In the Markets: China
8 min read
2024-09-23
Raymond Chan
Partner, Fangda Partners
Jeffrey Ding
Partner, Fangda Partners
Gil Zhang
Partner, Fangda Partners

executive summary

Gallery

No items found.
Need this in PowerPoint?
Enter your email to request the file.
Thanks! We'll send the file to your email shortly.
Something went wrong while submitting the form. Get in touch with 20Minds via info@twentyminds.com

article

Sample

How has AI investment evolved in China?

Most AI players in China have, so far, sought venture or growth capital rather than an M&A exit or listing. Alibaba has been the most active investor, but many other established tech giants and financial investors have made significant investments as well.

Large Language Models (LLMs)

Many of the notable AI-related transactions in China concern providers of LLMs:

As of 8 August 2024, 179 LLMs have completed the LLM go-live security assessment filing with the Cyberspace Administration of China (CAC), which is a strong indicator of the vitality of the market for LLMs in China.

Valuations in the Chinese LLM market have also experienced substantial growth, attracting investments from prominent players such as Alibaba and leading venture capital firms (e.g., Sequoia China, IDG Capital, ZhenFund).

Notable LLM Companies in China

Other areas of AI

Beyond the burgeoning LLM sector, other AI-related industries in China are also experiencing significant development and capital inflows:

Autonomous driving

Baidu has launched its 6th Generation Robotaxi (Luobo Kuaipao 萝卜快跑) in Wuhan, marking a critical moment in the evolution of smart mobility and unmanned vehicle travel service.

Tesla has begun its collaboration with Baidu to launch full self-driving (FSD) technology in China, which has been deemed as a significant milestone both for Tesla and for China’s autonomous driving industry.

WeRide and Momenta, both leading commercial-stage autonomous driving technology companies in China, have obtained approval from the competent PRC authorities for their respective IPOs in overseas stock exchanges.

Traditional automakers and new-energy vehicles such as Xpeng, BYD, Huawei and its partners, and GWM are also focusing on developing self-driving technologies.

Healthcare AI

QuantumPharm Inc. (also known as XtalPi), an AI-powered drug and new material discovery company, just completed its IPO on HKEX, which made it the first Specialist Technology Company listed under Chapter 18C of the Hong Kong Listing Rules that came into effect in March 2023.

Chinese companies, such as Insilico Medicine, are actively competing with their US counterparts in developing AI applications for rapid drug discovery.

Advanced semiconductors

Compared to the advanced semiconductors sector in the US, China’s AI chip companies are in the early stages, but there have been a few notable developments:

Cambricon and Intellifusion, focused on both AI chip and algorithm development, successfully completed their respective IPOs on the Shanghai Stock Exchange in 2020 and 2023, respectively.

Axera Semiconductor, a developer of AI perception and edge computing SoCs, secured RMB 800 million in a series A++ financing round in January 2022 from investors including Qiming Ventures, GGV, and Meituan.

Biren Technology, specialising in GPUs, closed its series B financing in 2021, raising a total of RMB 4.7 billion.

Neonexus, a company focused on AI chip integration, received approximately US$10 million in a series A and A+ financing round led by Primavera and Alibaba in August 2022.

Are there policies to help develop the Chinese AI sector?

China aims to become the world’s major AI innovation centre by 2030.

A development plan issued by the State Council outlines the goal of exceeding RMB 1 trillion (US$140.9 bn) in the AI core industry, with related industries surpassing RMB 10 trillion (US$1.4 tn).

In July 2022, China’s Ministry of Science and Technology and five other agencies unveiled a comprehensive plan to drive innovation and increased adoption of AI. The plan outlines principles and development goals for AI, targeting sectors like manufacturing, agriculture, finance, and transportation for accelerated AI adoption.  

China also considered enacting a comprehensive law to regulate artificial intelligence technology, according to the Legislative Work Plan for 2023 issued by State Council in June 2023.

A core principle of the plan is that AI development shall be enterprise-led, with the government playing a guiding role.

What is China’s approach to AI regulation?

China has adopted a cross-sectoral approach to AI regulation. Similar to the EU AI Act’s treatment of high-risk AI systems, China has set out an ex-ante filing obligation and full life cycle compliance requirements for certain types of AI services, such as generative and synthetic AI. It is worth noting that recently Chinese authorities also established a simplified filing process for online services that use filed LLM service as back-end API.

Additionally, different compliance requirements apply depending on the type of algorithm and use case.

Existing laws, such as the Personal Information Protection Law, the Cyber Security Law, the Employment Promotion Law, the Civil Code, various online content moderation rules, and other substantive regulations, intersect and continue to apply to AI. These laws ensure the protection of rights in the use of AI and the elimination and prevention of discriminatory and socially unacceptable practices.

What structures are used to invest in the Chinese AI sector?

Investments Into AI Companies Follows the Typical 'VIE Structure'

The investment structures and terms generally follow market standards for venture capital and growth financings in the Chinese technology, media, and telecommunications (TMT) sector: offshore USD investments into a Cayman Holdco and a “VIE structure” onshore in China to address PRC regulatory issues. These structures are widely used for both public and private TMT companies with a presence in China, and the contractual terms are also fairly standard.

Are there any restrictions on investing into the Chinese AI sector?

Foreign regulatory hurdles particularly stem from the US’s various measures to curb and restrict the growth of China AI sectors, such as “Reverse CFIUS” and the Executive Order on Preventing Access to Americans’ Bulk Sensitive Personal Data and United States Government-Related Data by Countries of Concern.

On the Chinese side, national security and data security reviews may be triggered concurrently. The national security review is initiated and managed by the National Development and Reform Commission (NDRC) and is distinct from the data security review managed by the Cyberspace Administration of China (CAC). However, the NDRC typically consults the CAC on cybersecurity, data compliance, and AI use cases. Importantly, not only onshore but also offshore transactions can trigger national security reviews by the NDRC.

An acquisition of “control” may also trigger a merger control filing. There are no specific merger control rules for AI, and so the review of acquisition and joint ventures concerning AI sectors will be carried out under the existing rules. We expect the latest Merger Control Review Guidelines for enforcement priorities in “technology” and “data” to be consulted in cases involving AI developers and deployers.

Export control-wise, specific types of AI-related technology, such as "AI interaction UI technology", are listed as restricted items in the Catalogue of Technologies Prohibited and Restricted from Export . This means that any domestic Chinese company wishing to export restricted AI technology to a foreign company, including exports within the same corporate group, needs to apply for an export control license with the competent local authority prior to that export.

What else distinguishes China's regulations on AI?

Key Compliance Requirements for Generative AI Services in China

Training Data

AI model developers are required to document the sources for training data and assess whether those sources comply with laws and regulations.

There has been a historical crackdown on web scraping, but recent interpretations of laws such as the PRC Cybersecurity Law and Personal Information Protection Law now consider the specific type and nature of data, making not all web scraping illegal. Although publicly available data is not strictly regulated from a cybersecurity standpoint, web scraping must adhere to intellectual property and data protection laws. However, ambiguities persist, notably regarding whether publicly disclosed personal information can legally be used for AI training under the “reasonable scope” defined by China’s Civil Code. The technical standard TC260-003 Basic Security Requirements for Generative Artificial Intelligence Service indicates the use of personal information scraped from public Internet sites for AI training purpose may be permissible so long as the relevant conditions, such as keeping a record of scraped data, can be met.

Copyright

AI providers must carefully manage the use of copyrighted materials such as images, videos, and texts for training AI, as Chinese courts typically view such use as an infringement. A significant case, the Ultraman Case , highlights the potential liabilities for AI providers using data scraped from the internet, especially when involving copyrighted video clips, photos and texts.

Use of Voice

Recent judicial decisions in China underscore the importance of respecting personal rights related to the human voice in AI training. Courts have held both AI providers and deployers liable for using voices without proper authorisation, though there’s still no consistent legal framework for assigning liability and risk among stakeholders.

Foreign AI Solutions

AI-based services in China utilizing foreign AI technologies face legal challenges if not filed with the CAC. Compliance with CAC’s registration and filing requirements is crucial for market entry, particularly for deep synthesis and generative AI technologies.

News and Social Media

Investors should be aware of the heightened scrutiny by Chinese authorities on AI applications used to generate news or rumors. Such uses are a law enforcement priority.

AI Output Copyrightability

In contrast to the U.S. position that generative AI outputs are generally non-copyrightable, a landmark decision by the Beijing Internet Court granted copyright protection to AI-generated content. The Court found that the output reflects originality and is a result of the user’s aesthetic choices and personal preferences.

Accountability & Liability

In contrast to the EU’s efforts to extend product liability to AI, the extent of liability for AI providers or deployers in China remains uncertain, with no current legal precedents. Although product liability laws apply to AI, defining what constitutes a “defect” is complex. Incidents involving autonomous driving are prompting legal developments, but existing laws hold the humans behind the AI system, such as doctors, accountable, potentially shielding AI providers from claims in areas like medical AI.

Also, AI providers are well advised to clearly label interactions as AI-generated and include disclaimers to mitigate potential claims. But the effectiveness of such disclaimers in court has yet to be fully tested.

Raymond Chan, Jeffrey Ding, and Gil Zhang are partners at Fangda Partners.

Sources

  1. The Information, (2024), Tecent, Alibaba Place Bets on Startups Racing to Become China's OpenAI
  2. South China Morning Post, (2024), China’s 4 new ‘AI tigers’ – Baichuan, Zhipu AI, Moonshot AI and MiniMax – emerge as investor favourites