“Is the US Lead Coming to an End?” Kimi K3 Shock Freezes AI Investment Sentiment as China Moves to Stabilize Markets and Support the Industry

Chinese state-backed investment firms step in to contain the global technology-stock sell-off
Low-cost, high-performance Kimi K3 shakes AI investment sentiment and pressures US leaders
China details a long-term AI development strategy, signaling a potential market realignment

Large Chinese state-backed investment companies have injected substantial amounts of capital into the country’s financial markets. As technology stocks came under selling pressure following the shock caused by Kimi K3, a high-performance open-weight model developed by Chinese artificial intelligence startup Moonshot AI, Beijing mobilized policy-linked funds to prevent the correction from spreading into a broader market crisis. Market observers increasingly expect the adjustment to represent more than a temporary decline in share prices, arguing that the emergence of increasingly competitive Chinese models could challenge an AI industry structure previously centered on expensive, proprietary systems developed in the United States and accelerate a broader realignment around China’s AI ecosystem.

AI Investment Sentiment Freezes

According to Reuters on July 20, China Reform Holdings Corporation and China Chengtong Holdings Group, two of the country’s largest state-backed investment companies, issued separate statements confirming additional purchases of shares in state-owned enterprises and exchange-traded funds. China Reform Holdings said it had completed the deployment of a total of RMB 50 billion, or approximately $7 billion, combining its own capital with funding made available through the People’s Bank of China’s special relending facility for share repurchases and equity acquisitions. The funds were used to support share buybacks by major listed subsidiaries within the group and to acquire stakes in centrally administered state-owned enterprises. China Chengtong also purchased approximately RMB 10 billion, or about $1.4 billion, in Chinese equities through the open market and announced that it would continue buying stakes in state-owned enterprises, strategically important high-technology companies, and broad-market ETFs.

The intervention followed a technology-stock sell-off that had spread across global financial markets. During the preceding several days, skepticism grew rapidly in the US market over the sharp rise in the valuations of AI and semiconductor companies. Major technology groups have committed enormous amounts of capital to data centers and AI chips, but investors remain uncertain about when those expenditures will begin generating stable and recurring cash flows. The selling pressure became clearly visible on July 15 and reached its peak on July 17, when the Nasdaq Composite fell 1.4% and the S&P 500 declined 1%. AI semiconductor companies including Nvidia, AMD, and Applied Materials all weakened, while the Philadelphia Semiconductor Index lost approximately 11% during the week and fell more than 20% from its late-June high.

The shock that originated in the United States quickly spread through Asian markets. Taiwan’s equity market fell by more than 6% on July 17, while Japanese stocks declined by approximately 4%. South Korea absorbed the impact somewhat later, with the Kospi dropping 4.5% on July 20. China’s Shanghai Composite has declined by more than 7% since the beginning of July, while the technology-focused ChiNext Index in Shenzhen has fallen by more than 20%. The STAR 50 Index, which includes many of China’s most advanced semiconductor companies, has also dropped approximately 25% from its July peak. The purchases by state-backed investment companies were therefore more than an attempt to acquire temporarily undervalued assets; they represented a policy intervention intended to prevent the global technology-stock correction from developing into a wider crisis across China’s capital markets.

Kimi K3 Shock Unsettles Global Markets

Some analysts argue that the rapid technological progress of Chinese AI companies contributed significantly to the correction by destabilizing confidence in the investment assumptions supporting US technology valuations. In particular, Moonshot AI’s Kimi K3 has disrupted sentiment across the AI sector. Kimi K3 is an exceptionally large open-weight model containing 2.8 trillion parameters, a commonly used measure of the scale and computational capacity of an AI system. Market participants have assessed the model as substantially larger than OpenAI’s earlier GPT-4 and potentially comparable in scale to the most advanced models offered by Google Gemini and Anthropic Claude. China has therefore demonstrated that it can independently develop a class of extremely large model once considered the exclusive domain of the largest American technology companies.

Kimi K3 is priced at $3 per million input tokens and $15 per million output tokens, making it one of the more expensive Chinese models while remaining cheaper than the most advanced US alternatives. Its performance has also been assessed as broadly competitive with leading American systems. Moonshot AI has emphasized that Kimi K3 can challenge major US models in complex coding, web navigation, and tasks requiring sustained operation over long periods. In benchmark testing conducted by AI evaluation organization Artificial Analysis, Kimi K3 ranked third behind Anthropic’s flagship Claude Fable 5 and OpenAI’s latest flagship model, GPT-5.6 Sol. Its open-weight structure also provides a considerable accessibility advantage. Moonshot AI has allowed users to access Kimi K3 free of charge after registering on its website, without requiring separate payment or complicated approval procedures, and plans to release the model’s complete weights on July 27 so that developers can use, modify, and adapt it independently.

The emergence of Kimi K3 has created significant uncertainty in global markets. A Chinese startup with substantially less capital and more limited access to advanced semiconductors than American technology groups has produced a model approaching the capabilities of leading US systems, raising concerns that the enormous capital expenditures of American companies may fail to preserve a lasting technological gap or generate monopoly-like returns. “As Chinese models become more capable and applicable across a wider range of tasks, companies and developers will inevitably become less dependent on expensive American systems,” one market participant said. “US companies are likely to face pressure to reduce usage fees, while the period required to recover their enormous investments in data centers and AI semiconductors could become substantially longer.”

China Details Its AI Industry Strategy

The global presence of Chinese AI is expected to continue expanding as Beijing more clearly expresses its intention to support the industry’s development. On July 17, the National Development and Reform Commission and other Chinese government departments jointly announced an Action Plan on AI Cooperation and Development during the 2026 World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance in Shanghai. Through the plan, China proposed creating a framework for the international sharing of data, one of the most important resources required for AI development. The government called for greater cross-border data flows and proposed establishing “cross-border trusted data spaces” in selected sectors to create a more efficient and secure system for international data exchange.

The plan also proposed jointly building high-quality AI training resources and industrial datasets, while developing and sharing multilingual corpora among participating countries. The initiative appears intended to diversify an international AI data environment that has largely developed around English-language materials. By supporting the creation of datasets in a wider range of languages and industrial contexts, China could reduce the structural advantage enjoyed by models trained primarily on English-language information and improve the international competitiveness of Chinese systems in markets that remain underserved by leading US platforms.

Another central element of the action plan is the joint development of an international AI open-source community and the expansion of cooperation among developers. China proposed sharing general-purpose large language models, core algorithms, development tools, and software components, while creating international open-source compliance systems and common safety guidelines. The plan places particular emphasis on helping countries pursue AI innovation suited to their own economic and social circumstances using open-source and open-weight models. The strategy is widely interpreted as an attempt to support Chinese AI companies that have released a series of openly accessible models and are building increasingly extensive developer ecosystems around them.

China also announced plans to expand international cooperation under the concept of “AI+,” through which AI systems would be integrated into established industries and social institutions. Beijing proposed creating cross-border industrial cooperation platforms and supporting developing countries in adopting AI models, computing infrastructure, and application services suited to their domestic industrial conditions and development needs. The initiative therefore extends beyond the simple export of individual AI models and appears designed to spread a China-centered technology ecosystem through infrastructure, software, data standards, and commercial applications. China additionally emphasized the need for a joint international education system for AI talent, proposing expanded cooperation among universities, research institutions, and companies to establish programs for the collective training of AI specialists. Together, these measures suggest that China is seeking not merely to narrow the model-performance gap with the United States, but to construct an alternative global AI ecosystem capable of competing on cost, accessibility, openness, and institutional reach.