Friday, July 19, 2024

Industry News
CST News
Tester FAQs
Industry News

New Sanctions to Drastically Cut China’s Demand for High-End AI Servers

Wednesday, November 15, 2023

A new update issued by the Department of Commerce on its export control regulations last month brings a new round of developments in the ongoing chip war between United States and China.

This time, the update included the targeting of sectors such as semiconductor manufacturing equipment and HPC chips (primarily AI chips) as well as the addition of two new companies onto the Entity List.

TrendForce points out that a significant change is the formal inclusion of the NXT:1980Di—which was previously in a grey area—into the list of controlled items. However, since ASML has already been green-lit for shipments post-application, the immediate ripples from this move remain to be seen.

Conversely, the landscape for HPC chips is set for a shake-up, as the current ban expands control over the A800, H800, and L40S series. This development is poised to see China’s tech giants—ByteDance, Baidu, Alibaba, and Tencent (BBAT)—curb their appetite for NVIDIA’s high-end AI servers from 5–6% of the global AI server market to a scant 3–4%.

While the clampdown has put the brakes on Chinese CSP’s demand for high-end AI servers, TrendForce anticipates a stockpiling dash by BBAT in this transitional phase. NVIDIA is also likely to attempt to allocate its currently scarce resources, such as the H800, for use by Chinese customers.

In the long term, observing the continuous AI chip bans imposed by the US since 2022, there’s been an accelerating trend among sizable or technologically capable Chinese firms to develop independent chips. Case in point, Alibaba’s Pingtouge is throwing its hat in the ASIC ring, and Huawei is investing in its Ascend series to build a local ecosystem in China.

Furthermore, the edge AI realm, with its smaller models and inference chips, has less stringent performance appetites, and here too, Chinese chip makers like Pingtouge and Hanergy are also seen to be speeding up their development efforts.

TrendForce anticipates that AI chip makers, like NVIDIA and AMD, will also plan for products that comply with new US regulations. These mandates might drive chip makers to develop more diverse solutions to adapt to the constraints of global geopolitics. For example, they may expand product lines with more modest TFLOPS prowess like TF16 or TF32, or a pivot toward beefier die sizes, thereby meeting the technical requirements of the ban without sacrificing market reach.

Additionally, fresh sanctions could catalyze a paradigm shift among China’s mainstay patrons such as BBAT and academic and research institutions and encourage them to consider renting AI training resources from regions outside China. This strategy would underpin foundational work in the Large Language Model (LLM) arena, subsequently segueing into training, fine-tuning, or AI inference pursuits for smaller models on home turf.

The shift might also stoke NVIDIA’s enthusiasm in championing its DGX cloud subscription and leasing model, anchored by the A100 or H100 AI servers, or even broadening its horizons to include more diverse cloud services (the L40S among others). This gambit could not only reel in Chinese clientele but also cast a wider net to encompass customers in regions grappling with their own geopolitical quagmires, thereby providing a versatile solution amidst a complex international landscape.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved

CST Inc. Memory Tester DDR Tester
Copyright © 1994 - 2023 CST, Inc. All Rights Reserved