Hong Kong, an international financial center, has also been at the forefront of global innovative technology. As early as the 1960s, Hong Kong began to develop the chip industry. Today, amidst the surging wave of artificial intelligence, Hong Kong is leveraging its ample capital, knowledge, technology, and talent advantages to build a new high ground for the AI technology industry.

Computing power is the upstream infrastructure of the AI industry, and Hong Kong's foray into the AI industry inevitably relies on solid and reliable computing power support. This year, Huawei Cloud began providing Ascend AI Cloud Services in Hong Kong, aimed at providing readily available AI computing power for large model training and inference.

Huawei Cloud, with continuous innovation based on over 30 years of ICT technology accumulation and leading experience, will not only continue to empower Hong Kong's intelligent transformation but also provide a digital base for Chinese enterprises pursuing global expansion, accelerating the release of digital productivity across all industries.

Nowadays, under the wave of digital economy globalization led by 5G and AI, "going global or being left behind" has become a consensus among Chinese enterprises. Many enterprises have converged overseas to explore new horizons and find incremental spaces.

Concurrent with the wave of Chinese enterprises going global, AI is stirring up the nerves of various industries. In 2023, AI giants like OpenAI's GPT-4 emerged and iterated rapidly, ushering in the first year of AI for human society. If 2023 was the first year of generative AI, then 2024 will be the first year of enterprise-grade AI. An increasing number of enterprise-grade AI applications are reshaping corporate productivity, with their powerful data processing, intelligent analysis, predictive capabilities, and personalized user experiences, ushering in profound industrial changes and the "AI era".

As going global has become an inevitable choice for enterprises, AI is also reshaping the business models and development strategies of overseas enterprises.

In the field of cross-border e-commerce, according to the Ministry of Commerce, China's cross-border e-commerce imports and exports reached 577.6 billion yuan in the first quarter, an increase of 9.6%, with rapid growth in e-commerce imports and exports with partner countries such as Hungary, the United Arab Emirates, Cambodia, and Brazil. Among them, the most representative examples are cross-border e-commerce platforms such as SHEIN, TEMU, and TikTok Shop, which are expanding their territories and capturing market share in overseas markets.

For cross-border e-commerce platforms and sellers, it is necessary to gain deep insights into local demand and launch localized products tailored to local conditions in various aspects such as design, product selection, operations, customer service, marketing, and supply chain management.

For example, SHEIN, a cross-border e-commerce unicorn that specializes in a flexible supply chain system, is a veritable AI-driven e-commerce company. Two key capabilities possessed by SHEIN: a quick-turn production model for small orders and digital management tools, both rely on data analysis and trend forecasting to generate design samples, ensuring that SHEIN's core fast fashion category can compress its production cycle to 5-7 days and become a fast fashion giant deeply loved by millennials.

In the field of automotive exports, AI is also rewriting the rules of the game. In 2023, China's automobile exports surpassed all other countries for the first time, driven primarily by the strong growth of new energy vehicles. However, China's overseas automotive market is not without challenges, facing strong competitors and numerous barriers. Nowadays, with the support of IoT, AI, and big data technologies, the ultimate battle between Chinese automotive brands and overseas giants is not only a competition of products and technologies but also a competition of application ecosystems—and AI's empowerment of human-vehicle interactive intelligent operating systems is undoubtedly crucial.

For example, at the Canton Fair held on April 15, iFLYTEK unveiled its latest, universally accessible large model, iFLYTEK Spark V3.5, trained on the nationally produced computing power platform "Feixing One." The multi-language products powered by the Spark large model are fully supporting China's automotive globalization strategy, currently providing in-vehicle voice solutions in 23 mainstream foreign languages, helping Chinese automakers provide highly localized smart vehicle interaction experiences for overseas users.

In the gaming and entertainment industry, AI's impact on the gaming industry has gradually evolved from cost reduction and efficiency enhancement to product innovation. Hu Zhipeng, Senior Vice President of NetEase Group, once told Xiaguang News, "Games are the most advanced in AI large models, as they are a natural integration platform for new technologies."

On the R&D and creation side, AI can assist game developers in character design, plot setup, and scene construction, greatly enhancing production efficiency and accelerating game development and release. On the operational side, AI can accurately identify overseas player needs through powerful and efficient data search and analysis capabilities, as well as multi-language processing capabilities, providing more precise personalized support and enhancing player identification and connectivity.

Taking the Middle East, which differs significantly from China's cultural context, as an example. When developing a game product for the Middle East market, all language rules and presentation logic need to be changed due to the right-to-left presentation of Arabic. Some game engine development, code, drawing tools, and even the simplest Excel, Word, and PPT, have poor compatibility with Arabic, possibly requiring the development of a dedicated font. In terms of worldview construction, user interface visualization, and localized operations, the cultural barriers faced are even more complex. The empowerment of AI large models can undoubtedly enable overseas game manufacturers to overcome barriers, deeply root themselves locally, and achieve long-term development tailored to local conditions.

Middle Eastern women experiencing esports games

It is evident that with the diversification, efficiency, and refinement of the needs of overseas enterprises, AI's advantages in overseas services are becoming increasingly important. AI empowerment is the key to enabling enterprises to maintain competitiveness in the fierce competition in overseas markets.

With the surging wave of artificial intelligence, the demand for computing power is booming, and the technical foundation of the AI era—the intelligent computing power industry—has become a key area of competition among major powers. According to the 2022 Global Computing Power Index Assessment Report, every one-point increase in the computing power index will drive a 3.5‰ increase in a country's digital economy and a 1.8‰ increase in GDP.

The deep integration and innovation of AI and industries have also led to explosive growth in enterprises' diversified demand for computing power. According to a report released by OpenAI, the computing power used for AI training tasks has doubled every 3-4 months since 2012; from 2012 to 2018, AI computing power demand increased by 300,000 times. Other studies predict that from 2018 to 2030, the demand for computing power in smart mobility will increase by 390 times, and the demand for computing power in smart factories will increase by 110 times. Huawei estimates that compared to 2020, the demand for computing power brought about by the AI boom will increase by 500 times in 2030, presenting significant investment opportunities.

As an AI superpower, China has witnessed the emergence of numerous AI large models. According to the "Beijing AI Industry Large Model Innovation Application White Paper (2023)," as of October 2023, there are a total of 254 large model manufacturers and universities/institutes in China with over 1 billion parameters, distributed across more than 20 provinces/regions. Both the inference and training scenarios of large models require significant computing power, such as large model training, which typically requires thousands of GPU cards and TB-level data, taking several months to train. Then, other users conduct reinforcement learning based on pre-trained large models, requiring hundreds of GPU cards and taking one to two months.

For overseas enterprises, whether it is product development, supply chain management, local operations, or intelligent marketing, AI can enhance the process efficiency and digital capabilities of every link.

Taking supply chain management as an example. Nowadays, the global supply chain landscape is undergoing profound restructuring, and Chinese enterprises are leveraging their existing capabilities to participate in this reshaping, achieving supply chain localization in various regions such as Europe, the Middle East, Southeast Asia, and Latin America, establishing a multi-dimensional and comprehensive industrial and supply chain system. Therefore, in numerous industries such as manufacturing, consumption, construction, medicine, automobiles, food, and new energy, from R&D to production to sales, enterprise supply chains have been greatly extended, with complexity increasing exponentially. AI technology can make enterprise procurement management and processes more intelligent, automated, and efficient, improving supply chain visibility, flexibility, and adaptability.

However, without sufficient computing power support, enterprises will face difficulties in areas such as product standardization, logistics scheduling, and intelligent risk control. Under the competitive landscape of supply chain decentralization, regionalization, and diversification, they will be caught in a passive situation.

Currently, NVIDIA and AMD, two American companies, are the leading global GPU manufacturers, accounting for the vast majority of the global GPU market share. However, due to geopolitical influences, the United States has repeatedly imposed chip restrictions on China since 2022, making it difficult for Chinese enterprises to obtain computing power support from American companies.

In October 2022, the Biden administration announced broad chip export restrictions on China and convinced the Netherlands and Japan to introduce similar measures. In October 2023, the United States updated its export controls on advanced semiconductors and computing equipment to China, and NVIDIA's China-specific GPU products, A800 and H800, were banned in China starting November 17. On March 29, 2024, the Biden administration further revised its chip restrictions on China, expanding the restrictions on chip exports to China to a broader range of consumer electronics and adopting a "case-by-case review" policy for AI semiconductor products exported to China, comprehensively restricting the sale of NVIDIA, AMD, and more advanced AI chips and semiconductor equipment to China.

Under the US technology blockade, Chinese enterprises are facing severe computing power anxiety. Since the second half of 2022, many Chinese companies have been frantically purchasing AI chips from NVIDIA and others, with the price of a single GPU once soaring to nearly 400,000 yuan. In August 2023, two months before the US ban on NVIDIA's A800 and H800 chips, Chinese internet giants Alibaba, Baidu, ByteDance, and Tencent spent a whopping $1 billion to purchase about 100,000 NVIDIA A800 processors. The enormous market demand and stringent restrictions have even given rise to a gray industry of "NVIDIA scalpers".

In the era of artificial intelligence, the foundation of computing power is the core driving force for technological progress. Undoubtedly, for Chinese enterprises, achieving overtaking through independent technological innovation is the long-term plan for the development of China's AI industry.

Faced with the dilemma and anxiety of insufficient computing power for Chinese enterprises, Huawei Cloud Ascend AI Cloud Services can provide readily available and robust AI computing power, becoming a solid computing power foundation and building another choice for the world.

AI computing power has extremely high requirements for data centers. Due to the significantly higher power density of AI servers compared to general-purpose servers, the power consumption of a single cabinet is 6-8 times higher than before and requires a dedicated liquid cooling system for heat dissipation to reduce PUE (Power Usage Effectiveness). Additionally, large model training often requires thousands or even tens of thousands of cards, posing significant challenges in terms of server interconnectivity, latency, reliability, and operations and maintenance.

Huawei Cloud has deployed four major AI cloud computing centers and over 30 edge computing centers in Gui'an, Ulanqab, Wuhu, and Hong Kong. Through AI cloud services, enterprises do not need to build or retrofit traditional data centers, invest in general AI technologies, or worry about operations, maintenance, and security pressures. They can readily access robust AI computing power and rich AI cloud services on demand.

Specifically, the differentiated competitiveness of Huawei Cloud Ascend AI Cloud Services is reflected in five aspects:

1. Convenience: Ready-to-use cloud computing power, eliminating the need to build or retrofit data centers. Enterprises can seamlessly access AI cloud computing nodes in Gui'an, Ulanqab, Wuhu, Hong Kong, and other locations, with readily available computing power at their fingertips.

2. Ease of Use: An efficient and user-friendly full-stack AI development platform, eliminating the need to invest in general AI technologies. Huawei Cloud Ascend AI Cloud Services possess full-stack service capabilities ranging from cloud computing power, model development, model hosting, to the ecosystem, enabling "integrated training and inference"—sharing resource pools between training and inference tasks, maximizing resource utilization efficiency through rapid resource scaling and elasticity.

3. Stability: Reliable and stable operation for large-scale training, eliminating the need for enterprises to worry about operations, maintenance, and security. In 2023, Huawei Cloud had zero P1 incidents, with industry-leading fault recovery time and no data breaches. A comprehensive security system safeguards enterprises' core data.

4. Openness: Fostering a fertile soil for hundreds of models and thousands of applications, eliminating the need to worry about model development and application deployment. The Huawei Cloud Ascend AI Cloud Services community provides over 70 optimized and enhanced open-source large models, enabling one-stop fine-tuning of customer-owned data and customization of downstream applications, linking pre-trained large models with rich cloud components, and shortening service deployment cycles. It connects with the cloud market to enable full asset monetization channels, collaborates with over 15 service partners to build end-to-end delivery capabilities for large models, and works with over 30 consulting and integration partners to build large model service capabilities.

5. Flexibility: Collaborative computing power across cloud, network, edge, and endpoints. Currently, the capabilities of large models are sinking to the edge, and Huawei Cloud leverages its cloud-network-edge collaboration advantages to enable endpoints to acquire more intelligent capabilities.

Huawei Cloud Ascend AI Cloud Services' readily available and robust computing power has already empowered numerous overseas enterprises to expand globally.

For example, iFLYTEK, a globally renowned leader in intelligent voice and AI, proposed to rapidly deploy the industry's first ultra-large-scale cluster, challenging to complete within a month, and further enhancing training efficiency to rapidly complete large model training.