To US controls are now joined by the Beijing himself, and point to his star chip
Nvidia has been dealing with the export controls that restrict the sending of their most advanced chips to China. But now, the company directed by Jensen Huang faces a double pressure: while trying to dodge Washington’s measures, it must also face new barriers imposed by Beijing, which threaten to further reduce their margin of maneuver in one of its key markets. A growing threat. As we said, the challenges for Nvidia are not new. In 2022, Joe Biden administration prohibited export of his chips A100 and H100specifically designed for artificial intelligence tasks. The company then warned that the measure could be up to 400 million dollars in sales losses in China. The answer soon: he decided to move. Thus, Nvidia designed a product designed exclusively to keep present in the Chinese market. To achieve this, he had no choice but to reduce the capacities of his chips and adapt them to regulatory demands. From there the A800 and H800 were born, cut versions of their star models. For a time they managed to be marketed in China, but a second round of controls ended up leaving them out of play. The H20 also wobbles. Once again, the team led by Jensen Huang got down to work to develop a chip adapted to the Chinese market. The result was the H20a cut version in front of its equivalents in the West, but raised as its great bet for China. Sales forecasts in 2024 exceeded one million units. However, obstacles have not taken long to appear. Now the pressures come from Beijing. Last year, the Chinese government began to recommend to local businesses to acquire NVIDIA GPUS. In full global career for artificial intelligence, the measure seemed contradictory. But the context explains it: just then, several Chinese manufacturers were finalizing their own alternatives to reinforce the country’s computation capacity without depending on foreign technology. Huawei has not stayed with crossed hands. The company has opted strong with products such as the Ascend 910C, a chip that, as points out Tom’s hardwarereaches inference to inference about 60% of the NVIDIA H100 yield. In addition, it is optimized for large language models and has already begun to be adopted by Chinese giants such as Baidu or Bytedance. Huawei also has other variants, such as the Ascend 910b. But there is more. In the middle of last year, The Chinese government presented an action plan to boost the “ecological development of data centers.” The objective was clear: to improve its energy efficiency. To measure the advances, the authorities chose a metric known as Pue (Power Usage Effectiveness)which relates the total energy consumption of the center – including air conditioning, lighting and other auxiliary systems – with the energy used exclusively by IT equipment, such as servers, networks or GPU. The objective of the plan was to reduce the can of the data centers below 1.5 by 2025. It should be remembered that the more the Pue can the value 1, the more efficient the data center is. One of the keys to achieve this goes to use more efficient graphics cards, which generate less heat and, therefore, reduce the energy consumption of the cooling system. The problem, As the Financial Times points outis that H20 chips do not finish fit in this equation. According to the aforementioned newspaper, the National Development and Reform Commission is urging local companies to use only chips that meet demanding energy efficiency standards, both in new data centers and in extending of the existing ones. In practice, this translates into increasing pressure on Chinese technology to reduce – or directly abandon – their GPUS dependency manufactured in the United States. For now, the regulations do not apply strictly, but everything indicates that that could change. In the horizon a possible hardening of control appears: inspections in situ, economic sanctions and harder requirements. If that scenario materializes, Nvidia could be seen before an even greater blow in which, until now, it is its second most relevant market: China represents 13% of its global sales, with more than 17,000 million dollars in annual income. Images | ABODI VESAKARAN | Nvidia In Xataka | The US suspects that Nvidia chips are arriving in China through a third surprising country: Malaysia
