In 2019, inside Tsinghua University’s Knowledge Engineering Lab, a small team of researchers quietly started building something ambitious.
That team became Zhipu AI (now branding itself Z.ai).
Their mission was simple in spirit but huge in scale: design large language models (LLMs) that can think, talk, write, reason and help people build AI tools, a kind of digital “cognitive engine.”
In the early days, they worked quietly, building the GLM series of models.
‘GLM-4, GLM-4.5’ became familiar names in AI circles, almost like secret codes among tech nerds.
But then Zhipu made a bold move: open a developer platform (bigmodel.cn), let outside engineers tap into their models via APIs, and enable more people to build on top of them.
Gradually, they expanded into multimodal realms (not only text but images, agents, video): “Ying,” “AutoGLM,” and more.
At home in China, Zhipu came to be regarded as one of the “AI Tigers,” the next generation of Chinese AI powerhouses.
According to data firm IDC, Zhipu has become one of the top three LLM players in China.
IPO ambitions
Then the story entered a new chapter: IPO.
In April 2025, Zhipu officially filed for IPO counseling with the Beijing Securities Regulatory Bureau, hiring China International Capital Corporation (CICC) as the guiding institution.
This was a milestone: Zhipu became the first among China’s LLM “six tigers” to openly start the journey toward going public.
They plan to finish IPO-counseling by October, and aim to submit a listing application as soon as possible.
Going public offers investors exposure to China’s fastest-growing AI infrastructure play, one backed by local governments, private capital, and national strategy.
As this was happening, money flowed in.
Local governments from Chengdu, Hangzhou, Zhuhai poured strategic investments.
In March, Zhipu raised over 1 billion yuan led by state-backed funds.
These injections did more than boost cash, they anchored Zhipu to China’s broader mission: building sovereign AI infrastructure, especially in emerging markets eager for alternatives to Western tech.
Rivals, roadblocks and reinvention
But of course, in every captivating story there is tension. Enter the rivals.
A newcomer, DeepSeek, surprised many by launching DeepSeek-R1 early in 2025.
Its models claimed to match leading AI systems, at a much lower cost – challenging the high-investment, high-scale model of companies like Zhipu.
Suddenly, questions multiplied: can Zhipu keep up? Will investors flock to the cheaper challenger?
Things inside Zhipu weren’t entirely calm either.
Over recent months, several top executives resigned – the strategy head, the finance VP, others.
For a company stepping toward IPO, these shakeups raised eyebrows. Observers wondered if these were signs of internal friction or just the growing pains of hypergrowth.
Meanwhile, the world was watching.
OpenAI (yes, the OpenAI) flagged Zhipu publicly. In a blog post, it said Beijing-backed Zhipu had made “notable progress” and was on the “front line” of China’s ambitions in AI.
US regulators also intervened – Zhipu and its subsidiaries were added to the US Commerce Department’s Entity List, restricting its access to certain American technologies.
Risk and moat
So where does Zhipu stand now, and can it truly challenge US models like GPT-4?
Well, on one hand, Zhipu claims that some versions of GLM surpass GPT-4 on benchmarks, and that certain GLM models match rival products “at one-thirtieth of the operational cost.”
Meanwhile, by open-sourcing some model versions (32B, 9B), Zhipu seeks to attract research and developer loyalty and increase transparency.
On the other hand, scaling and profitability remain huge challenges.
AI models eat capital. In 2024, Zhipu reportedly generated revenue of about 3 billion yuan, but incurred net losses around 2 billion yuan, as high R&D costs and competitive pricing squeezed margins.
Also, the price war is real: Zhipu’s GLM-4-Flash token fees plunged (by some accounts, down 99.99 %) as players raced to offer “cheap AI.”
For investors, those numbers might look daunting, but they also reveal the moat.
What this means for investors
Zhipu’s burn rate reflects deep technical investment rather than marketing fluff.
It’s doing the hard yards: optimising compute, building proprietary data pipelines, and expanding its ecosystem overseas.
In markets like Indonesia, Malaysia, and the UAE, where US cloud services face regulatory barriers, Zhipu could perhaps become the go-to AI infrastructure partner.
