China AI ETFs Rally as Kimi 3 Model Boosts Cloud Demand
China AI software and cloud-focused ETFs had a strong July, benefiting from the release of Moonshot AI’s Kimi 3 Large Language Model. As shown in Figure 1, the KraneShares CSI China Internet ETF (KWEB) significantly outperformed both the Roundhill Magnificent Seven ETF (MAGS), which is dominated by U.S. hyperscalers, and the iShares MSCI China ETF (MCHI), which represents the broader Chinese equity market. CFRA continues to be bullish on the U.S. hyperscalers, but the success of China’s open-weight models gives investors more diversification options in the AI model and cloud computing trade.
Kimi 3 Likely to Spur Cloud Demand and Drive Operational Efficiencies
While Kimi 3 is a threat to the model businesses of the Chinese AI firms held in KWEB, this impact will be offset by external cloud demand and internal AI adoption. Kimi 3 is a 2.8 trillion parameter open-weight model that was released by Chinese firm Moonshot AI on July 16, 2026, and its performance is very close to leading U.S. models like Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6. Although Kimi 3 competes with the Hunyuan model from Tencent, CFRA has a Strong Buy rating on Tencent since it benefits from the adoption of Kimi 3. Tencent has pivoted its FinTech and Business Services segment towards external cloud and AI enterprise services. Additionally, it is also pursuing AI-led monetization across its advertising and gaming businesses. We see a similar dynamic with Alibaba, whose Qwen model will be challenged by Kimi 3. However, its Alibaba Cloud business benefits from AI training and inference demand flowing through its infrastructure. Both Tencent and Alibaba have equity stakes in Moonshot AI, providing additional upside.
The proliferation of low-cost open-weight models also benefits Chinese firms with large tech-enabled consumer franchises. Meituan, a top 5 holding in KWEB, can use cheaper open-weight models to drive operational efficiencies in its core food delivery business, including route optimization, rider dispatch and dynamic pricing. It can also be used to build agentic workflows for its consumer-facing applications. Meituan is also an investor in Moonshot AI through its investment division. CFRA expects NetEase, a provider of online PC and mobile gaming, to have greater visibility on its next slate of titles, such as Sea of Remnants, Blood Message and Ananta. Open-weight models should be a useful tool in that development and growth process through coding assistants and AI-assisted content creation. CFRA also has a Buy recommendation on JD.com, a provider of e-commerce and logistics services and another top 10 holding in KWEB. In our view, JD.com’s AI adoption will further increase its revenue potential by improving user and merchant experiences. This includes the integration of AI into its search and recommendation systems, as well as its merchant tools, including order management and data analysis. JD.com can also use AI to improve productivity in areas such as video content review and software development.
China’s July Gains Were Primarily in AI & Software-Enabled Businesses
The July 2026 price gains in China ETFs were primarily in the mega-cap AI and software-focused funds. KWEB, which is concentrated in AI-enabled stocks, was up 16.4% in U.S. dollar terms, and the Hong Kong-focused iShares China Large-Cap ETF (FXI) was up 15.5%. FXI holds large cap Chinese equities listed on the Hong Kong Stock Exchange, with a focus on state-owned banks as well as tech-enabled communication services and consumer firms. In contrast, MCHI also holds mainland China-listed A-shares and B-shares, which didn’t benefit as much from this AI-driven rally in July. The hardware-focused VanEck China Semiconductor ETF (SMHC) was down 35% in July 2026 due to a global pullback in semiconductor stocks amid concerns about AI-related capital expenditures by mega-cap firms and supply chain pressures stemming from international trade restrictions.
The token market share of Chinese-origin models has accelerated dramatically in the trailing 12 months. Open Router processes over 20 trillion tokens per week across hundreds of AI models. Its usage rankings provide the clearest picture of what developers choose when given access to every major model through a single API. According to data from Open Router, Chinese models had a roughly 60% market share of total token usage on its platform as of June 2026. This is up from merely 1.2% in October 2024. Chinese model market share on Open Router increased meaningfully to 10% in March 2025 after the launch of DeepSeek V3 and had increased to around 50% by April 2026 (see Figure 3 for provider share in April 2026).
The high performance of the Kimi 3 model will further accelerate the adoption of Chinese open-weight models, benefiting the holdings in KWEB. The decision to allow downloadable model weights is proving to be a winning distribution strategy for the Chinese model providers. These models are also cheaper to serve than their U.S. counterparts since export controls forced the Chinese providers to be more efficient. This combination of open weights and lower costs in the Chinese models has resulted in U.S. model provider market share falling from over 70% in 2024 to between 30%-40% in mid-2026. However, U.S. providers continue to capture a large share of model-related revenue.
Looking Ahead
A key variable to watch will be Chinese government regulation as AI adoption increases in the country. The government could look to restrict international access to downloading of its open-weight models, which would negatively impact the holdings in KWEB. State intervention in China can often prioritize national security and data privacy concerns over short-term corporate profitability. Major players like Tencent, Alibaba, and Baidu already operate under tight security and content monitoring rules. Geopolitical tensions between the U.S. and China will be another key area to monitor, particularly around the export of high-performance chips from the U.S. to China, and the export from China of critical minerals that impact the technology industry.
