The Wall Street Journal reports that, following U.S. regulations prohibiting NVIDIA from selling its high-performance artificial intelligence chips to China, the company’s engineers quickly designed a new product series to comply with the tightened rules. However, this American tech company faces a bigger challenge: Chinese cloud computing companies, including NVIDIA’s largest global customer, are not enthusiastic about purchasing its low-power artificial intelligence chips.
Since November, China’s largest cloud computing companies have been testing NVIDIA samples. Alibaba Group and Tencent have informed NVIDIA that the quantity of chips they plan to order this year will be significantly less than originally planned when the now-banned products were available.
In the short term, the downgrade of NVIDIA processors has narrowed the performance gap with local alternative products, making chips manufactured in China increasingly attractive to buyers.
The sources mentioned that Alibaba and Tencent are redirecting some advanced semiconductor orders to local companies such as Huawei Technologies Limited, relying more on chips developed in-house. Two other major buyers in China—AI pioneer Baidu and ByteDance, the owner of TikTok—are doing the same.
In the long run, Chinese buyers are uncertain about NVIDIA’s ability to continue supplying them, as U.S. regulatory agencies have committed to regularly reviewing chip export controls and may further tighten performance restrictions. Tech companies are adjusting business strategies to cope with the potential inability to access NVIDIA products in the future and avoid the costly process of continuously adjusting technical parameters to adapt to new chips.
For NVIDIA, headquartered in Santa Clara, California, meeting the requirements of U.S. regulatory agencies while supplying chips to Chinese customers is becoming increasingly challenging. The company has billions of dollars in early chip orders yet to be fulfilled, and China is one of its largest markets, historically accounting for around one-fifth of its revenue.
At present, there is a high demand for NVIDIA chips. However, the long-term risk of losing sales in the world’s second-largest economy due to geopolitical tensions poses a significant challenge, especially as China considers the development of artificial intelligence a strategic focus.
Analyst Frank Kung from technology research firm TrendForce notes that currently, about 80% of high-end artificial intelligence chips used by Chinese cloud computing companies come from NVIDIA, and this proportion may decrease to 50%-60% in the next five years. He adds that the tightening of future U.S. chip controls will exert greater pressure on NVIDIA’s sales in China.
NVIDIA states that it is working diligently to provide products that comply with U.S. regulations to its global customers. The company previously mentioned that it doesn’t anticipate short-term financial impacts due to restrictions on shipping artificial intelligence chips to China, as it can find other buyers for these chips. However, CFO Colette Kress expressed last year that in the long run, the prohibition of selling AI chips in China would hinder the U.S. industry’s ability to compete and lead in one of the world’s largest markets.
In the past two years, the Biden administration has implemented two rounds of export sanctions to restrict China’s access to advanced chips and technology. NVIDIA CEO Jensen Huang has expressed his desire to continue supplying high-end processors to China and is working with Chinese customers to obtain export licenses.
Since the initial restrictions in October 2022, NVIDIA modified its chips sold in China, reducing their performance to fall below the thresholds supervised by the U.S. government. In 2023, the company sold over $1 billion worth of such chips to Chinese customers.
A year later, the U.S. further cut off unauthorized exports of NVIDIA chips to China, prompting the chip manufacturer to develop another series of processors with lower performance for Chinese buyers, planned for release early this year. Last month, NVIDIA launched the GeForce RTX 4090 D, an upgraded version of its top gaming chip, adjusted after the latest U.S. restrictions.
Chinese companies have been testing samples of NVIDIA’s upcoming highest-performance AI chip, H20. Some testers indicate that this chip allows efficient data transfer between multiple processors, making it a better choice than domestic alternatives for building chip clusters needed to process artificial intelligence workloads. However, testers note that they need more H20 to simulate the computing power obtained from NVIDIA’s previous chips, increasing their costs.
Developers state that advanced Chinese chips, like their U.S. counterparts, can handle inference (predictive results from trained AI models) and less complex training tasks.
Huawei, acknowledged by NVIDIA’s Huang as a “strong competitor” in China, is seen as winning the market at the expense of U.S. chip manufacturers. Insiders reveal that in 2023, Huawei secured orders for at least 5000 Ascend 910B chips from major Chinese internet companies. This chip is considered the closest Chinese alternative to NVIDIA’s high-performance A100, which is banned for export.
Due to U.S. sanctions, Huawei faces production restrictions, and the delivery of these chips will extend into 2024. Official procurements in China, such as those by state-owned telecom operators, require the use of domestically produced chips like those from Huawei.
Insiders state that Huawei has intensified efforts to expand its software ecosystem and plans to launch a new high-end artificial intelligence chip as early as the second half of 2024.
Since the U.S. implemented restrictions in 2022, some government-backed AI computing centers have procured chips from Huawei.
Informed sources reveal that Alibaba’s chip division, Pingtouge, is also developing a new specialized AI processor named “Han Guang.” An executive at Alibaba Cloud mentioned, “If restrictions may only become stricter in the coming years, it’s better to start considering alternatives now.”
Earlier last year, the fervor for generative AI stimulated demand for advanced NVIDIA chips as large enterprises and startups in China sought to develop their large language models. Now, many smaller companies are scaling back such efforts and focusing on AI applications.
Kenneth Yang, co-founder of a Shanghai-based medical AI startup, plans to skip the latest NVIDIA chips targeting China and instead lease AI processing power from Baidu or Huawei as he develops a healthcare application.
Engineers at Chinese tech companies indicate that NVIDIA chips will remain a priority in procurement over the next 12 months because NVIDIA’s product ecosystem is more extensive, and local alternatives are still in high demand.
Kevin Xu, founder of hedge fund “Interconnected Capital,” suggests that in the long run, U.S. restrictions are likely to drive China to develop its technology. Once the current hoarding phase ends, NVIDIA’s Chinese business may become a casualty.
Related:
- Nvidia Forms ASIC Division, Aggressive Talent Hunt Underway
- Inside Story: Why U.S. Chip Leaders Might Stumble in China
- NVIDIA Draws the Line: Bans CUDA Emulation on Other Chips
- Unveiling the Speed Gap Between SSDs and USB Drives
- Intel, Qualcomm to Stop Huawei Chip Exports in 2024
- Honda China Cuts 1700 Jobs: Nikkei Exclusive
- Complete Step-by-Step Guide to M4 Mac Mini SSD Upgrade