HomeVideos

Why Nvidia is rooting for Chinese AI

Now Playing

Why Nvidia is rooting for Chinese AI

Transcript

110 segments

0:00

There is a very hot debate in Washington

0:02

and in Silicon Valley about whether to

0:04

regulate Chinese models, open-source

0:06

models. And um Jensen Huang weighed in

0:09

on this, the Nvidia CEO, of course. He

0:11

he talked with Axios. And I'm going to

0:13

play for you what he had to say. He

0:15

basically said the US has nothing to

0:16

fear from China's open-source AI models.

0:20

>> These China Chinese models are

0:21

excellent. Uh open open-source models

0:24

uh that are excellent should be used.

0:26

And so the market has misunderstood the

0:29

impact of DeepSeek the first time. It's

0:31

misunderstood the

0:33

Yeah, it's it's misunderstood the the

0:35

impact of Kimia again this time. I think

0:37

first of all, with great AI

0:40

open models, it's great for the whole

0:42

industry. Obviously, when there's more

0:44

if there's great AI, even if it's open,

0:47

uh wherever it comes from, uh there will

0:49

be more use. Whenever there's more use,

0:51

you'll have to sell a lot more Nvidia

0:53

computers, we'll have to build more data

0:54

centers, we'll have more services, the

0:57

technology will diffuse into more

0:58

industries. And so

1:00

starting point is great models lead to

1:03

great use, which leads to great growth.

1:08

>> So interesting that he's talking about

1:10

that cuz a couple things stand out to

1:11

me, Dan. One, um

1:13

it's good for Nvidia if there's more use

1:15

of [laughter]

1:16

of AI. Um

1:18

two, you know, Nvidia has stakes in a

1:21

lot of the companies that will be

1:22

competing with the open-source models

1:25

out there. Um so in that way it's not

1:26

necessarily in their financial interest.

1:28

And three, he doesn't even touch and

1:30

maybe he does elsewhere in the

1:31

interview, to be fair. Um he doesn't

1:32

even touch the Chinese security concerns

1:35

about these open models.

1:37

>> Yeah, I think one of the things to keep

1:38

in mind with these open open models is

1:41

they're they're open weights so that you

1:42

can you can make certain adjustments but

1:44

not, you know, fully kind of go in and

1:46

and kind of tool with it on your own.

1:48

You can customize it to a degree. What

1:50

What are the things that's

1:52

uh noteworthy about open-source software

1:55

is that so many people, because it's

1:57

open source,

1:58

usually are messing with it that they're

2:02

addressing cybersecurity issues on the

2:05

spot. So, you know, that's what makes

2:07

open source software so kind of secure.

2:10

It seems stupid because you're like,

2:12

well, everybody has access to it. Why

2:14

would it be safe? But it's because

2:16

everybody has access to it that it is.

2:18

They're all, you know, trying to ensure

2:20

that it's it's not, you know, something

2:22

that people can can break. Um when it

2:25

comes to what Jensen said uh on the, you

2:28

know, uh

2:29

deep seek versus Kimmy uh K3 is, you

2:34

know,

2:35

kind of what what they continue to say

2:36

is

2:37

if it becomes easier to run models, more

2:40

people will buy our our hardware because

2:43

they'll want to run more models.

2:44

>> Mhm.

2:45

>> So, that the thinking is, okay, we have

2:48

more companies that can afford to run

2:50

models or get access to models, they're

2:53

going to need to power that somehow, and

2:55

they'll use our soft our hardware. The

2:58

other flip side of that is that uh

3:01

his his regular uh kind of uh approach

3:03

is

3:04

the more hardware you put behind a

3:06

model, the better the responses are. So,

3:09

in other words, the the, you know, the

3:10

the bigger the engine, the faster the

3:11

car goes. Uh that's what what he's

3:15

regularly applying to these kind of AI

3:17

models as well. So, you know, I don't I

3:20

don't see this as as necessarily a

3:22

negative for Nvidia. Um but, you know, I

3:26

I think it it kind of leads to the

3:28

question of how far into this argument

3:31

can he go because of the geopolitical

3:33

implications, you know, Nvidia's trying

3:35

to get back on sale in China. China's

3:38

saying maybe some companies will take

3:39

this. The US has, you know, banned

3:41

certain chips, then kind of flipped on

3:44

that. And so, it's it's interesting to

3:46

see him kind of go out and and talk

3:48

about this kind of stuff. Yeah, it's

3:49

crucial to his business, but also, you

3:51

know, could raise uh could be like

3:54

kicking a hornet's nest in Washington if

3:57

>> Yeah.

Interactive Summary

This video covers Nvidia CEO Jensen Huang's recent perspective on the debate regarding Chinese open-source AI models. Huang argues that excellent open-source models are beneficial for the entire industry, ultimately driving more demand for Nvidia's hardware. While analysts discuss the potential geopolitical risks and security implications of these models, the consensus remains that increased accessibility and use of AI technology generally favor Nvidia's business strategy of hardware expansion.

Suggested questions

3 ready-made prompts