies investing in building data centers in America.
00:13How far AI will take us and how fast may depend in part on a basic choice about the overall approach to sharing or withholding information, a choice often mentioned in passing, but one that investors may not have identified as key.
00:28I want to see AI everywhere. You know, we were such so early in the real usage of the technology.
00:37Lisa Su is chair and CEO of AMD, the chip maker with hundreds of billions of dollars at stake in the development of AI.
00:45When it comes to the software that runs on their products, she and her company have made their choice in favor of open source technology, sharing where they can.
00:55The difference with an open source or open standards is the idea that, you know, you set out a set of standards that people can follow and then different companies can choose to implement in a different way.
01:09And, you know, the whole idea is to provide platforms that the entire ecosystem, lots and lots of developers can develop and that you can be able to interchange sort of the best of breed from different sources.
01:24Why is there so much passion around open? I mean, is it a matter of business or is there a matter of philosophy?
01:30The philosophy is, you know, do you believe that there's any one company or any one group that will have all of the best ideas in the world?
01:39Or do you believe that if you have an open ecosystem where different researchers and different groups can contribute, that you will end up with a better product overall?
01:49And philosophically, I think, you know, from an AMD standpoint, you know, actually, if I think about, you know, throughout my career, I look at sort of the different inflection points on technology.
01:59Things often start, new ideas often start in a closed environment.
02:04But when you actually think about when they get big and when, you know, a lot of people adopt, you like to be in open environment.
02:11And, you know, that's philosophically. Now let's talk about the business aspects of it.
02:16You know, the business aspects of it are, you know, we all have to make choices of where we invest.
02:20And do you want to trust, you know, all of your crown jewels and all of your data in a closed ecosystem that, you know, may or may not be the most competitive at any given time or, you know, may have, you know, a flaw or any of those things.
02:34Whereas, you know, in an open ecosystem, you have choices. You can decide who is the best at any given time.
02:40You can decide if I have my data in a cloud environment, I can easily move it to another environment.
02:47If I have my applications that are built on one chip, if there's a better chip, you know, a few years later, you can move, you know, quickly.
02:55That's kind of the idea of both.
02:57What, broadly speaking, are the advantages of open?
03:00And let me throw one out, for example, in a new technology, is it more likely, all of the things being equal, that the technology will develop more quickly if it's open?
03:09I would say that the main advantage of an open ecosystem is that you can get many, many more developers on that ecosystem and they will contribute, you know, sort of in their special way.
03:23So, you know, an example I can give you is, you know, when we think about supercomputing, like the largest supercomputers in the world, we usually like to develop them on open software stacks so that researchers from all different labs can actually contribute to those applications and those learnings.
03:41We see that a lot in open ecosystems, which is the notion that we want more developers.
03:47Like Linux is a great example, right? Linux is, you know, one of the examples where with a open operating system, you can see that it's now really become, you know, the standard for a lot of computing going forward.
04:00For all the advantages of the open source approach to AI, many of the most important players, including most recently Meta, with a new model expected to debut next year, have gone the other way, keeping much of their generative AI tech proprietary.
04:16And there are some advantages to that approach, including preventing bad actors from having access to source code that they can manipulate.
04:24That's something Sue believes can be remedied by community policing.
04:29I think there is a definite view that there's a place for open and then there's a place for proprietary.
04:36And when I say that, I mean, look, there are amazing groups of researchers that are working on the largest foundational models.
04:44When you think about, you know, what's happening at OpenAI and what's happening at Google and what's happening at Anthropic, what's happening at Meta.
04:52These are phenomenal labs that are doing a tremendous research in AI.
04:57There are also a number of open models.
04:59You know, OpenAI has an open model.
05:01You know, Meta, their LAMA series has had open models.
05:05And what you find when those models are open is there are more researchers that are able to build on top of that.
05:11And I think there's significant advantages there.
05:14So, you know, our thought process in how this evolves over time is you're going to have both in the ecosystem.
05:21And just like my very early example of sort of the Apple iOS ecosystem and the Google Android ecosystem, it's not like you're going to have all of one takeover, all of the other takeover.
05:34But what you do see is there's value and richness in having, let's call it, you know, a little bit of competition between the ecosystems.
05:41And most importantly, what we want is what's best for the consumer is to get the best overall product experience and the best overall capabilities.
05:51And you do that when you allow a playing field that allows good competition across the board.
06:01The debate between open source and proprietary is not a new one.
06:05This IBM 705.
06:07The computer industry had to confront a similar choice long before the current rush to artificial intelligence models.
06:14Sam Palmisano was the CEO of IBM from 2003 through 2011.
06:20How this whole thing began was basically in the late 90s.
06:23And there were two different models.
06:25There were proprietary models that existed.
06:27IBM in the mainframe, Microsoft client server, et cetera.
06:31They were all proprietary.
06:32And we believed at IBM that a better model was one that would drive more innovation, more growth if it was open source so everybody could participate.
06:41So that led to this Linux initiative, which was the open source in this institute that was created by the industry, which we were a part of.
06:49And obviously what happened over time, now we put a lot behind it.
06:54We made a billion-dollar bet that Linux would become commercial and that in the future, Linux and Apache would be kind of the operating environment of the Internet.
07:03Well, it worked out.
07:04But it could have been a mistake, but it worked out.
07:06So that gave it a lot of momentum.
07:07But the whole point was open innovation is going to win over single-company proprietary approaches because they have a limited amount of resource and a limited market.
07:18Even if they have a large share position, it's still a smaller market than the entire industry.
07:23When you made the decision to go open source with Linux, was there controversy about it?
07:28What were the pluses and the minuses of that decision?
07:30Huge internal debate.
07:31Think of all the stuff IBM invented over the history of the industry.
07:34It was all proprietary, whether that was not just in mainframe or storage and database and all those things were all invented by IBM.
07:41So there's the camp that said, hey, you guys, this is crazy.
07:44We have this proprietary model.
07:46It's very profitable.
07:47Slower growth, but it's very profitable.
07:49And it's, you know, from a business perspective, the alternative was that, no, if we could open up the opportunity and then participate in a bigger opportunity than the one we participated in, IBM will be better off long-term.
08:01And that was the decision.
08:02The IBM move to open source for operating systems 25 years ago helped spur what became a software revolution as firms adopted the system and built their own proprietary applications on top of it.
08:16Looking back on it now, it's hard not to conclude that it was good for the business overall and ultimately good for companies like IBM.
08:24But is history likely to repeat itself this time as the world experiments with various generative AI approaches?
08:31The chip is actually quite small.
08:33We want to present a framework where we're going to get the best ideas coming out of this.
08:37And from an AMD standpoint, we'd love you to run on our chips.
08:41But more importantly, we want you to run on an open ecosystem so you can decide, you know, two years from now or four years from now, if you made a choice of AMD, that doesn't mean you're locked in to our chips for the next five years.
08:55It means that you have an open ecosystem where you can benefit from all of the competitive capabilities that come on board over the next five to 10 years.
09:05Which sounds like it should be good for the business long term, overall the business.
09:09Is it also good for AMD in the sense that in that world where you can switch and you don't have to be committed, you'll sell more chips over the long term?
09:15I believe we will because we are giving people the opportunity to choose.
09:20And look, we always have to be best in class, right?
09:23There's no there's no question that it's there's so many innovations to come on board.
09:29But the idea that if you develop on AMD, you have choice and you also have the ability to work closely with us in terms of how we develop this ecosystem together.
09:41That this is a case where one plus one is going to be greater than three because, you know, we're we're getting the smartest people from all all of the ecosystem versus just, you know, we're going to figure out everything on our own.
09:56And it's not just the big U.S. tech companies making their bets on open versus proprietary.
10:02China is casting its vote as well and perhaps ironically, siding with the open approach where we started.
10:09It was mostly proprietary and the winners at that point in time were meta, Google, open A.I., Microsoft, you know, et cetera.
10:17They were the winners and were growing like crazy with these large language models.
10:21Then all of a sudden this thing happens in China.
10:23And there's two things that happen in China that are open sourced.
10:26And they're based on this view that innovation will scale faster if we have an open source approach.
10:33DeepSeq, which got a lot of press, a lot of coverage.
10:36The other one's called Huggy Face, which you hardly hear about.
10:39But if you look at the numbers compared to the growth before 2024, it was heavily dominated by the three companies, the hyperscalers I talked about.
10:48And then since then, it's been it's been these other companies in the China approach.
10:52So since 2024, China actually is outgrowing the proprietary models.
10:57But it's the open source innovation and they've gone from nothing to millions of people now using these capabilities.
11:05To your point, Sam, there's this chart that we were provided actually by your folks out at Stanford, which actually illustrates the competitive nature of the large amount of language models in China versus the United States over time.
11:17Correct.
11:17You look at it, the point I was making up until 2024, it was clearly a blue curve led by the United States, the companies, but led by the United States.
11:26It flipped in 2024 and China went vertical and it's almost caught up to the United States now.
11:32This thing is rocketing.
11:33As happens so often, it ultimately comes down to business, albeit business with profound ramifications well beyond the specific companies involved.
11:44AMD's Sue has little doubt that what works best for AI development overall will also be the best way forward for her company.
11:52The ability to collaborate actually helps us go faster because you only have to differentiate on the things that are, let's call it the most secret sauce.
12:00And if we can come up with standards so that we're not doing the same work multiple times at different companies, that's actually a good thing.
12:08And the secret sauce for you is in the chip.
12:10Yes.
12:10The secret sauce is in the chip.
12:12It's in how we marry the chip to what will eventually be the application.
12:17And in our world, David, what I find the most interesting is the idea that every application that we see going forward, every device that you see is going to have AI as part of its essential element.
12:31And we want to be as much as we can the provider of that essential AI capability from the hardware side.
12:38We've seen, you know, sort of the advent of ChatGPT now turn into, you know, is there a AI bubble?
12:46And I say that we're just starting to see the real utility of AI.
12:51Yes, the investments are high.
12:53But why are we all so, so confident that there's a payoff at the end?
12:57The way AI has been improving over the last, you know, let's call it a couple of years, has been faster than any other technology that I've seen.
13:07The adoption rates have been faster than any other technology.
13:10The potential is faster than any other technology.
13:13But what I'm here to tell you is that it's nowhere near its peak capability.
Comments