Farsight, July 11 – A report from Emerging Intelligence reveals that Tang Jie, the founder of Zhipu, issued an internal memo today, entitled 'The Giant Wave Has Arrived', announcing the initiation of the 'Touch High' initiative.

It is reported that in June 2026, roughly six months after its Hong Kong Stock Exchange debut, Zhipu's market capitalization soared into the trillion HKD club, marking a tenfold surge from its initial public offering (IPO) valuation and nearly tripling Baidu's market cap. This capital market enthusiasm stems largely from Zhipu's stellar performance. Since 2025, the company has honed in on AI Coding capabilities. By March 2026, models like GLM-4.5 and GLM-5.2 propelled the annualized recurring revenue of its MaaS platform to 1.7 billion yuan, a staggering 60-fold increase year-over-year.

In his internal memo, Tang Jie articulated a strategic departure from the industry norm of expediting scenario implementation and commercialization. Instead, Zhipu will temporarily halt short-term business expansion to channel more resources into foundational AGI capabilities. He emphasized that the enduring value of AI enterprises hinges on the upper echelons of their intelligence capabilities, rather than mere product function iterations. According to Zhipu's vision, AGI must demonstrate the capacity to generate 'relativity-level' original knowledge and approach the aggregate intelligence of humanity.

Against this backdrop, Tang Jie unveiled Zhipu's 'Touch High' initiative – a commitment to forgoing immediate application monetization in pursuit of AGI's next frontier.

Tang Jie outlined four core tenets of the 'Touch High' plan: accomplishing long-term tasks throughout the project lifecycle, constructing an agent society boasting tens of thousands of agents with diverse professional 'personalities' and 'skills,' enabling AI to self-train and continuously evolve using synthetic data, and committing hundreds of billions to researching safety governance and mechanical interpretability.

In a parallel move, Zhipu open-sourced GLM-5.2, featuring a 1 million-context window and licensed under MIT. Tang Jie asserted, 'True safety is not predicated on technological secrecy and barriers but on extensive co-construction, sharing, and oversight in the open.'

Farsight observes that Zhipu is strategically repositioning itself from an AI application service provider to a general intelligence infrastructure provider. Sustaining R&D investment post-initial stock lock-up expiration may introduce short-term pressures but also underscores the company's unwavering commitment to its technological trajectory. The pivotal question looming for Zhipu is whether these foundational capabilities can be seamlessly integrated into products to yield sustainable commercial returns.