Key Takeaway: Zhipu AI's Tang Jie Affirms 'Counter-Intuitive' Path, Betting on AGI Research Over Short-Term Monetization

As an investor, I've seen too many entrepreneurs wavering between "technical idealism" and "commercial reality." Tang Jie's "Touch High Plan" reminds me of the days when DeepMind stuck with AlphaGo and OpenAI persisted with the GPT route. But history doesn't simply repeat itself. This time, Zhipu AI chose a more extreme path: at a time when the industry generally pursues implementation, revenue, and profit, they instead increased R&D investment, refusing to trade model capabilities for cash.

Comparing Two Routes: Zhipu's "Touch High" vs. Industry's "Gold Rush"

Dimension	Zhipu Route (Touch High)	Mainstream Industry Route (Gold Rush)
Core Goal	Explore AGI boundaries, break through model capability ceilings	Quickly acquire users, revenue, market share
Commercial Action	Cautiously open APIs, control pace of commercial partnerships	Low-price or free API access, actively output industry solutions
Valuation Logic	Monopoly pricing power brought by potential technical breakthroughs	User scale, revenue growth rate, GMV
Competitive Moat	Top talent, computing power, data, original algorithms	Channels, scenarios, ecosystems, brand
Risk Points	Uncertain technical breakthroughs, high capital consumption, team fatigue possible	Homogeneous competition, thin profits, shallow technical moats

Business Model Assessment: Valuation Logics Are Completely Different

Zhipu's "Touch High" is essentially a bet on bargaining power brought by a "technological singularity." If AGI is truly achieved in the future, the first company to reach AGI will possess nearly infinite commercial space—just like NVIDIA with GPUs today. But this valuation model requires extremely harsh prerequisites: First, the technical path must be correct; Second, funding must be sufficient to burn until the breakthrough point; Third, the team must maintain execution amidst long-term uncertainty.

The mainstream industry route (such as ByteDance's Doubao, Baidu's Wenxin, SenseTime's RiRiXin) is closer to "tool-based monetization": Outputting models as API services or solutions, earning fees per call or project payment. The advantage of this model is controllable risk, but the disadvantage is a clear ceiling—once model capabilities become homogeneous, price wars are inevitable, and profit margins get compressed. I've invested in several AI companies, and during their early commercialization stages, they all fell into the dilemma of "selling models being worse than selling courses."

Competitive Moat Analysis: Is Zhipu's "Counter-Intuitive" Approach Sustainable?

Tang Jie emphasizes "counter-intuitiveness," which is exactly the label investors need to be most wary of. Counter-intuitive means fighting against mainstream market cognition. Historically, only a very few companies succeeded (like SpaceX, Huawei HiSilicon). But failure cases are more numerous: Webvan, Juicero. Zhipu's moat lies in the academic backgrounds of founders like Cao Xiang and Tang Jie, and Tsinghua University's accumulation in NLP. But "team execution is key"—if the "Touch High" plan leads to vague R&D goals, such as "exploring AGI" turning into "trying every direction," funds will dissipate rapidly.

Additionally, Zhipu's current main revenue source is B-side model licensing and technical services. Completely abandoning short-term monetization means cash flow might drop off a cliff. As I understand it, Zhipu's subsequent financing rhythm might be affected because investors value certainty more. But Tang Jie's open letter might be sending a signal to the capital market: "We adhere to long-termism; please value us based on future expectations."

Investment Judgment: Opportunities and Risks Coexist

If you ask me whether I'd invest in Zhipu, my answer is: Yes, but critical milestones must be set. "Touching high" is not "touching in the dark." Zhipu needs to prove to investors: First, capital efficiency—for example, can annual improvements in model capabilities be quantified (via evaluation metrics, parameter count, inference cost reduction); Second, phased results—for example, reaching OpenAI levels on specific tasks or achieving a certain technical breakthrough; Third, talent retention—are core scientists willing to stay in a long-term research environment?

By comparison, investing in the "gold rush" route might yield returns earlier, but the ceiling will also appear sooner. Several large model startups I've recently contacted are desperately lowering API prices, with gross margins already below 10%. This kind of competition is a typical "red ocean," not my style.

Action Advice for Readers

If you are an investor or entrepreneur focused on the AI field, think like this:

1. Don't easily dismiss any path. Both Zhipu's "Touch High" and the industry's "Gold Rush" have their rationality. It depends on your capital cycle and risk appetite. If you have patience for 10+ years, keep an eye on Zhipu; if you hope to see returns within 3 years, look more at application-layer projects.

2. Pay attention to the specific execution details of Zhipu's "Touch High Plan." Tang Jie said "continue focusing on AGI research," but what is the definition of AGI? Is it human-level multimodal understanding, or general reasoning? If there is no clear technical... internally

---

Original Link: https://www.ithome.com/0/975/620.htm