Chinese giants trace the future of AI without NVIDIA

USA’s rigid restrictions on advanced chips export accelerate tests of domestic alternatives; Chinese technology groups prepare for a future of AI without Nvidia

China’s largest technology companies have started a complex and time consuming process: migrating artificial intelligence development to locally manufactured chips. The movement arises amid increasingly scarce inventory of Nvidia processors and the hardening of the commercial sanctions imposed by the United States. According to industry executives, giants such as Alibaba, Tencent and Baidu have already begun testing alternative semiconductors to meet the growing internal and client demand for AI solutions.

The measure was accelerated after the administration of former President Donald Trump intensifying, last month, restrictions on the sale of the favorite chip of these companies: Nvidia’s H20, a less powerful version developed to comply with the rules imposed during Joe Biden’s government.

Limited stocks and search for alternatives

Sources close to the subject reveal that existing Nvidia chip stocks should sustain AI development only by the beginning of next year. As new requests take three to six months to be delivered, and there is no guarantee that Nvidia will be able to launch a competitive processor compatible with the new export rules, Chinese companies are running against time.

Shen Dou, leader of Baidu’s cloud AI group, recently said that the company has several chip options – especially for inference processing (crucial stage in solving AI problems) – which can replace 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.”declared Shen.

Adaptation and Strategies of the Tech Giants

In recent communications, executives from major Chinese companies signaled changes in their strategies. Eddie Wu, CEO of Alibaba, stressed 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 chips and evaluating alternatives.

“We have enough stock of high performance chips to train our models for some generations”said Lau, adding that Tencent can “Use other processors” To meet the demand for inference, which has been growing with the popularization of AI in the country.

Huawei emerges as an alternative, but enlists challenges

A Think-Tank linked to the Chinese State Ministry of Security stressed that while US sanctions are harmful, they “They encouraged a wave of independent innovation on high -performance AI chips, with Huawei’s Ascend series as the main example.”. In fact, Chinese entities have already begun to take Huawei chips on a large scale, according to the China Institutes of Contemporary International Relations.

Initially, the largest buyers were state -owned companies such as China Mobile, and sensitive sectors such as defense, health and finance. It is now expected that a broader range of technology companies will play Huawei chips. However, much avoid publicly commenting on ascend processors after the US warned that their use “Anywhere in the world” It can lead to criminal penalties.

Nvidia in search of a solution, but with limitations

GF Securities analysts estimate that Nvidia can start producing a new US export -compatible chip in July. However, the processor, based on Blackwell architecture, will have no high bandwidth memory (HBM), an essential component for processing large data volumes with agility. Crucial details, such as the presence of NVLINK technology (used for fast chips), are still uncertain.

In a recent conference, Jensen Huang, CEO of Nvidia, admitted that the company does not yet have a definitive solution for the Chinese market: “At the moment, we have nothing”he declared.

Migration Costs and Challenges

The transition to local chips is not simple. Adapting Training Codes developed for the NVIDIA CUDA platform to the Huawei Framework is a time consuming process and requires intensive support for adjustment and optimization engineers. An executive from a large technology company estimated that migration would cause about three months late in AI development.

Most companies are considering a hybrid approach: Continue using Nvidia chips for model training while adopting local inference processors, whose demand fired the expansion of AI in China.

Huawei tries to increase production through partners and even with a new factory, but the offer does not yet follow the demand. Meanwhile, other Chinese manufacturers, such as Cambricon and Hygon, are also being tested by technology giants. Baidu and Alibaba, in turn, develop their own processors to reduce external dependence.

Conclusion: A technological race with global impacts

The dispute for technological sovereignty is redefining the global semiconductor market. While China accelerates the search for alternatives, Western companies monitor the developments of a trade war that can shape the future of artificial intelligence.

With information from Financial Times*