In a windowless room at Tsinghua University in 2019, a small team of researchers planted the seed for what would become one of China’s most formidable artificial intelligence companies. Fast forward to today, and Zhipu AI stands as a testament to China’s technological resilience and innovation—thriving despite American attempts to ground its ambitions.
The unstoppable innovation machine
Despite intensifying sanctions from the United States aimed at thwarting China’s technological advancement, the Chinese innovation machine shows no signs of slowing. Washington’s stated goal of preventing Beijing from outpacing America in the tech domain—already achieved in several instances—seems increasingly quixotic as Chinese firms continue to forge ahead with new breakthroughs.
The recent launch of Zhipu AI’s latest artificial intelligence model offers a stark reminder of China’s resilience. This burgeoning startup has thrown a spanner into the global AI landscape with the unveiling of its new AI agent, AutoGLM Rumination.
During a launch event in Beijing on Monday, Zhang Peng, Zhipu’s chief executive, outlined the agent’s capabilities for conducting in-depth research, generating comprehensive reports, and assisting users in planning a myriad of activities and tasks—all without charge.
David versus Goliath
The narrative of Chinese tech firms challenging American dominance despite facing overwhelming odds continues to gather pace. Zhipu AI boasts two major models, with AutoGLM Rumination not only excelling in deep research but also offering functionalities such as web searches, travel planning and research report writing.
The timing could hardly be more pointed. AutoGLM Rumination’s debut comes shortly after another Chinese AI sensation, Manus, which sparked interest with its claim of being the world’s first general AI agent. While Manus charges a steep monthly fee of up to $199, Zhipu’s offering will be available free of charge through the company’s official channels, including its GLM model website and mobile app.
Established in 2019 as a spin-off from a Tsinghua University laboratory, Zhipu AI has quickly risen to prominence among China’s leading AI startups. The company’s GLM series of models has garnered attention, with its latest large language model, GLM4, claiming to outperform OpenAI’s GPT-4 on several key benchmarks.
Swimming against the tide
The geopolitical context of Zhipu’s rise cannot be overlooked. In January, the American Commerce Department placed the firm and its subsidiaries on the export control entity list, effectively blocking them from obtaining American-made components.
This move came just weeks after Manus AI stunned the global AI community by outperforming OpenAI’s deep research tool on the GAIA benchmark, which evaluates AI agents on solving real-world problems.
Zhipu AI strongly objected to the US government’s stance that exhibited American insecurities regarding China’s technological progress and innovation.
Yet, despite this setback, Zhipu AI’s success story continues to unfold. The company’s recent $69m Series D funding round, following a $137m raise, places it at a valuation of $2.74 billion. Notably, Zhipu AI has also received backing from tech giants Tencent and Alibaba, as well as several Chinese state-owned entities.
Zhipu AI’s success proves the Chinese logic of building another door when one door closes, and their firm belief that necessity can breed any form of innovation, which is shown in their ability to excel without American chips, which countries like India can’t achieve despite having greater access to American chips and technology.
The broader ecosystem flourishes
Zhipu is merely one player in China’s increasingly vibrant AI landscape. The ecosystem is currently experiencing a surge of growth, with giants like Alibaba and Tencent unveiling new AI models at an unprecedented pace.
Recently, Alibaba introduced QwQ 32B, an AI model with 32 billion parameters, which, despite its relatively modest scale, is reported to achieve performance comparable to DeepSeek-R1, which boasts 671 billion parameters (with 37 billion activated parameters).
Meanwhile, the latest model from DeepSeek, the DeepSeek V3-0324, has now claimed the top spot in benchmarks among all non-reasoning models. Reuters has reported that DeepSeek is planning to release R2 “as early as possible,” with the company considering an accelerated launch timeline.
The dragon’s resilience
The story of Chinese AI companies like Zhipu and DeepSeek underscores the nation’s commitment to technological advancement, even in the face of external challenges. This resilience reflects a broader pattern in China’s approach to innovation—adapting to constraints rather than being defined by them.
As Western governments continue to grapple with how to respond to China’s technological rise, Chinese firms are demonstrating that borders or regulations cannot easily contain innovation. The country’s AI sector appears determined to chart its own course, crafting technological solutions tailored to domestic needs while increasingly setting global standards.
For the rest of the world’s tech sector, watching this unfold from afar, the message is clear: in the new multipolar technological world, dismissing Chinese innovation as merely derivative or dependent on Western technology would be a dangerous miscalculation. The dragons have learned to fly on their own.
