Chinese AI firms are thriving despite US chip restrictions

Tencent and Baidu, two of China’s top tech giants, say they have found ways to stay competitive in the global race for artificial intelligence even as the United States tightens controls on key semiconductors.

In April, the U.S. tightened export rules on certain chips made by Nvidia and AMD, even after the Trump administration lifted one controversial rule put in place by President Biden. These moves have prompted Chinese firms to adapt quickly to keep their AI plans on track.

Both companies discussed their strategies during recent earnings calls. Tencent’s president, Martin Lau, and Baidu’s AI cloud head, Dou Shen, each laid out how they plan to push ahead despite limits on high-end processing units.

Tencent has a “pretty strong stockpile” of GPUs

Lau said Tencent has built up “a pretty strong stockpile” of graphics processing units, or GPUs, which are critical for training large AI models. By buying ahead, the company secured enough chips to fuel its research for the next several “generations” of models.

GPUs offer the raw computing power needed to sift through immense volumes of data and help models learn patterns. But Lau argued that adding more GPUs isn’t always the best route to better results. Instead, Tencent has focused on squeezing more performance out of the chips it already holds.

“That actually sort of helped 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 said.

For running AI tasks, known as inferencing, Tencent is using “software optimization” to make each GPU more efficient. Lau added that the company is exploring smaller, leaner AI models that demand far less computing power and can still deliver strong outcomes.

“We just need to sort of keep exploring these venues and spend probably more time on the software side, rather than just brute force buying GPUs,” Lau said. He also noted that Tencent can lean on custom-designed chips and semiconductors produced within China.

Baidu can build apps using its full-stack AI

Baidu, which runs the country’s biggest search engine, pointed to its “full-stack” setup. This means it controls everything from the cloud servers where data lives to the AI models themselves—like its ERNIE chatbot—and the applications built on top of those models.

“Even without access to the most advanced chips, our unique full-stack AI capabilities enable us to build strong applications and deliver meaningful value,” said Dou Shen, president of Baidu’s AI cloud division.

Baidu’s leaders also highlighted their software tricks to lower the cost of running AI workloads. Because Baidu owns much of its technology stack, it can tweak each layer—from the infrastructure up—to get more out of every GPU it owns.

“With foundation models driving up the need for a massive computing power, the abilities to build and manage large scale GPU clusters and to utilize GPUs effectively has become key competitive advantages,” Shen said.

To cushion the blow from U.S. chip curbs, Baidu and others have also turned to Chinese-made semiconductors. Shen said homegrown chips, paired with an ever-more efficient local software stack, will form “a strong foundation for long-term innovation” in China’s AI sector.

China has been pushing hard to build its own chip industry in recent years. While most experts agree that domestic GPUs and AI chips still lag behind U.S. offerings, they say progress is clear. 

Analyst Gaurav Gupta of Gartner noted that stockpiling is just one tactic; China’s firms have also made steady gains across materials, equipment, chip design and packaging.

“They have achieved decent success,” Gupta said in an email, adding that these homegrown chips may not yet match U.S. leaders but “continue to make progress.”

In Washington and Silicon Valley, some U.S. executives have urged a rethink of export limits. Nvidia’s CEO, Jensen Huang, called the curbs a “failure” this week, arguing they do more harm to American companies than to Chinese buyers.