Zhipu AI's Existential Crisis

"Everything for Existence"

"AI products launch daily, foundation models update monthly. Benchmarks and PowerPoints everywhere, invite codes and agents flying in tandem. Yet no one cares except AI content creators.

And does anyone even remember, before Hangzhou's Six Little Dragons, the previous generation of gods were the AI Six Little Tigers? 😭

In light of this, I hereby declare this week as the Funeral-Love Savage Review Week. Going through the AI Six Little Tigers one by one, keeping it real.

One post per day. Today is Zhipu AI, tomorrow is Baichuan."

Zhipu AI is going public.

Born with a "Tsinghua University" silver spoon in its mouth, armed with massive state capital, once regarded as the most credible contender for "China's OpenAI."

Today, Zhipu AI faces a severe existential crisis. Its niche has been usurped by DeepSeek.

Zhipu AI's original positioning was clear: seize the "safe, controllable, autonomous" domestic large model niche. Focus on B2B and government markets sensitive to data security. This positioning appeared rock-solid.

After all, many Chinese enterprises genuinely cannot use OpenAI. Government and state-owned enterprise concerns over data security gave Zhipu AI what seemed like an unassailable moat.

But DeepSeek's emergence fundamentally changed the game.

DeepSeek struck precisely at Zhipu AI's foundation. With stronger performance and lower costs, it offered an equally "safe, controllable, autonomous" alternative. More critically, it also supported on-premise deployment.

This was devastating for Zhipu AI. Those government and SOE clients who previously had no choice but Zhipu AI suddenly had a better option.

For instance, on June 23, Linfen People's Hospital tendered for a DeepSeek-based medical system to be deployed within the hospital, with a budget of 15.69 million yuan.

Government and SOE information system procurements across various regions have also been riding the DeepSeek name.

These cases tell the market: Zhipu AI's moat in the government sector is far less formidable than imagined.

Zhipu AI's own product capabilities have also exposed problems.

Take AutoGLM Chensi — clearly an agent product made to follow the trend. But Zhipu AI can't even get the frontend right.

AutoGLM Chensi's chain-of-thought runs excessively long, yet lacks even a basic auto-collapse function. Reviewing historical conversations requires manually scrolling upward for several seconds.

This gives the impression that Zhipu AI can't even copy properly, with zero concept of consumer-facing products.

How did no one think to add auto-collapse?

Riding the story of "China's AGI hope," Zhipu AI raised billions of RMB. This valuation rested on one premise: it was China's OpenAI, a platform-scale tech giant.

But now the problem emerges.

On the large model path, it has already been technologically surpassed by DeepSeek and others. Zhipu AI has lost its core competitiveness.

Pivot to applications? That path is also blocked.

Once it officially announces a pivot to applications like 01.AI did, it amounts to admitting defeat in the large model strategy. The valuation logic collapses from "platform-scale tech giant" to "application software company." The bubble bursts instantly.

The hottest agent, Manus, is valued at roughly $500 million. The star AI application YouWare, valued at under $200 million, is already considered an absurdly high price.

However popular the application becomes, it cannot sustain Zhipu AI's valuation of 20 billion RMB.

This is the paradox Zhipu AI faces: it can't beat others at large models, but it also can't pivot to applications. Because once it pivots, the valuation bubble immediately pops.

Funding inertia, equity structure, and high valuation dictate that Zhipu AI can only stubbornly persist down the large model path.

So Zhipu AI keeps burning cash, keeps doing pre-training. (This does reflect genuine technical idealism, to be fair.)

In September 2024, Haidian District's investment platform Beijing Zhongguancun Science City Innovation Development led a round of several billion yuan. In December 2024, Zhipu AI secured another 3 billion yuan from multiple strategic investors and state capital.

Here's a detail: In March 2025, Intelligent Emergence reported that Zhipu AI had received over 1 billion yuan in strategic financing from Hangzhou Capital. But in fact, most of this financing came from that late-2024 round — nearly 1 billion yuan in funding was counted twice.

Now, many RMB funds, especially state capital, are willing to support Zhipu AI for essentially one reason only: the IPO. Their investment logic has shifted from "investing in a good company" to "investing in an exit-able project."

Only if Zhipu AI continues raising money can it sustain itself until listing day. How the actual business performs matters less than whether it can go public.

Zhipu AI's trajectory closely mirrors the previous generation of AI Four Little Dragons. Top-tier academic credentials, grand AI narratives, and massive funding.

SenseTime's cohort already proved this out. Smart people doing AI ultimately end up doing project-based work, grinding through the hard, dirty labor.

AI security couldn't beat traditional players like Hikvision. All kinds of random projects meant dispatching a few engineers to camp out at client sites. This low-integration, labor-intensive business model cannot make a company profitable.

No matter how high-tech AI sounds, this is what actual implementation looks like.

Zhipu AI faces more severe challenges than SenseTime.

The capital market has already been "educated" once by SenseTime. Secondary markets have a more sober understanding of AI company valuations and profit models 😭

Caijing magazine reported that Zhipu AI's 2024 losses reached 2 billion yuan, with revenue of roughly 200 million — and this was in the pre-DeepSeek world.

More awkwardly, Zhipu AI's current valuation is already approaching half of SenseTime's market cap, yet its revenue is less than 1/15th of SenseTime's.

Two months ago, Zhipu AI filed with the Beijing Securities Regulatory Bureau to launch its IPO, with CICC Capital as sponsor.

I'm genuinely eager to see Zhipu AI's financials.

(Illustrations in this article generated by ChatGPT o3, with writing assistance from an AI product currently in development.)