Key Points:
- Zhipu AI’s founder Tang Jie published an internal memo defending open-source AI, arguing that transparency offers stronger safeguards than restrictions.
- The defense comes as both Washington and Beijing consider restricting access to advanced open-weight models due to biosecurity and cyber risks.
- Zhipu recently surpassed a HK$1.27 trillion ($163 billion) market capitalization in Hong Kong, surpassing tech giants Xiaomi and Baidu.
- The company is launching its “Touch High” plan, temporarily setting aside short-term profits to focus on foundational AGI research.
The global debate over the security and accessibility of cutting-edge artificial intelligence has taken a major turn, as the leader of China’s most valuable AI startup issued a powerful defense of open-source software. Tang Jie, the founder and chief scientist of Beijing-based Zhipu AI, released an internal memo to employees titled “The Giant Wave Has Arrived,” affirming that frontier artificial general intelligence must remain open and accessible to everyone. The strategic memo outlines the company’s counter-intuitive decision to temporarily set aside short-term monetization in favor of foundational scientific research. The move places a major open-weights pioneer directly at odds with both Western export controls and potential domestic regulatory clampdowns.
This bold ideological commitment comes from a position of extraordinary financial strength. Operating globally under the brand Z.ai and listed on the Hong Kong Stock Exchange under the ticker 2513, the company has emerged as a stock market sensation. Since its initial public offering in January 2026, the company’s share price has surged by more than 2,200% from its debut. This blistering rally has pushed the company’s market capitalization past HK$1.27 trillion ($163 billion), eclipsing established Chinese technology giants like Xiaomi and Baidu. This massive capital market success provides the startup with a historic cash cushion to fund its next generation of research without immediate profit-generating pressures.
Rather than capitalizing on its high valuation to aggressively expand commercial software sales, the company is launching its ambitious “Touch High” (reach-up) program. The two-year initiative marks a deliberate strategic pivot, temporarily pausing the monetization of its highly successful coding applications to fully return to foundational Artificial General Intelligence research. The firm is reallocating the bulk of its newly acquired capital to build four core engines: long-horizon tasks, autonomous agentic systems, fully self-training models, and extreme safety governance. The goal is to scale up the raw intelligence of its models to break past the boundaries established by closed-source Western systems.
In his memo, Tang Jie articulated a dual-sided philosophy of technical advancement, arguing that high-end capabilities and accessibility are two sides of the same coin. He wrote that while one hand must reach upward to challenge the limits of machine intelligence, the other hand must lay the road downward, making those frontier capabilities as open and widely accessible as possible. He emphasized that the technological heights the company touches belong to all of humanity, and the road it builds belongs to everyone. This pro-openness stance argues that real AI security stems from public participation, transparency, and collaborative global oversight rather than restrictive security barriers.
The startup’s vocal defense of open-source software occurs against a highly complex, hostile geopolitical backdrop. Recently, the United States administration took the unprecedented step of ordering developer Anthropic to suspend foreign national access to its newest Fable 5 and Mythos 5 models. Because these high-end systems can autonomously identify and exploit zero-day software vulnerabilities, Western regulators are taking aggressive steps to contain them. These strict export controls have locked out major technology labs in China, Europe, and the Middle East, sparking intense concern regarding technological sovereignty and the danger of relying on proprietary U.S. software.
The company’s response to these Western export blocks was swift and decisive. Just days after Washington restricted Anthropic’s models, the Beijing-based startup released its latest flagship model, GLM-5.2, under the highly permissive, open-source MIT license. By releasing the model weights with no usage restrictions and no regional locks, the company provided global software developers with a free, highly capable alternative. Built on a massive Mixture-of-Experts architecture containing 744 billion total parameters, GLM-5.2 is designed specifically for complex systems engineering and long-horizon tasks, successfully running on a solid, highly reliable 1-million-token context window.
The open-source release has quickly captured the attention of major enterprise developers in Silicon Valley. Independent benchmark tests reveal that GLM-5.2 is already performing on par with Anthropic’s flagship Claude series in advanced coding, while trailing other premium models by only a single percentage point on long-horizon SWE-bench tasks. More importantly, the model delivers these cutting-edge capabilities at a fraction of the cost of its closed-source competitors. For example, database software giant Snowflake recently confirmed that its engineering teams completed complex data transformation tasks using the open-weight model at half the operational cost of using proprietary American APIs.
However, the startup’s open-source philosophy is facing a delicate regulatory balancing act at home. As open-weight models rapidly close the performance gap with closed-source frontier systems, policy analysts warn that governments in both the West and the East are realizing that open-weight code is becoming too dangerous to leave unrestricted. Recent reports suggest that Beijing is currently considering regulatory measures to restrict foreign access to China’s most advanced open-source AI models due to potential cyber and biosecurity risks. If these rules materialize, they would force local developers to geofence their software, creating a severe contradiction for a firm built on the belief that code belongs to everyone.
Ultimately, the clash over the future of open-source AI highlights the complex, multi-speed nature of the modern digital economy. While venture capital funds and corporate boards continue to chase short-term software profits, the developers who can build open, trusted, and highly accessible model architectures are finding a viable path to long-term market dominance. By leveraging its historic trillion-dollar valuation to fund foundational research while keeping its models open to all, the South Korean and Chinese innovators are proving that tech sovereignty cannot be achieved through restrictions alone. The coming years will reveal how successfully the firm can navigate these rising geopolitical barriers, but the foundations for an open-weight AI future are now firmly in place.
