Special Features

AI Infrastructure Month

Chip bans? LOL! Chinese web giant Tencent says it has enough GPUs for future AI model training

Partly because America does AI wrong and it can get more done with less


Chinese web giant Tencent says it has enough high-end GPUs to train new AI models for years, in part because it’s found more efficient ways to do so.

Speaking on the company’s Q1 2025 earnings call, company president Martin Lau said Tencent has “a pretty strong stockpile of chips that we acquired previously”.

The company will use some “for the applications that will generate immediate returns for us” – mostly advertising and content recommendation.

We have enough high-end chips to continue training models for a few more generations

Tencent will use others to train more large language models using techniques that mean it won’t need to acquire more silicon.

“Over the past few months, we start to move off the concept or the belief of the American tech companies, which they call the scaling law, which required continuous expansion of the training cluster,” Lau told investors. “And now we can see even with a smaller cluster you can actually achieve very good training results.”

The company president added his view that “There's a lot of potential that we can get on the post-training side which do not necessarily meet very large clusters. So that actually help us to look at our existing inventory of high-end chips and say we should have enough high-end chips to continue our training of models for a few more generations going forward.”

Lau noted that agentic AI and chain of chain of thought workloads need more GPUs, but said software optimization offers Tencent “quite a bit of room for us to keep on improving the inference efficiency. So if you can improve inference efficiency 2x, then basically that means the amount of GPUs get doubled in terms of capacity.”

Tencent therefore intends to invest in efficiency initiatives, including training smaller models tuned to the needs of certain applications that require fewer resources.

We need to spend more time on the software side rather than just force buying GPUs

Lau also said Tencent is looking at alternatives to GPUs.

“We can potentially make use of other chips, compliant chips available in China or available for us to be imported, as well as ASICs and GPUs in some cases for smaller models. We just need to sort of keep exploring these avenues and spend probably more time on the software side rather than just force buying GPUs.

Lau’s remarks suggest the US-led effort to stop the flow of high-end GPUs to China has not succeeded and instead have spurred Tencent to focus on innovation and optimization. The USA’s GPU crimp was designed to preserve western tech companies’ lead in AI, and to stop China putting the tech to work in its military.

Contradictory cloud?

CEO Pony Ma also commented on GPUs in the context of Tencent’s cloud business, with remarks that don’t accord with Lau’s.

“GPU rental is directly related to cloud business and that's more like a reselling business mostly,” Ma said. “And to a large extent right now, we are putting it on a lower priority because especially when there's a short supply of GPUs then the GPU rental is lower priority for us.”

Whatever the state of Tencent’s GPU fleet, the company is in rude health.

Q1 revenue rose 13 percent year-over-year to $25.1 billion, while gross profit rose 20 percent to $14 billion. The company has 1.4 billion monthly active users of its Weixin and WeChat services and is turning them into more cash with an upgraded advertising platform that uses Generative AI to generate ads and “to deepen our system's understanding of merchandise and of user interests across our apps and so deliver better ad recommendations.” Those tools help merchants who operate “mini shops’” – e-commerce facilities embedded in messaging apps – to sell more stuff and helped to drive Tencent’s own growth.

The company is China’s number one long-form video streamer, and its top audio streamer too. Games revenue surged for both new and old titles.

Asked about future prospects, Lau told investors that he felt consumer spending “has bottomed out and [will] start to slowly recover.” But then the US/China trade war happened.

“I think we just have to see what will be happening in the next quarter. And it's very dynamic, right?” Lau said. “There's tariffs, but then what's the extent of the tariff?”

Lau said China’s government “is very supportive with stimulus” and “still has a lot of room for rolling out more stimulus.”

“So, I think tariff and the interplay of that, the impact of it on the economy and then the counter effect of the stimulus would be the two factors to watch.” ®

Send us news
20 Comments

MiniMax M1 model claims Chinese LLM crown from DeepSeek – plus it's true open source

China's 'little dragons' pose big challenge to US AI firms

China's asteroid-and-comet hunter probe unfurls a 'solar wing'

Plus: Hitachi turns graybeards into AI agents, Tiananmen anniversary censorship, AWS in Taiwan, and more!

Enterprises are getting stuck in AI pilot hell, say Chatterbox Labs execs

Security, not model performance, is what's stalling adoption

Are you a big AI business vendor making terrible AI business decisions? We can help

The word Microsoft does not appear in this article. Why would you think otherwise?

The launch of ChatGPT polluted the world forever, like the first atomic weapons tests

Academics mull the need for the digital equivalent of low-background steel

ChatGPT used for evil: Fake IT worker resumes, misinfo, and cyber-op assist

OpenAI boots accounts linked to 10 malicious campaigns

Meta offered one AI researcher at least $10,000,000 to join up

Mark Zuckerberg reached out to our source directly

AI coding tools are like that helpful but untrustworthy friend, devs say

A survey from AI biz Qodo finds robo-coding productivity gains are unevenly distributed

Trump administration's whole-government AI plans leaked on GitHub

The AI.gov repository and staging site vanished when we asked questions, but don't worry – we captured backups

Canva to job candidates: Thou shalt use AI during interviews

Design software slinger warns it won't hire devs who aren't good at modern tools

American coders are most likely to use AI

Baseball, apple pie, and assisted programming

As AI gallops through the federal workforce, lawmakers once again call for expanded training

The last effort never went anywhere