Zhipu AI's Tang Jie Issues Internal Memo: Betting on AGI, Abandoning Short-Term Monetization
Taylor Wilson
Zhipu (智谱) founder Tang Jie published an internal letter launching the "Touch High" plan — a full commitment to AGI research over short-term profit. With the stock up 24× from its IPO price and a fresh HK$31.4 billion placement just completed, the letter is effectively a message to investors: the money goes into long-term research, not near-term earnings.
What exactly is the "Touch High" plan?
Tang listed four peaks on the road to AGI: long-horizon tasks, autonomous agent systems, fully self-directed training, and ultimate safety governance.
In plain terms = the first three mean evolving AI from "answering a single question" to "completing an entire job on its own and teaching itself to improve." The fourth ensures the process stays under control.
Safety governance was singled out for special emphasis. Zhipu plans to commit tens of billions of yuan to "mechanistic interpretability" — making model decisions transparent and human-readable instead of black-box.
This means → Zhipu is treating safety as part of the technical roadmap itself, not just a compliance cost — an uncommon stance among Chinese AI companies.
Why release this letter now?
Zhipu listed on the HKEX in January at HK$116.2 per share. The stock peaked at HK$2,980 — more than 24× the IPO price — briefly pushing market cap past HK$1.3 trillion.
On July 8, over 25 million shares held by 11 cornerstone investors were unlocked. The expected sell-off never materialised; the stock rose instead.
The next day Zhipu placed new shares at HK$1,588 each, a ~13% discount, raising roughly HK$31.4 billion — the largest single AI-company placement on the HKEX this year.
This means → with a massive cash injection just landed, Tang immediately went public saying "the money goes to AGI research." The letter's real audience is investors: do not expect short-term profits; prepare for a long horizon.
Where does the technology stand?
Zhipu's GLM-5.2 model is considered in comprehensive benchmarks to have reached the performance frontier of leading overseas models, and its open-source approach has earned recognition in the developer community.
The company said placement proceeds will go mainly to foundation-model R&D, compute infrastructure, commercialisation, and global ecosystem development.
In plain terms = the technology has caught up to the first tier's threshold. The new capital is meant to push through — but the gap between "reaching the threshold" and "holding a first-tier position" is exactly what the market needs to watch.
What should investors take away?
Tang wrote explicitly that the company's view on an "upper-bound leap in intelligence" is "the very perception we most want to convey to investors" — tying strategic direction directly to capital-market communication.
This reflects a clear signal: Zhipu's management has deliberately chosen a "high valuation + long payback" narrative over a "monetise fast to prove the business model" path.
Two nodes will determine how the market assesses Zhipu going forward: whether "Touch High" creates differentiation in the AGI race, and the real progress of the multi-billion safety-governance investment — the first decides technical standing, the second decides regulatory and public trust.
Content is for reference only, not financial advice.