Niphon | Istock | Getty images
Tence and BaiduTwo of the largest Chinese technological companies have revealed how they were careful in the world race of artificial intelligence, even though the United States tightens borders on key semiconductors.
Company methods include flea storage, making it more effective AI models and even the use of local semiconductors.
While the administration of American president Donald Trump abandoned a controversial rule of the Biden era chip, it has always tightened the exports of certain semiconductors of companies, of which Nvidia And Dmla in April.
The big names in the sector addressed the problem during their last telephone conferences on the results.
Martin Lau, President of Tencent – The operator of the largest Chinese messaging application WeChat – said that his company had a “fairly solid stock” of fleas which she had previously bought. It referred to graphic processing units (GPU), a type of semiconductor that has become the Order for the formation of huge AI models.
These models require a powerful computing power supplied by GPUs to process high data volumes.
But, said Lau, unlike the conviction of American companies that GPU clusters must develop to create a more advanced AI, Tencent is able to obtain good training results with a smaller group of these chips.
“This has somehow helped us to look at our existing inventory of high -end fleas and to say that we should have enough high -end chips to continue our training of models for some additional generations in the future,” said Lau.
With regard to inference – the process of carrying out an AI task rather than training – Lau said that Tence are uses “software optimization” to improve efficiency, in order to deploy the same amount of PPI to perform a particular function.
LAU added that the company also plans to use smaller models that do not require such important calculation power. Tencent also said that he could use chips and semiconductors made to measure currently available in China.
“I think there are a lot of ways [in] What we can meet the needs for expansion and growing inference, and we just need to continue to explore these places and probably spend more time on the software side, rather than just buying GPUs by force brute, “said Lau.
Baidu approach
Baidu, the largest Chinese research company, praised what he calls his “complete” capacities – the combination of his cloud computing infrastructure, AI models and actual applications based on these models, such as his Chatbot Ernie.
“Even without access to the most advanced fleas, our unique unique battery AI capacities allow us to create solid applications and offer significant value,” said Dou Shen, president of Baidu’s Ai Cloud Business, on the call for the company’s profits this week.
Baidu also praised software optimization and the possibility of reducing the cost of execution of its models, because it has a large part of the technology of this battery. Baidu’s management has also spoken of efficiency which allows it to make the most of the GPUs it has.
“With foundation models stimulating the need for massive computing power, the ability to build and manage GPU clusters on a large scale and effectively use GPUs have become key competitive advantages,” said Shen.
The executive of Baidu has also praised the progress made by national Chinese technology companies in AI semiconductors, a decision which, according to him, would help to reduce the impact of the American edges of the fleas.
“Auto-associated depict developed at the national level, as well as [an] The increasingly effective software battery will jointly constitute a solid base for long -term innovation in the AI ecosystem in China, “said Shen.
Focus of domestic fleas in China
China has accelerated the development of chips designed and manufactured on its soil at home for a few years. Most experts agree that Beijing remains overall behind the United States in the field of GPUs and IA chips, but there have been a few advances.
Gaurav Gupta, analyst covering Gartner semiconductors, said storage is uniform for Chinese companies treat export restrictions. In addition, some progress has been made in semiconductor technology in China, even if it stays behind the United States, Gupta added.
“China has also developed its own domestic semiconductor ecosystem, from materials to equipment to chips and packaging. Different segments have made different levels of progress, but China has been surprisingly extremely coherent and ambitious in this goal, and it must be admitted that they have experienced a decent success,” Gupta to CNBC by e-mail.
“This provides them with an avenue to get AI chips, which may not compete with those of the leaders of the American chips but continue to progress.”
Many American leaders have urged Washington to remove export restrictions in the light of China’s progress. The CEO of Nvidia, Jensen Huang, called the borders of “failure” this week, saying that they cause more damage to American companies than in China.
