On July 11, Zhipu founder Tang Jie sent an all-hands letter, "The Great Wave Is Here," with one line that became tech-media headline fodder: "Not reaching the summit is failure."
Even more jarring was the timing. Three days before the letter, Zhipu had just placed shares to raise about HK$31.4 billion-the largest single AI-company deal on the Hong Kong exchange this year-with market cap swinging wildly between HK$731.4 billion and HK$906.3 billion. Having raised that much, by all logic they should rush to tell a commercialization story. But Tang Jie said the opposite: over the next two years, no chasing short-term monetization, strategic investment in AGI. The plan is called "Touch High"-in Chinese, "reaching up."
What Is "Touch High" Reaching For
Tang Jie drew the "three mountains" toward AGI: long-horizon tasks (letting the model plan and execute over months, not just answer in real time), fully autonomous agent systems (from "one-person company" to "fully automated company NPCs," thousands of digital employees collaborating 24/7), and self-evolution (AI training AI, using synthetic data and self-play to break through the human-data bottleneck). Past these three mountains, AI will learn what "I" means, and beyond that lie emotion, consciousness, straight toward superintelligence ASI. He cited DeepMind's "From AGI to ASI" reasoning: if AGI instances grow tenfold per year, in five years there'd be 100 million of them, and at the population level that equals ASI.
Where the Billions Go
Investment concentrates on four engines: the first three mountains plus extreme safety governance. Tang Jie stressed the latter most-investing billions-level resources to crack "mechanistic interpretability," opening the AI black box to figure out which neuron is firing in a decision. Reaching up with one hand while paving the road down with the other: Zhipu simultaneously open-sourced GLM-5.2 (million-token context, MIT license)-in his words, "one hand reaching up, the other paving down."
Why It's Unusual
"Others ring the bell, we reset to zero." Zhipu listed this January; the stock rose as much as 24×, then raised another HK$31.4 billion-ammo's full, they can afford the hardest path. The judgment is also clear: this AI shift isn't product innovation, it's technology raising the "ceiling of intelligence"; whoever pushes that ceiling up an inch redefines every industry. Building tools only traps you in price wars; Zhipu chose to ascend, directly contesting next-gen tech-defining power. One side proof: Tencent News reported that after its Coding API raised prices 83%, usage instead grew-the market will pay a premium for tech leaps.
Risks and Take
"Two years no monetization" sounds bold, but can investors wait? The wild market-cap swings are the market's hesitation. "AI training AI" also carries academia's "model collapse" worry-repeatedly using synthetic data is like a copier copying a copy, increasingly distorted. Further out is ethics: driverless companies are coming, what happens to employment structure?
Whatever the outcome, this is the first time a Chinese AI company has walked from chaser into no-man's-land to define the frontier. For regular people, it clearly sketches a trend happening now: regardless of when driverless companies arrive, an "AI colleague" may join your work group as soon as next year. The great wave is here-be the surfer or get washed ashore, your call.
References
- Zhipu launches "Touch High" plan, Tang Jie says "not reaching the summit is failure" (NetEase): https://www.163.com/dy/article/L1NCHE7K0511KIFE.html
- Zhipu founder Tang Jie internal letter: true summiters turn the mountain into a road (East Money/Leidi): https://caifuhao.eastmoney.com/news/20260711210618298945890
- Zhipu completes HK$31.4B placement, Tang Jie: we will push upward (Sohu): https://m.sohu.com/a/1049836546_430392
- Amid the unlock storm, Zhipu hits its "great wave" moment (Tencent News): https://news.qq.com/rain/a/20260713A09WV900