Jack Ma-Backed Ant Touts AI Breakthrough Using Chinese Chips

An anonymous reader quotes a report from Bloomberg: Jack Ma-backed Ant Group used Chinese-made semiconductors to develop techniques for training AI models that would cut costs by 20%, according to people familiar with the matter. Ant used domestic chips, including from affiliate Alibaba and Huawei, to train models using the so-called Mixture of Experts machine learning approach, the people said. It got results similar to those from Nvidia chips like the H800, they said, asking not to be named as the information isn't public. Hangzhou-based Ant is still using Nvidia for AI development but is now relying mostly on alternatives including from Advanced Micro Devices and Chinese chips for its latest models, one of the people said.

The models mark Ant's entry into a race between Chinese and US companies that's accelerated since DeepSeek demonstrated how capable models can be trained for far less than the billions invested by OpenAI and Alphabet Inc.'s Google. It underscores how Chinese companies are trying to use local alternatives to the most advanced Nvidia semiconductors. While not the most advanced, the H800 is a relatively powerful processor and currently barred by the US from China. The company published a research paper this month that claimed its models at times outperformed Meta Platforms Inc. in certain benchmarks, which Bloomberg News hasn't independently verified. But if they work as advertised, Ant's platforms could mark another step forward for Chinese artificial intelligence development by slashing the cost of inferencing or supporting AI services.

Maybe, but can we trust them?

[TheMiddleRoad]

Lots of claims come from China, and most of them are false. Hard to know what to believe these days.

Re:

[Anonymous Coward]

I remember when Musk was the star of the Democrat party. It was only two years ago. He was doing solar, and EVs, and building trains and going to Mars.

Re:

[DrMrLordX]

Given how DeepSeek wound up, probably not.

Re:

[DamnOregonian]

Depends. Can probably trust some parts, and not others.
Parts that are easily verified, probably more likely to be true.

Like DeepSeek may have been completely full of shit on how much it costed, and what was used to train V3 and R1, but it is still a fact that R1 sits in the top 5 leaderboards, and now QwQ (another Chinese LLM) sits just below it, and it can run on commodity hardware (only 32B parameters).

Re:

[DamnOregonian]

Maybe. I have read conflicting reports, and I'm not 100% sure what to believe.
Your explanation sounds very reasonable, though.

Jack Ma

[Anonymous Coward]

Didn't that guy get "re-educated" by the Chinese government a few years back? What happened to him?

Re:

[Gavino]

I think he had a lobotomy where they took out the part that resists authoritarian control. Peng Shuai got the same treatment.

Re:

[ndsurvivor]

I was going to say something similar.

China has no FABs

[sinij]

I am skeptical, as Chinese do not have ability to FAB something like AI chips in mass. Then again, I am basing this on Peter Zeihan [youtube.com] analysis.

Re:

[ceoyoyo]

China has fabs. Quite a few of them, and they're building quite a few more.

https://en.wikipedia.org/wiki/... [wikipedia.org]

Re:

[EreIamJH]

I think it's important to search out the industry specific commentators. Industry specialists value add (and have the news weeks earlier), while generalist commentators just work to support their own narrative. The Taiwan based industry commentators are mostly talking about China's rapid improvement. China's plan is to reach generational parity by 2030 using 100% Chinese owned IP, but the talk is that they're on track to hit that target a year or two earlier.

Re:

[DamnOregonian]

China has shit-tons of fabs.
They've even got magical DUV 7nm.

Of course, given the amount of ICs made with it, I'm guessing it has early Intel 10 yields... but they do have it, and engineering problems have a way of being surmounted.

Their number crunchers aren't as good as our number crunchers, but the thing with number crunchers, is that they scale really, really well.
If yours isn't as good per-unit, make more units.

20%!!

[larryjoe]

Hmm, a claim of 20% seems really underwhelming. If Chinese hardware only saves 20% compared to Nvidia, that is fantastic news for Nvidia.

Reading the paper, the innovations are totally with the model and how to train it. It's like the DeepSeek news but with not so impressive results.

Re:20%!!

[AmiMoJo]

Most of the cost is probably electricity and the other overheads from running a large datacentre. Still, 20% is nothing to be sniffed at, it's a solid competitive advantage in a very competitive space.

Most importantly though, it proves that China has already developed chip technology that rivals Nvidia. It may not be quite as fast, but it's cheaper, and you can just buy more of them to make up the performance gap.

Re:

[larryjoe]

Most of the cost is probably electricity and the other overheads from running a large datacentre. Still, 20% is nothing to be sniffed at, it's a solid competitive advantage in a very competitive space.

Most importantly though, it proves that China has already developed chip technology that rivals Nvidia. It may not be quite as fast, but it's cheaper, and you can just buy more of them to make up the performance gap.

The Chinese aren't looking for 20% cheaper. They're looking for anything that works sort of the same for sort of similar costs. So, 20% cheaper and 50% more expensive but built in China are similar.

We'll see if this is real news or not. It'll be easy to tell. Nvidia's revenue to Singapore and Malaysia should tumble in the near future. Otherwise, it's just yet another PR piece in the Chinese PR stream. This is also the way to tell how useful hyperscalar ASICs are in displacing the need for Nvidia GPUs.

Yet more evidence...

[MpVpRb]

...to support my claim
There are a LOT of smart people in China. Trying to prevent them from getting tech is futile and counterproductive
There are a LOT of smart Chinese people working around the world in top research labs. They could easily return home if they see a better opportunity
There is a real possibility that China may become the world leader in science, tech, medicine, etc
We need to end the cold war and figure out how to work together

Re:

[Gavino]

>>We need to end the cold war and figure out how to work together That won't happen anytime soon. Not while the Chinese education system and state media is constantly throwing shade on the West, because they are still butthurt from the centuries-old Opium Wars, and because the USA helped defend Taiwan after the Korean War. China will not back down and "work together" in any meaningful way until they control Taiwan, which they see as theirs, even though Taiwan has never been under Communist rule. If t

Re:

[misexistentialist]

Unfortunately it's probably a reason for a hot war to kill those smart people. Behind all the talk about the benefits of collaboration there is a zero-sum reality.

The Chinese are leaping ahead

[mspohr]

Lots of naysayers here but I think recent Chinese advancements in robotics, AI, ADAS, and chips have shown that they are moving ahead.

What a great name!

[fortunatus]

I thought human personality was dried up in the tech sector. But what a great name, the Jack-Ma-Backed Ant Group! What is a jack-ma-backed ant? It has a funny, crooked back?