Zhipu AI, one of China’s most prominent artificial intelligence startups, is carving out a new internal unit dedicated to product development — a structural move that reveals how the Beijing-based company is shifting from pure research toward commercial viability. The restructuring, first reported by The Information, marks an inflection point not just for Zhipu but for the broader Chinese AI industry, which is under mounting pressure to turn expensive foundation models into revenue-generating businesses.
The new unit is designed to accelerate the development of consumer-facing and enterprise products built on top of Zhipu’s large language models. It’s a bet that the company’s future depends less on building bigger models and more on figuring out what people will actually pay for.
That distinction matters enormously right now. Across both Silicon Valley and Zhongguancun, AI companies are confronting the same uncomfortable question: how do you monetize models that cost hundreds of millions of dollars to train? OpenAI has its ChatGPT subscription revenue and a growing enterprise business. Anthropic is courting large corporations. Google is embedding Gemini across its product line. But in China, the path to profitability has been murkier, complicated by fierce domestic competition, government oversight, and a consumer market that has historically been resistant to paying premium prices for software.
Zhipu, founded in 2019 as a spinoff from Tsinghua University’s Knowledge Engineering Group, has positioned itself as a technical leader among Chinese AI firms. Its GLM family of models has drawn comparisons to OpenAI’s GPT series, and the company has attracted significant backing. In 2024, Zhipu raised approximately $400 million in a funding round that valued it at roughly $3 billion, with investors including Saudi Arabia’s Prosperity7 Ventures and several Chinese state-backed funds. The company has been one of the “Six Little Tigers” — the informal grouping of China’s most promising AI startups that also includes Moonshot AI, MiniMax, Baichuan Intelligence, 01.AI, and StepFun.
But capital alone doesn’t build a business. And Zhipu’s leadership appears to recognize that the window for turning research prowess into commercial dominance is narrowing fast.
The creation of a dedicated product unit suggests Zhipu is borrowing a page from the playbook of more mature tech companies, where organizational structure directly reflects strategic priorities. When a company pulls product development out of its research organization and gives it independent leadership, it’s sending a clear signal: shipping matters as much as — or more than — publishing papers.
This is not an abstract corporate reshuffling. China’s AI sector is entering a phase where the survivors will be determined not by benchmark scores but by revenue. The country’s leading tech giants — Baidu, Alibaba, Tencent, and ByteDance — have all launched their own large language models and are aggressively integrating them into existing products with massive user bases. For startups like Zhipu, competing against incumbents who already own distribution channels requires a different kind of organizational muscle. Pure research labs don’t build go-to-market strategies. Product teams do.
The timing of Zhipu’s restructuring also coincides with a period of intense turbulence in the Chinese AI market. DeepSeek, the Hangzhou-based AI lab funded by quantitative trading firm High-Flyer, stunned the global AI community in January 2025 with its R1 reasoning model, which demonstrated performance competitive with leading Western models at a fraction of the training cost. DeepSeek’s emergence reshaped the competitive calculus for every Chinese AI startup, proving that brute-force spending on compute isn’t the only viable strategy — and raising the bar for what investors and customers expect.
Zhipu has responded by doubling down on its own model capabilities while simultaneously pushing toward applications. The company’s ChatGLM chatbot and its Zhipu Qingyan consumer product have gained traction domestically, but neither has achieved the kind of breakout adoption that would justify a multi-billion-dollar valuation on its own. The new product unit appears designed to fix that gap, bringing together engineering, design, and business development under a single organizational roof focused on shipping products rather than advancing model architectures.
There’s a broader pattern here. Across China’s AI industry, 2025 is shaping up as the year of commercialization — or reckoning. Moonshot AI, maker of the Kimi chatbot, has been aggressively pursuing consumer users and recently expanded its model’s context window capabilities. Baichuan Intelligence has pivoted toward enterprise solutions. MiniMax has focused on creative and entertainment applications. Each company is staking out a different niche, trying to find the specific application layer where it can build a defensible business before the funding environment tightens further.
And it will tighten. Chinese venture capital investment in AI, while still substantial, has become more selective. Investors who once wrote checks based on model benchmarks and team pedigree are now asking harder questions about unit economics, customer retention, and path to profitability. The era of funding AI research for its own sake is ending. What’s replacing it is a demand for products that work, scale, and generate cash.
Zhipu’s structural move also reflects lessons learned from the American AI market. OpenAI’s internal tensions between its research mission and commercial ambitions have been well documented — culminating in the dramatic boardroom crisis of late 2023 and the company’s subsequent restructuring into a for-profit entity. The lesson for Chinese AI companies watching from afar was clear: it’s better to resolve the tension between research and product early, before it becomes a governance crisis.
By creating a distinct product unit now, Zhipu is attempting to institutionalize that balance. Research continues. But it no longer drives every decision.
The Chinese government’s posture adds another dimension. Beijing has made AI development a national priority, channeling funds through state-backed investment vehicles and providing preferential access to computing resources. But the government also expects results — not just in the form of scientific papers, but in economic output, industrial applications, and technological self-sufficiency. Companies that can demonstrate real-world product traction are more likely to receive continued state support. Those that can’t may find themselves sidelined in favor of competitors who deliver tangible outcomes.
Zhipu’s close ties to Tsinghua University — one of China’s most prestigious institutions and a key node in the country’s AI research infrastructure — give it certain advantages in talent recruitment and government relationships. But those advantages are table stakes, not guarantees. Several of the Six Little Tigers have similar academic pedigrees and political connections. What separates winners from also-rans will be execution at the product level.
So what might Zhipu’s product unit actually build? The company has already shown interest in several verticals: enterprise knowledge management, code generation, multimodal AI applications combining text and image understanding, and consumer-facing assistants. A dedicated product organization could accelerate development in all of these areas by removing the bottleneck that occurs when product managers must compete with researchers for engineering resources and organizational attention.
The enterprise market in China represents a particularly large opportunity. Chinese companies across manufacturing, finance, healthcare, and logistics are actively exploring how large language models can improve operations. But enterprise adoption requires more than a good model — it demands integration work, customer support, compliance with data regulations, and the kind of sustained product iteration that research teams are rarely equipped to provide. A standalone product unit can own that entire lifecycle.
There’s risk, too. Organizational restructurings can be disruptive. Key researchers may feel marginalized if the company’s center of gravity shifts toward product development. Talent retention in China’s AI sector is already fiercely competitive, with engineers and scientists regularly poached by rivals offering higher compensation and more ambitious mandates. Zhipu will need to manage the cultural transition carefully to avoid losing the technical talent that made it valuable in the first place.
The financial pressure is real and growing. Training frontier AI models requires enormous capital expenditure on GPUs and cloud computing — costs that don’t decrease as models get larger. Zhipu’s $3 billion valuation implies that investors expect significant revenue growth in the near term. Without a product organization capable of converting model capabilities into paying customers, that valuation becomes increasingly difficult to defend in subsequent funding rounds.
What Zhipu is doing isn’t unique. But the fact that it’s doing it now — in the middle of 2025, with DeepSeek reshaping competitive dynamics and Chinese tech giants flooding the market with their own AI offerings — tells you something about the urgency the company’s leadership feels. The race in Chinese AI has moved past the question of who can build the best model. The question now is who can build the best business.
And that requires a fundamentally different kind of organization than the one Zhipu started as. The Tsinghua spinoff that began life as a research-first lab is becoming something else: a company that ships products, signs enterprise contracts, and measures success in revenue, not just perplexity scores. Whether that transformation succeeds will depend on execution — on whether the new product unit can move fast enough to claim market share before the window closes.
For the Chinese AI industry as a whole, Zhipu’s restructuring is a bellwether. If one of the country’s most research-oriented AI startups is reorganizing around product development, it signals that the entire sector has entered a new phase. The science still matters. But the business matters more.
WebProNews is an iEntry Publication