The artificial intelligence (AI) landscape is experiencing a seismic shift, with Chinese technology companies at the forefront of this revolution. DeepSeek, a relatively unknown startup just a few months ago, has emerged as a formidable challenger to established AI giants, marking what many are calling China's "ChatGPT moment." This development is not only reshaping the global AI industry but also propelling Chinese internet companies into a new era of innovation and competition.

DeepSeek: Disrupting the AI Industry

DeepSeek, particularly with its R1 model, has sent shockwaves through the AI community. The company's sudden rise to prominence is attributed to several key factors:

1. Cost-Efficiency: DeepSeek R1 was developed at a fraction of the cost compared to its Western counterparts. The company claims to have spent only $6 million on computing power to train the model, significantly less than the estimated costs for models like ChatGPT or Google's Gemini.
2. Performance: Despite its lower development costs, DeepSeek R1 has demonstrated impressive capabilities. It has earned high marks for its performance, rivaling larger competitors across key benchmarks.
3. Pricing: DeepSeek offers substantially lower costs per token than OpenAI's models, making it a cost-effective solution for developers and enterprises.
4. Open-Source Approach: Unlike many proprietary models, DeepSeek has released its R1 model in a relatively open-source manner, allowing researchers and developers to access and modify the code freely.
5. Efficiency: DeepSeek R1 employs a Mixture of Experts (MoE) architecture, activating only 37 billion of its 671 billion parameters per forward pass. This design ensures scalability without proportional increases in computational cost.

The impact of DeepSeek's emergence has been profound. It has challenged the notion that developing cutting-edge AI models requires vast resources, prompting a reevaluation of AI development strategies globally. The company's success has also highlighted the potential for innovation under constraints, as DeepSeek developed its model despite U.S. chip export restrictions.

China's Internet Giants: Accelerating AI Investments

DeepSeek's breakthrough has catalyzed an AI arms race among China's internet giants. Companies like Alibaba, Tencent, Kuaishou, Baidu, and ByteDance are now doubling down on their AI investments, recognizing the potential for AI to transform their businesses and maintain global competitiveness.

These companies are among the major players investing heavily in AI research and development. This surge in investment is not just about keeping pace with global competitors; it's also driven by the unique opportunities and challenges in the Chinese market. With a vast user base and diverse industrial applications, Chinese companies are well-positioned to develop AI solutions tailored to local needs while also competing on the global stage.

Alibaba's latest and most advanced large language model (LLM) is Qwen 2.5-Max, released in January 2025. This model utilizes a MoE architecture, like DeepSeek, and has been trained on over 20 trillion tokens. Qwen 2.5-Max has demonstrated impressive performance across multiple benchmarks, outperforming models like DeepSeek-V3, GPT-4o, and Llama-3.1-405B in various tests. The model is available in different sizes, ranging from 3 billion to 72 billion parameters, and includes base and instruction-tuned versions. Qwen 2.5-Max excels in language understanding, coding, mathematics, and reasoning. It also features multimodal capabilities, allowing it to process text and visual inputs. Alibaba has made the model accessible through APIs on its generative AI development platform, Model Studio, enabling developers worldwide to leverage its capabilities.

Moonshot AI, an Alibaba-invested AI start-up, launched its latest model, Kimi k1.5, in January 2025. This multimodal reasoning model has demonstrated performance comparable to OpenAI's o1, particularly excelling in math tasks. Kimi k1.5 features an extended reinforcement learning (RL) context window of 128k tokens and employs advanced techniques to enhance RL policy optimization. The model has shown impressive results across various benchmarks, including a score of 77.5 on AIME and 96.2 on MATH 500. Kimi k1.5 also excels in multimodal reasoning tasks, such as MathVista, which require visual comprehension of complex subjects like geometry and IQ tests. Moonshot AI has introduced effective long2short methods, allowing the model to provide high-quality responses while significantly reducing inference costs. The company's focus on long-context processing and multimodal reasoning has positioned Kimi k1.5 as a versatile and powerful tool in the evolving AI landscape.

Tencent's flagship LLM is Hunyuan-Large, an open-source model with 389 billion total parameters and 52 billion activated parameters. Released in late 2024, Hunyuan-Large utilizes an advanced MoE architecture to achieve performance equivalent to a dense model with seven times the activation parameters. The model has demonstrated strong capabilities in both Chinese and English language processing, outperforming Meta's Llama 3.1 405B across many key benchmarks. Hunyuan-Large can handle contexts of up to 256,000 tokens, making it suitable for applications requiring extensive context and detailed analysis. Tencent has made the model available on developer-friendly platforms like HuggingFace and GitHub, encouraging widespread adoption and innovation in the AI community.

Kuaishou's leading LLM is KwaiYii, which has shown remarkable progress since its launch. As of early 2025, KwaiYii has significantly surpassed GPT-3.5 in overall performance and is approaching the capabilities of GPT-4 in certain metrics. The model excels in content creation, information consultation, and mathematical problem-solving, with its performance nearly matching that of ChatGPT. KwaiYii is part of Kuaishou's comprehensive AI model matrix, which also includes recommendation models and visual generation models. The company has been actively expanding its AI capabilities, recently introducing Kling, a text-to-video model capable of generating high-quality videos up to two minutes long with 1080p resolution. Kuaishou's AI innovations aim to reshape its content creation and commercial ecosystem, providing users with advanced tools for video generation and creative expression.

Baidu, China's leading search engine company, continues to make significant strides in AI with its Ernie LLM. The company plans to release a new version of Ernie in early 2025, aiming to enhance its performance, accuracy, and support for diverse user needs. As of late 2024, Ernie was handling an impressive 1.5 billion daily calls, processing over 1.7 trillion tokens of text each day - a 30-fold increase from the previous year. Baidu's CEO, Robin Li, predicts an "exponential" surge in AI applications by 2025, driven by the rapid deployment of generative AI across various industries. The company has also introduced new AI-powered tools, including I-RAG, a text-to-image generator, and Miaoda, a no-code application builder, as part of its strategy to commercialize AI technologies.

ByteDance's Doubao-1.5-pro, released in January 2025, represents a significant advancement in the company's AI capabilities. This model utilizes a sparse MoE architecture, achieving performance equivalent to a dense model with seven times the activation parameters. Doubao-1.5-pro has demonstrated impressive results across various benchmarks, surpassing models like GPT-4o and Claude 3.5 Sonnet in knowledge, coding, reasoning, and Chinese language processing. The model features a "Deep Thinking" mode, which enhances its reasoning abilities through extensive Reinforcement Learning techniques. ByteDance has also introduced multimodal capabilities to Doubao, including text, image, and audio processing, with an upcoming text-to-video feature. Despite its advanced capabilities, ByteDance has maintained competitive pricing for Doubao, making it widely accessible to individuals and businesses.

Implications for China's Internet Companies

The rapid advancement of AI models by Chinese companies has far-reaching implications:

These AI models will enable Chinese internet companies to offer more sophisticated and personalized services to their vast user base. From improved search capabilities to more engaging social media experiences, AI will transform how users interact with digital platforms.

AI models like DeepSeek R1 and Qwen2.5-VL can significantly boost operational efficiency. Alibaba's model, for instance, could revolutionize e-commerce by improving product recommendations and streamlining logistics.

The multimodal capabilities of these AI models open possibilities in sectors like healthcare, finance, and education. Chinese internet companies can leverage these technologies to develop innovative solutions tailored to local needs.

As these AI models continue to improve, they position Chinese companies to compete more effectively in the global market. This could lead to increased adoption of Chinese AI solutions worldwide.

Since ChatGPT launched in November 2022, the Nasdaq 100 Index has gained 81.6%, while the KraneShares CSI China Internet ETF (Ticker: KWEB) has only gained 17.5% over the same period.¹ We believe DeepSeek could be the ChatGPT moment for China internet companies and potentially lead to a revaluation of these companies that is more in line with their US peers.

Conclusion

In conclusion, DeepSeek's breakthrough has not only demonstrated China's capability to produce world-class AI models but has also spurred a new wave of innovation and investment across the Chinese technology landscape. As these companies continue to push the boundaries of AI technology, we can expect to see transformative changes in how digital services are delivered and consumed, both within China and globally. The race is on, and China's internet giants are poised to play a leading role in shaping the future of AI.

---

For KWEB top 10 holdings, risks, and other fund information, please click kraneshares.com/kweb.

Citations:

1. Data from Bloomberg as of 1/29/2025.

Index Definitions:

Nasdaq-100 Index (NDX Index): A stock market index comprised of equity securities issued by 100 of the largest non-financial companies listed on the Nasdaq stock exchange. It is a modified capitalization-weighted index.

Definitions:

Large Language Model (LLM): A type of artificial intelligence that is specifically designed to understand and generate human language by being trained on massive amounts of text data, allowing it to perform tasks like writing, translating, summarizing, and answering questions in a human-like way; essentially, it's a complex AI system that can process and generate text with a high degree of accuracy and fluency.

Mixture of Experts (MoE): A machine learning technique where multiple specialized models, called "experts," work together, with a gating network deciding which expert is best suited to handle each input, essentially dividing a complex problem into smaller, more manageable subtasks based on specific expertise; this allows for more efficient and accurate predictions compared to a single monolithic model.

Reinforcement learning (RL): A machine learning (ML) technique that trains software to make decisions to achieve the most optimal results. It mimics the trial-and-error learning process that humans use to achieve their goals.e other lists maintained by the U.S. Government for sanctions or export control measures, inclusion in the CMC List relates only to U.S. defense procurement, which does not impact the business of the Group.