Tencent and Baidu, two of China’s top tech giants, say they have found ways to stay competitive with the global race for artificial intelligence, despite the US strengthening its management of major semiconductors.
In April, the United States tightened the export rules for certain chips created by Nvidia and AMD, even after the Trump administration lifted one controversial rule set by President Biden. These moves have prompted Chinese companies to adapt quickly to keep their AI plans on track.
Both companies Discussed Strategies during recent revenue call. Tencent president Martin Lau and Baidu’s AI Cloud Head Dou Shen each had planned how they planned to push despite the limitations of high-end processing units.
Tencent has a “pretty strong stockpile” of GPUs
Lau said Tencent We have built a “very powerful stockpile” of graphics processing units or GPUs. This is important for training large-scale AI models. By purchasing in advance, the company has secured enough tips to promote research for the next few “generations”.
GPUs provide the raw computing power needed to sift through huge amounts of data and help models learn patterns. However, Lau argued that adding more GPUs is not necessarily the best route to better results. Instead, Tencent is focusing on narrowing down more performance from the chips it already holds.
“It really should help us look into existing inventory of high-end chips and have enough high-end chips to continue training the model for generations to come,” Lau said.
To perform AI tasks known as inference, Tencent uses “software optimization” to make each GPU more efficient. Lau added that the company is exploring smaller, more lean AI models that require less computing power and can deliver strong results.
“We need to continue exploring these venues and spend more time on the software side, not only buying GPUs,” Lau said. He also said Tencent could rely on custom designed chips and semiconductors produced within China.
Baidu can build apps using full stack AI
Baiduwhich operates the country’s largest search engine and pointed out the “full stack” setup. This means controlling everything, from the cloud server where the data resides, to the AI models themselves (such as the Ernie Chatbot) and applications built on top of those models.
“Even if you don’t have access to the most advanced chips, our unique full-stack AI capabilities allow us to build powerful applications and deliver meaningful value,” said Dou Shen, president of Baidu’s AI Cloud Division.
Baidu’s leaders highlighted software tricks to reduce the cost of running AI workloads. Baidu owns a lot of the technology stack, so you can adjust each layer over your infrastructure to get more of every GPU you own.
“The foundational model that increases the need for large-scale computing power has made the ability to build and manage large GPU clusters and effectively utilize GPUs,” Shen said.
To mitigate the impact from US chip curbs, Baidu and others have also relied on Chinese-made semiconductors. Shen said homemade chips paired with more efficient local software stacks will form a “strong foundation for long-term innovation” in China’s AI sector.
China has been working hard in recent years to build its own chip industry. Most experts agree that domestic GPUs and AI chips are still behind our offerings, but they say progress is clear.
Gartner analyst Gaurav Gupta noted that stockpile is just one tactic. Additionally, Chinese companies are steadily making profits across materials, equipment, chip design and packaging.
“They’ve had decent success,” Gupta said in an email, adding that these homemade chips are still not in line with the US leaders, but they are “continuing to make progress.”
In Washington and Silicon Valley, some US executives are calling for a rethink of export restrictions. Nvidia CEO Jensen Huang calls Curbs a “fail” this week, claiming it is doing more harm to American companies than Chinese buyers.
Cryptopolitan Academy: Are you tired of the market shaking? Learn how Defi can help you build a stable passive income. Sign up now
