Zhipu’s World-First Large-Model Listing: CEO 张鹏 Interview

Summary

• 张鹏 said early investors could not understand Zhipu’s business model at all, with some even suggesting that its valuation be cut in half during the downturn. To outside observers, Zhipu looked more like “concrete” or a Tsinghua STEM guy: smart, capable and dependable, but light on emotional appeal and buzz; open source won it credibility among developers and programmers.
• He acknowledged that he did not become CEO only after being fully prepared; what truly wears you down is confronting unfamiliar problems after commercialization and repeatedly aligning the understanding of customers, investors and the team. The company also had to clear organizational-management hurdles at 50 people, 200 people and more than 500 people; its latest challenge is public-company compliance.
• After its shareholding reform, Zhipu leaves major decisions to the board, while day-to-day business is decided collectively by professional committees. 唐杰 leads research, the chairman handles regulators, government, ministries and financing, while 张鹏 focuses on daily operations and market commercialization. The company admits that past explorations in multimodality, video and other areas were spread too broadly; going forward, it plans to narrow its focus and serialize work where appropriate to improve resource efficiency and control cost and risk.
• 张鹏 believes all 3 predictions Zhipu made in early 2025—the continued improvement of foundation models and multimodality, the rise of agents as a major direction, and internationalization—have been validated. Its only long-term bet is AGI; in the near term, the focus is on agents, new scaling laws, new computing paradigms and reinforcement learning.
• 张鹏 summed up the goal as “making machines think like humans,” while stressing that the endgame is to empower people and generate real value. He believes the team must balance grand ideals with disciplined execution; even if forced to choose, he would choose achieving AGI over building a highly profitable company, while arguing that the two are not inherently in conflict.
• He recalled the Shenzhen client contract, the phone agent handing out red envelopes and the low-key launch of GLM-4.5, concluding that real preparation means accumulating, day after day, in the right direction—not predicting the future. He hopes Zhipu’s footnote in AGI history will read: “Zhipu was a pioneer in the history of AGI.”

Deep dive

1. Investors Couldn’t Understand the Business Model, and Zhipu Admits It Wasn’t “Cool Enough”

张鹏 recalled that early investors could not understand what Zhipu was doing at all, repeatedly asking, “What is this?”, “How does it make money?” and “How do you commercialize it?” One investor even asked during a particularly difficult economic period: “How about cutting your valuation in half?”

张小珺 once relayed a description of Zhipu: technically native and broad-minded, but somewhat boring, like concrete. It could execute beautifully, but offered little emotional appeal. 张鹏 called the assessment “fair enough,” comparing Zhipu to a smart, capable Tsinghua STEM guy who does not particularly command attention.

One thing he admires about 杨植麟 is his ability to capture ordinary people’s attention and his stronger grasp of promotion and mainstream-user needs. Zhipu has experimented in that direction, but its positioning and culture remain closer to an engineering team. Open source, meanwhile, has earned it a good reputation among developers and programmers: its grounded approach, engineering culture and product strengths all resonate with that audience.

2. No CEO Is Ever Fully Prepared; the Hardest Part Is Constantly Aligning Everyone’s Understanding

When asked about the idea that he was pushed into the CEO role without adequate preparation, 张鹏 first pushed back, then admitted the assessment was basically right. He later reached a similar conclusion in conversations with several co-founders: people are never completely prepared, no matter the moment. Once the direction is clear, a willingness to learn and improve is enough reason not to be overly afraid.

He believes Tsinghua’s biggest contribution to the team was teaching it how to learn, giving it the ability to learn, and instilling the desire to keep learning. None of that, however, amounts to a ready-made answer. What truly made him suffer was the stream of problems that emerged after commercialization—users, investors and customers all saw the company differently, forcing the team to communicate repeatedly and bring those perspectives onto the same page.

That is why 张鹏 would go to the front lines and explain the company directly to customers. Hardship itself was never the problem; what mattered was whether the work ultimately produced a return.

3. 50 People, 200 People and 500 People: Three Management Inflection Points

Academician 张钹 once warned the team that startups typically encounter several inflection points: around 50 people, 200 people, and 500 people or more. After living through them, 张鹏 came to understand that the key was not the exact headcount, but the shift in the company’s development stage.

With a few dozen people, the priorities are to make money and build team confidence—to convince everyone that the business can continue. Around 100 or 200 people, R&D, product, commercialization and daily operations begin to separate, bringing the management costs of communication, coordination and goal alignment. If each group runs its own fiefdom and no one can integrate the whole, the team can fracture.

At several hundred people, and especially beyond 500, layers and middle managers begin to appear. Information travels farther, alignment becomes harder, and management, compliance and security costs all rise. 张鹏 went from knowing almost everyone at the company to realizing that he could no longer put names to some employees—a sign that his personal line of sight no longer covered the organization.

This is not merely a psychological adjustment, but a management blind spot. The company must rely on systems, mechanisms and management infrastructure to keep events outside any individual’s line of sight within a safe and controllable range. Zhipu’s latest inflection point is the higher compliance bar that comes with being a listed company.

4. Governance Is Taking Shape, but the Academic Culture Still Has to Overcome Commercial Inertia

After completing its shareholding reform, Zhipu established a board to decide major matters, while professional committees jointly decide day-to-day business and operations. 唐杰, the chief scientist, leads research and participates in major decisions as a member of the founding team. The chairman handles regulators, government, ministries and financing. 张鹏 is primarily responsible for daily operations, especially market-facing work and commercialization.

The team is not made up solely of Tsinghua alumni; it also includes people from Fudan, SJTU, Peking University, ByteDance, Alibaba and Tencent. 张鹏 describes the overall culture as relatively open. As for the label “academic,” he considers it broadly accurate, since he did come out of academia.

The common weakness of an academic culture is a heavy emphasis on research and innovation, with commercialization receiving less attention. Because Zhipu had early market exposure and began generating revenue while still in the lab, 张鹏 believes it “will not make particularly serious mistakes in the broad direction” on this front, though it still needs continuous adjustment.

张鹏 also does not believe the company is already perfect. In the past, it pursued many parallel explorations in areas including multimodality and video generation; some projects later slowed because of resource constraints and other factors. Going forward, he believes Zhipu can narrow its exploration bandwidth and serialize certain tasks to use resources more effectively while balancing time, cost and risk.

5. Zhipu’s Only Long-Term Bet Is AGI; the Near-Term Focus Is Agents and New Computing Paradigms

张鹏 said Zhipu made 3 calls in early 2025: foundation-model capabilities would continue to improve, including multimodality and hybrid foundation models combining multiple types of data; agents would become an important direction; and internationalization would continue. Looking back, he believes all 3 have been validated.

The only long-term bet is still AGI. Over shorter time horizons, agents are critical because they connect model capabilities to real-world applications. Zhipu will also track new scaling laws, new computing paradigms and the new paradigms that reinforcement learning may unlock.

张鹏 is reluctant to judge companies simply by who says they are pursuing AGI, because people define AGI very differently. Zhipu’s slogan is “making machines think like humans,” but the endpoint is not to build a system that merely thinks. It is to have machines empower people, society and human history in return—to make society better.

He describes the team as relatively balanced between idealism and realism. It can set ambitious goals, but once the path and interim targets are clear, it executes steadily and delivers results at each stage.

6. Technological Idealism Has to Reach the Customer’s Site

张鹏 says Zhipu would not be satisfied with “making money without technical output or a contribution to the industry.” Asked to choose between achieving AGI and becoming a highly profitable company, he chose AGI—but stressed that the two are not opposites. If it truly achieves AGI, the company could also become a great business.

His version of technological idealism is not just spending every day in the lab writing code, running experiments and tinkering with machines. It also means turning technology into reality so users can actually put it to work. The team goes to customer sites to solve practical problems; the engagement does not end when the model is sold.

Customers are not buying a parameter file to put on a shelf. They want models, tools, products and services that solve real business problems. The team’s sense of accomplishment is fundamentally different when users say, “This is genuinely useful—it solved my actual problem.”

7. The Team Is in Good Shape; the CEO’s Job Is to Build Bridges

Assessing the team ahead of the listing, 张鹏 said morale was generally high. Recent model technology, model launches and commercialization results have broadly met expectations. Cloud revenue is growing rapidly, while To B revenue is also expanding steadily and quickly. As R&D and technology converge, cost investment is being optimized continuously, so he believes the path to break-even “should not be too long”; specific forecasts should be left to the financial statements.

The company holds weekly alignment meetings, with particular attention to commercialization, market changes and how the technology and research teams can work with the commercial organization, preventing research and commercialization from becoming “two separate skins.”

张鹏 sees the CEO’s role as building bridges and creating the platform for different teams to deploy their capabilities, imagination and execution. He also enjoys watching a young team occasionally deliver results beyond expectations.

8. Several Moments Worth Remembering

While serving a major client in Shenzhen, 张鹏 stayed there from July through the end of the year, eventually returning to Beijing with a client contract worth tens of millions of yuan. The experience made him proud: customers were genuinely willing to pay for the technology, and it could create value for them.

At another product launch, the team had a phone-based agent hand out red envelopes to the audience in real time. A small bug caused the amount entered for the envelopes to be off by one digit, but the money still went out. Someone later commented that this was the first red envelope AI had sent to humans—a line 张鹏 remembers clearly.

When GLM-4.5 launched, only a few dozen people were in the room. The release was relatively low-key, conducted mainly online and through open source, yet it drew strong reviews overseas. 张鹏 said U.S. companies such as Windsurf use their models; Cerebras also integrated the model to provide services for Windsurf. There are reportedly also vendors that distill and prune open-source models before repackaging them.

In the end, he believes the work itself has to be done well. However it is marketed, the verdict ultimately comes back to real-world performance. GLM-4.6 and GLM-4.7 were not accompanied by the kind of large-scale launch events used in the past, partly because the industry had grown somewhat fatigued with promotional spectacle. 张鹏 sees that as one of the ways DeepSeek has influenced the sector.

9. “Preparation” Means Long-Term Accumulation, Not Precise Forecasting

张鹏 remembers his master’s adviser saying: “Opportunities always go to those who are prepared.” He later added that when a plank drifts by at sea, you still have to start paddling before you can grab it.

Preparation does not mean accurately predicting what will happen tomorrow; the future is too difficult to forecast precisely. Real preparation means doing what you believe is right, day after day and year after year, accumulating steadily without being distracted by noise. When an opportunity appears, the team is then capable of seizing it.

He believes Zhipu’s journey reflects both luck and long-term accumulation. Timing, circumstances and friends provided help at critical moments, while the team’s continuous preparation ensured that it did not miss its opportunities. In Chinese terms, timing, conditions and people all had to align.

10. Hoping to Become a Pioneer in the History of AGI

When 张鹏 attended Moore Threads’ bell-ringing ceremony, he joked that he had gone to learn “how to ring the bell.” He reflected that reaching this point had been difficult for everyone, and that companies that make it this far are, in a sense, all heroes.

If Zhipu appears in an artificial-intelligence history book 100 years from now, he hopes its footnote will read: “Zhipu was a pioneer in the history of AGI.” When 张小珺 pressed him on why it should not be “innovator,” 张鹏 replied that pioneers are generally innovators—the people who open the road.