Ia Chinese follows his own way without Nvidia

In response to US sanctions, Chinese companies invest in their own chips to keep the leadership in artificial intelligence

Amid the climb of the restrictions imposed by the United States on advanced chips export, large Chinese technological companies are accelerating efforts to replace imported components with alternatives manufactured within the country. The goal is to maintain the pace of the development of Artificial Intelligence (AI) without depending on the American giant Nvidia. According to information gathered from the sector, companies such as Alibaba, Tencent and Baidu have already started tests with local semiconductors, seeking to meet the growing internal demand for AI solutions.

Read also: Chinese giants trace the future of AI without NVIDIA

The change was intensified after the previous US government, led by Donald Trump, reinforced last month the prohibitions for selling the H20 chip – a version adapted by Nvidia to meet previous Joe Biden government demands.

Inventories run out and pressure increases

Reports of sources near the sector indicate that current NVIDIA chip stocks should last by early 2025. As new orders take between three and six months to be delivered, and there is no guarantee that the company will be able to launch a model compatible with new export rules, Chinese companies face a race against time.

Shen Dou, responsible for Baidu’s cloud AI group, recently said that the company has several chip options for the inference stage – fundamental for the practical application of AI models – capable of replacing those of Nvidia.

“We believe that over time, locally developed chips, combined with increasingly efficient software, will form a solid foundation for innovation in China’s AI ecosystem.”said Shen.

Adjusted strategies to deal with new reality

In recent weeks, executives from the largest technological companies in the country have signaled changes in their supply strategies. Eddie Wu, CEO of Alibaba, mentioned that the company is “Actively exploring diverse solutions to meet the growing demand of customers”. Already Martin Lau, president of Tencent, said the company is optimizing the use of its existing chips and evaluating alternative options.

“We have enough stock of high performance chips to train our models for some generations”guaranteed Lau, adding that Tencent can resort to “Other Processors” to supply the growing need for inference driven by the popularization of AI in the country.

Huawei emerges as the main alternative, but faces obstacles

An Institute of Studies linked to the Chinese State Ministry of Security stressed that, although American sanctions are challenging, they also “They encouraged a wave of independent innovation on high -performance AI chips, with Huawei’s Ascend series as the main example.”

According to the China Institutes of Contemporary International Relations, Chinese entities are already adopting Huawei chips widely. Initially, they were mainly state companies, such as China Mobile, and strategic sectors such as defense, health and finance. Now other technology giants are expected to use these components. However, much avoid talking publicly about tests with the Ascend chips, fearing US reprisals, which warned that their use “Anywhere in the world” It can result in criminal penalties.

Nvidia seeks solution, but faces technical limitations

GF Securities Analysts estimate that Nvidia can start the production of a new US export -compatible chip from July. However, the component, based on Blackwell architecture, will not include high bandwidth memory (HBM), an essential piece for rapid processing of large data volumes. Also, it is still unclear whether it will feature NVLINK technology, used to connect chips ultraly.

During a recent conference, Jensen Huang, CEO of Nvidia, admitted that the company does not yet have a definitive response to the Chinese market: “At the moment, we have nothing”he said.

Migration brings high costs and technical complexity

The transition process for local chips is complicated. Adapt codes developed for Nvidia’s Cuda Platform for Huawei’s Cann environment requires time and great support from engineers. An executive of a large company estimated that this migration could cause a delay of about three months in the development of AI projects.

Given this, many companies are opting for a hybrid strategy: continue using Nvidia chips for model training while adopting local processors for inference, an area whose demand exploded in China.

While Huawei tries to expand its productive capacity through partners and the construction of a new factory, the offer does not yet follow high demand. Other Chinese companies, such as Cambricon and Hygon, are also being tested by the big techs. In addition, Baidu and Alibaba are investing in their own development of chips, aiming to reduce external dependence.

The dispute for technological autonomy is transforming the global scenario of semiconductors. While China accelerates in search of domestic alternatives, western companies carefully observe the developments of a trade war that promises to shape the future of artificial intelligence worldwide.

With information from Financial Times*