Who is winning the US-China AI race?

[00:00:00]Let's bring in Tiachi Wheels digital lead, Abdulwahab, a YouTuber to get more analysis on this. Welcome to Abdulwahab.

[00:00:05]So, we're just talking about the Kimik K3 model topping some of the US models. So, the Chinese projects have really gained ground, but then as I read, they ran out of graphics processing units.

[00:00:21]So, obviously not everything is a smooth ride. They had to pull some of their sign-ups. So, I mean, as fast as they're developing, is there still limitations in China?

[00:00:31]>> Yeah, definitely this Moon Shot's K3 Kimik K3 release is marking the headlines tech headlines across the world. And exactly from like because it just on July 16th is released this Kimik K3 2.8 trillion parameters. Just to put into context the AI frontier models like uh uh Cloud Code or GPT-5, uh they have the similar parameters, which is which is the capacity to produce knowledge on GPU on training base. So, it's approaching them. This is has shocked everybody. But what follows just 48 hours later, uh they stopped subscriptions because of the lack of computer to to to face all the demand across the world. So, this shows two things.

[00:01:20]First, that China has the talent, has the capacity to produce fast and cheap models, which can even match the US. But the second the key challenge is China as bottleneck is the lack of GPUs, lack of advanced advanced like GPU where the the tokens are being produced, trained, and inference, which is reasoning of this knowledge. So, US advantage is exactly here.

[00:01:52]>> So, China has the So, the US has the advantage early on. It kind of got that early head start. China's gaining.

[00:01:59]I mean, how how long will the US stay in front, would you say?

[00:02:02]>> Uh well, as we saw the latest Stanford AI index is saying that it's just very much approaching.

[00:02:11]Just for context, last year we were talking about the year ahead US is in frontier models in comparison to China. Now we're talking two to six months and even with the Kimi K3, we talk about probably weeks behind it. So China is coming to really fast following the US. Still US is leading in uh closed source model models because of the advantage on chips, advanced chips from Nvidia, advanced HBM memories and other networking like systems. But China is with its open models open to everybody basically around the world and trying to gain masses. So by by by using US frontier models for training. It just uses that knowledge to create inference reasoning. So in that part, China is really much catching up.

[00:03:08]>> I've heard that the heads of the various big tech companies say that whoever gets there first in AI, whoever wins the race, that's the only position that matters. Everybody else, second, third, fourth, they would just fade into insignificance. Given this the importance of this race, we've got the the profit earnings coming up soon from the big companies. Are we going to see them plowing all the profits back in into the into AI?

[00:03:30]>> Oh yeah, definitely. I mean, this release I mean just showed up that the real bottleneck is is compute like advanced compute, GPUs, CPUs, HBM memories and electricity without increasing prices, especially the US.

[00:03:47]So I do expect that all magnificent seven, especially like Google tomorrow will uh and and Amazon next week. I expect to increase uh their capital spending on AI infrastructure, definitely, because the only way to uh go through and to gain uh market capital. I mean, to to to produce the best uh most advanced uh AI uh is to invest a lot on the on the infrastructure, which uh I mean, we see China doing that with the state support.

[00:04:21]So far, US is dealing This is a main difference. While US is doing this on the private investment from the big companies, like especially the Magnificent Seven, China uh is uh the state itself uh has a huge stakes and uh support a lot these frontier models.

[00:04:38]>> Yeah, the AI race is truly heating up.

[00:04:41]Great analysis. Thank you so much, Abdul Waheed That's our TMT digital editor, Abdul Waheed AUP, with me.