In order to comply with new regulations on the export of chips to the United States, NVIDIA has been consistently releasing AI chips and graphics cards tailored for the Chinese market.
However, according to sources cited by The Wall Street Journal, since November 2023, major cloud service provider (CSP) in China such as Alibaba and Tencent have been testing samples of NVIDIA’s special chips. These Chinese enterprises have conveyed to NVIDIA that the quantity of chips they plan to order in 2024 will be significantly lower than their initial plans.
According to a report from The Wall Street Journal, in October 2023, the United States announced new regulations preventing NVIDIA from selling advanced AI chips to China. However, NVIDIA swiftly developed a “special edition” chip for China, allowing them to continue selling chips in the Chinese market without violating regulations.
Nevertheless, NVIDIA is facing another challenge: major Chinese CSPs are not actively purchasing the “downgraded” performance versions of the chips.
Chinese enterprises have been testing the highest-performance version, H20, of NVIDIA’s “special edition” AI chips. Some testers have mentioned that this chip enables efficient data transfer among multiple processors, making it a better choice than domestic alternatives for building chip clusters required for processing AI computational workloads.
However, testers also indicate that they need more H20 to compensate for the performance gap compared to previous NVIDIA chips, which increases their costs.
The report indicates that in the short term, the performance advantage of NVIDIA’s “downgraded” chips over domestic Chinese products is diminishing, making Chinese-made chips increasingly attractive to buyers.
Informed sources cited from the report suggest that major players like Alibaba and Tencent are redirecting some advanced semiconductor orders to domestic companies and relying more on internally developed chips. This trend is also observed with the other two major chip buyers, Baidu and ByteDance.
Looking ahead in the long term, Chinese customers are uncertain about NVIDIA’s ability to continue supplying them with chips, as U.S. regulatory authorities have committed to regularly reviewing chip export controls, potentially tightening restrictions on chip performance further.
Baidu developed its first self-researched ASIC AI chip, Kunlunxin, in early 2020, with its second generation scheduled for mass production in 2021 and the third expected to launch in 2024. Post-2023, Baidu aimed to use Huawei’s Ascend 910B acceleration chips and expand the use of Kunlunxin chips for its AI infrastructure.
After Alibaba’s acquisition of CPU IP supplier Zhongtian Micro Systems in April 2018 and the establishment of T-Head Semiconductor in September of the same year, the company began developing its own ASIC AI chips, including the Hanguang 800.
TrendForce reports that T-Head’s initial ASIC chips were co-designed with external companies like GUC. However, after 2023, Alibaba is expected to increasingly leverage its internal resources to enhance the independent design capabilities of its next-gen ASIC chips, primarily for Alibaba Cloud’s AI infrastructure.
According to the data from TrendForce, currently, around 80% of the high-end AI chips used by Chinese cloud computing companies are sourced from NVIDIA. However, in the next five years, this proportion may decrease to 50% to 60%.
On October 17th, the U.S. Department of Commerce announced an expansion of export control, tightening further restrictions. In addition to the previously restricted products like NVIDIA A100, H100, and AMD MI200 series, the updated measures now include a broader range, encompassing NVIDA A800, H800, L40S, L40, L42, AMD MI300 series, Intel Gaudi 2/3, and more, hindering their import into China. This move is expected to hasten the adoption of domestically developed chips by Chinese communications service providers (CSPs).
Chinese CSPs Strategically Invest in Both In-House Chip Development and Related Companies
In terms of the in-house chip development strategy of Chinese CSPs, Baidu announced the completion of tape out for the first generation Kunlun Chip in 2019, utilizing the XPU. It entered mass production in early 2020, with the second generation in production by 2021, boasting a 2-3 times performance improvement. The third generation is expected to be released in 2024. Aside from independent R&D, Baidu has invested in related companies like Nebula-Matrix, Phytium, Smartnvy, and. In March 2021, Baidu also established Kunlunxin through the split of its AI chip business.
Alibaba, in April 2018, fully acquired Chinese CPU IP supplier C-Sky and established T-head semiconductor in September of the same year. Their first self-developed chip, Hanguang 800, was launched in September 2020. Alibaba also invested in Chinese memory giant CXMT, AI IC design companies Vastaitech, Cambricon and others.
Tencentinitially adopted an investment strategy, investing in Chinese AI chip company Enflame Tech in 2018. In 2020, it established Tencent Cloud and Smart Industries Group(CSIG), focusing on IC design and R&D. In November 2021, Tencent introduced AI inference chip Zixiao, utilizing 2.5D packaging for image and video processing, natural language processing, and search recommendation.
Huawei’s Hisilicon unveiled Ascend 910 in August 2019, accompanied by the AI open-source computing framework MindSpore. However, due to being included in the U.S. entity list, Ascend 910 faced production restrictions. In August 2023, iFLYTEK, a Chinese tech company, jointly introduced the “StarDesk AI Workstation” with Huawei, featuring the new AI chip Ascend 910B. This is likely manufactured using SMIC’s N+2 process, signifying Huawei’s return to self-developed AI chips.
Some Chinese Companies Turn to Purchasing Huawei’s Ascend 910B, Yet It Lags Behind A800
Huawei’s AI chips are not solely for internal use but are also sold to other Chinese companies. Baidu reportedly ordered 1,600 Ascend 910B chips from Huawei in August, valued at approximately 450 million RMB, to be used in 200 Baidu data center servers. The delivery is expected to be completed by the end of 2023, with over 60% of orders delivered as of October. This indicates Huawei’s capability to sell AI chips to other Chinese companies.
Huawei’s Ascend 910B, expected to be released in the second half of 2024, boasts hardware figures comparable to NVIDIA A800. According to tests conducted by Chinese companies, its performance is around 80% of A800. However, in terms of software ecosystem, Huawei still faces a significant gap compared to NVIDIA.
Overall, using Ascend 910B for AI training may be less efficient than A800. Yet with the tightening U.S. policies, Chinese companies are compelled to turn to Ascend 910B. As user adoption increases, Huawei’s ecosystem is expected to improve gradually, leading more Chinese companies to adopt its AI chips. Nevertheless, this will be a protracted process.
According to IThome’ report, Alibaba Group CEO Eddie Wu, speaking at the 2023 World Internet Conference Wuzhen Summit today, announced that Alibaba is gearing up to open-source a massive model with 72 billion parameters. This model is set to become the largest-scale open-source model in China.
Wu expressed that with AI becoming a crucial breakthrough in China’s digital economy innovation, Alibaba aims to evolve into an open technology platform. The goal is to provide foundational infrastructure for AI innovation and transformation across various industries.
It’s reported that Alibaba has, up to this point, open-sourced Tongyi Qianwen’s Qwen-14B model with 14 billion parameters and the Qwen-7B model with 7 billion parameters.
According to guandian’s report, Eddie Wu mentioned at the 2023 World Internet Conference Wuzhen Summit that AI technology will fundamentally transform the ways knowledge iteration and social collaboration occur, creating a profound impact on productivity, production relationships, the digital world, and the physical world.
He emphasized that society is currently at a turning point from traditional computing to AI computing, with AI eventually taking over all computing resources. The dual drive of AI and cloud computing is the underlying capability that Alibaba Cloud relies on to provide services for the future AI infrastructure.
In addition, on October 31st, Alibaba Cloud announced the release of the large-scale model Tongyi Qianwen 2.0 at the Apsara Conference. On the same day, the Tongyi Qianwen app was officially launched on major mobile application markets. Compared to the 1.0 version released in April, Tongyi Qianwen 2.0 has shown improvements in capabilities such as complex instruction understanding, literary creation, general mathematics, knowledge retention, and illusion resistance.