Skip to main content

Zhipu's Big Bet: Ignoring the Bell, They Hit Reset on AI Strategy

In the fast-paced world of AI, most companies are racing to ring the bell—launching products, chasing revenue, and trying to prove their worth. But Zhipu, one of China's leading large model firms, has decided to hit reset instead.

On July 11, founder Tang Jie sent out an internal letter that turned heads. The company is launching what it calls the "Touch High Plan"—a strategic investment in four core areas over the next two years: long-range tasks, autonomous intelligent agent systems, fully self-trained models, and extreme security governance. The goal? Nothing less than the next generation of general artificial intelligence (AGI).

A New Focus on Long-Range Tasks

Zhipu's latest open-source model, GLM-5.2, already supports a massive 1 million token context length—think of it as the AI equivalent of having a photographic memory for long conversations or documents. But the company isn't stopping there. They're pushing models beyond simple Q&A toward handling grand, multi-step projects. Imagine an AI that can plan and execute a complex research project or manage a supply chain over weeks, not just answer a single question.

Building a Workforce of AI Agents

Another key piece of the plan is creating thousands of specialized intelligent agents. These aren't just chatbots; they're digital workers with specific skills, designed to work together. Zhipu envisions a future where these agents form a "fully automated company"—a concept they call NPC (Non-Player Character, borrowing from gaming). It's a bold vision: a business run almost entirely by AI, from customer service to logistics.

Self-Training and Synthetic Data

To fuel this, Zhipu is building a high-quality synthetic data factory. Instead of relying solely on human-labeled data, they'll use AI to generate its own training material through a process called "self-play." Think of it as AI sparring with itself to get better, much like how AlphaGo trained by playing millions of games against itself. This approach could help Zhipu scale its models faster and more efficiently.

Safety First, Even at a Cost

Perhaps the most striking part of the plan is the emphasis on safety. Zhipu says it will invest billions of yuan into "mechanical explainability"—making AI's black-box decisions transparent. They're also pushing forward with super alignment research, ensuring that as AI gets smarter, it stays aligned with human values. In an industry where speed often trumps caution, this is a refreshing—and risky—bet.

The Stock Rollercoaster

This strategic pivot comes after a wild ride for Zhipu's stock. Listed on the Hong Kong Stock Exchange on January 8, 2026, the company saw its shares soar to HKD 2,410 by June 22, briefly making it the first Chinese large model firm to exceed a trillion Hong Kong dollars in market value. But then the market cooled. By July 10, the stock had dropped to HKD 1,640, trimming the market cap to HKD 731.2 billion.

While many in the industry are rushing to commercialize and show profits, Zhipu is choosing to "break through upwards"—investing heavily in R&D even as the market grows skeptical. It's a high-stakes gamble that reflects the tension between technical ambition and financial reality.

What This Means for the AI Landscape

Zhipu's move is a reminder that the path to AGI isn't a sprint—it's a marathon. While others are focused on incremental improvements and monetization, Zhipu is betting that long-term investment in foundational capabilities will pay off. Whether they succeed or not, their strategy is a fascinating case study in how AI companies navigate the delicate balance between innovation, safety, and market pressure.

Key Points

  • Zhipu announced the "Touch High Plan," a two-year strategy focusing on long-range tasks, autonomous agents, self-training, and safety.
  • The company released GLM-5.2, an open-source model with 1 million token context length.
  • Zhipu aims to build a "fully automated company" using thousands of specialized AI agents.
  • The plan includes a synthetic data factory and heavy investment in AI explainability and alignment.
  • Zhipu's stock experienced a sharp rise and fall, with market cap dropping from over HKD 1 trillion to HKD 731.2 billion.