Zhipu's Clockwork

Yesterday, the stock price of Zhipu rose by 48% during intraday trading, reaching HK$1,620 directly.

A friend in the group said that there is a pattern with Zhipu's stock: every time a new model is released, the stock price goes up. I thought to myself at that time, could it be that magical? So, after looking at the historical data, I found that it was indeed the case.

It went public in January at an issue price of HK$116.2, and barely managed to avoid breaking the issue price at the closing.

Then, take a look. The stock price trend in the past five months almost perfectly matches the rhythm of Zhipu's model releases.

In February, the flagship model GLM - 5 was launched, and on the same day, it was announced that the API price would increase by 30%. The stock price rose by 42% that day, and its market value exceeded that of Kuaishou and JD.com in one day.

Not long after, GLM - 5 - Turbo was released, and the API price increased by another 20%. It wasn't a major version upgrade, just a tool - type model for specific scenarios. The rhythm continued.

In March, the annual report was released. The ARR of the MaaS platform was 1.7 billion, a 60 - fold increase year - on - year. After the API price increased by 83%, the call volume not only didn't decrease, but actually increased by 4 times during the same period. The next day, the stock price rose by 32%, and the market value exceeded HK$400 billion for the first time.

Immediately afterwards, GLM - 5.1 was released. Its selling point was the "ability to work continuously for 8 hours", the first among global open - source models to achieve this. The stock price rose by 18% during intraday trading on that day.

In May, the large - model sector had a collective upsurge, and Zhipu's stock price rose by 37%. Its market value exceeded HK$500 billion. At the end of the month, the stock price once soared to nearly HK$2,000, only HK$7 away from the round - number mark. Since its listing, the increase has exceeded 1,600%.

Then, something interesting happened.

In the afternoon of that day, the stock price dropped from HK$2,000 to HK$1,500 within half an hour. The amplitude exceeded 30%.

In June, GLM - 5.2 was fully opened. On the same day, the US Department of Commerce took a tough measure, requiring Anthropic to restrict non - Americans from accessing. Zhipu, on the other hand, chose to open - source under the MIT license without any geographical restrictions.

Driven by both technical and geopolitical factors, the stock price rose by 48% during intraday trading yesterday.

In five months, the stock price has gone from over HK$100 to HK$1,600. There is a story behind each sharp increase: either a new model was released, or the price was adjusted, or both.

It can almost be said that Zhipu's rhythm is like a well - wound clock, with a release almost every month, and each release can push up the stock price. I counted, and there were 6 obvious sharp increases in 122 days.

Five of them were due to Zhipu's own moves: releasing models, adjusting prices, and releasing financial reports.

The increase in May was due to the overall movement of the sector, driven by the financing news of Yuezhianmian and Jieyuexingchen. If we only consider Zhipu's own rhythm, it's about once a month. The shortest interval was only 8 days.

By the way, the models themselves are indeed improving.

The programming ability of GLM - 5 can already compete with the world's top closed - source models. GLM - 5.1 can handle long - term tasks for 8 hours, and GLM - 5.2 can handle a context of up to 1 million tokens.

The increasing call volume after the price increase shows that the market really recognizes these products.

But there is a problem. I've been thinking that MiniMax has also been releasing models this year. It released M2 in March and M3 in May. Its performance is not bad, and it can also handle a 1M context.

The two companies have similar technical routes, similar revenue scales, and similar levels of losses. They went public only a few days apart. By the end of May, Zhipu's market value was over HK$700 billion, while MiniMax's was HK$263.4 billion, a difference of 2.7 times. How do you explain this?

......

To answer this question, we need to look at a figure first. The proportion of Zhipu's truly tradable shares in the total share capital is less than 4%.

I looked through the prospectus. The total share capital is about 446 million shares. When it went public in January, 37.41 million H - shares were issued, which increased to 43.04 million after the over - allotment option was exercised. Among them, cornerstone investors subscribed for 25.68 million shares, with a 6 - month lock - up period.

Calculated, the number of shares that can actually be traded in the market after listing is about 17.35 million.

What does 17.35 million shares mean? Out of the 446 million total share capital, only this many are in circulation. The remaining 96% are all locked up and no one can trade them.

Imagine throwing the same - sized stones into a swimming pool and a bathtub. Will the splashes be the same? Zhipu's tradable shares are like the bathtub. Any buy or sell order will cause much bigger waves in this pool than in a normal situation.

I checked. MiniMax's free - float ratio is between 4% and 5.5%, slightly larger than Zhipu's. However, the more crucial difference lies elsewhere.

On June 8th, Zhipu was included in the Hang Seng Tech Index with a weight of 0.53%. 0.53% may seem insignificant, but behind the Hang Seng Tech Index, there is a huge pool of funds.

I checked. There are 13 ETFs tracking the Hang Seng Tech Index. In the first two months of this year, the cumulative net subscription of these ETFs was 41.6 billion. Including related products such as the Hong Kong Stock Connect Internet and the Hang Seng Internet, the total inflow this year is about 70 billion.

I saw in the 21st Century Business Herald that the figure is even larger, saying that the cumulative net subscription of the Hang Seng Tech ETFs in half a year exceeded 100 billion.

Zhipu accounts for 0.53% of the weight. A portion of these funds must be allocated according to this weight. It's a "must - buy". The task of index funds is to replicate the index, not to judge whether a company is overvalued or not.

These funds are price - insensitive. Whether Zhipu has a negative P/E ratio or a negative P/B ratio doesn't affect their buying decisions. If the index says to buy, they will buy.

It's not just index funds.

In Zhipu's shareholder structure, state - owned assets account for a significant proportion. I won't mention the specific names as they can all be found. These early investments mean that their holding period is measured in years and they are insensitive to short - term price fluctuations.

So, this is the situation you see:

4% of the chips are facing a group of buyers who don't care about the price. Index funds are driven by regulations, and state - owned assets are for long - term holding. What's lacking on this trading board is buyers who will bargain.

Normal price discovery requires a game between buyers and sellers. One side thinks the price is too high and wants to sell, while the other side thinks it's cheap and wants to buy. The price finds a balance in the middle.

On Zhipu's trading board, the sellers are locked up, with 96% unable to trade; and the buyers don't bargain. Under this structure, the price discovery mechanism partially fails.

This is not a problem unique to Zhipu. Similar things have happened several times in the history of the Hang Seng Tech Index.

I looked through the records: In 2021, 5 stocks entered the Hang Seng Tech Index through the "fast - track inclusion mechanism": JD Health, Kuaishou, Baidu, Bilibili, and Autohome.

After inclusion, all 5 stocks declined, with declines of 42%, 81%, 37%, 37%, and 47% respectively. None was spared.

In 2022, there was an even more absurd case. When SenseTime was included in the Hang Seng Tech Index, the Hang Seng Index Company mistakenly set its free - float factor from the actual 2% to 35%. Passive funds bought crazily according to the 35% weight.

Later, when the mistake was corrected, SenseTime's stock price tumbled in a single day.

These two historical events show one thing. Being included in the index is a structural event. The forces behind it are mechanical and institutional, and have no direct relation to whether the company is good or not.

This force pushes the stock price up when it rises and pulls it down when it corrects.

Zhipu was included in the Hang Seng Tech Index on June 8th, and MiniMax was included on the same day. However, due to its dual - class share structure, MiniMax can enter the Hong Kong Stock Connect at the earliest in August, two months later than Zhipu.

They are in the same industry with similar model capabilities. Even the time windows for passive funds to enter the market differ by two months.

By now, the 2.7 - fold difference in market value is not entirely a mystery. Zhipu has a smaller tradable share, earlier entry of passive funds, and more concentrated price - insensitive buyers. The difference in amplitude comes from the structure.

......

The combination of these two factors can explain the stock price amplitude, but there is something more crucial than the amplitude.

What is it?

My expression may be a bit sharp. For Zhipu, a high stock price is like oxygen. Without it, the financing chain won't work, and the rhythm of model iteration can't be maintained.

Just take a look at the financial reports.

In the 2025 annual report, the revenue was 724 million, and the net loss was 4.718 billion. Part of the net loss is non - cash items and can't be directly used to calculate the cash - burning rate. I roughly calculated the operating cash outflow, and it burns about 3 billion a year, or two to three hundred million a month.

I also looked at the balance sheet. At the end of 2025, there was 2.26 billion in cash on the books, plus 490 million in wealth management products, making a total of 2.75 billion in liquid reserves.

It went public in Hong Kong in January, and received about 3.6 billion RMB. 2.75 billion plus 3.6 billion is a total of 6.3 to 7 billion. Sounds like a lot, right? But with an annual cash - burning rate of 3 billion and an accelerating rate, the money will run out in less than two and a half years.

This is assuming that the cash - burning rate doesn't increase further. In fact, the training and inference costs of the GLM - 5 series will only be higher. When GLM - 5 was launched in February, Zhipu publicly apologized for the lack of computing power.

On June 1st, Zhipu announced its plan to list on the STAR Market, raising 15 billion. 12 billion will be invested in large - model R & D, 2 billion in the MaaS platform, and 1 billion to supplement working capital.

What does 15 billion mean? It's more than four times the amount raised in the Hong Kong stock market.

The prospectus says it's for "R & D" and "platform construction". Look at this from another perspective. The money on the books is enough to last until the second half of 2027. If the A - share fundraising is successful, it can extend its life for more than five years. If not, it can only rely on bank credit.

So, there is a question worth thinking about.

Everyone is discussing whether Zhipu's stock price is "worth" the current level, as if a high stock price is a reward for good model performance. I think it's the other way around. For Zhipu, a high stock price is oxygen.

Because to maintain the rhythm of model iteration, continuous huge R & D investment is needed. Where does the money come from? Financing. What's the prerequisite for financing? The valuation needs to be supported. What supports the valuation? The stock price. What supports the stock price? A catalyst every 20 days to create enough amplitude in the 4% tradable shares.

You see, it's a closed - loop.

The model drives the stock price, the stock price supports the valuation, the valuation supports the financing, the financing supports the R & D, and the R & D produces the next model. If this loop breaks at any link, the whole chain will reverse.

If the model iteration slows down, the catalyst will disappear. Without the catalyst, the amplitude in the 4% tradable shares will amplify the decline. Once the stock price drops, the valuation can't be supported, and the conditions for the next round of financing will deteriorate. Without enough funds, the model iteration will only slow down further.

So, I think the difference in market value between MiniMax and Zhipu is not just about the tradable shares and passive funds. Zhipu needs this high stock price.

The revenue ratio of the two companies is 1.27 times, while the market - value ratio is 2.7 times. If the difference is purely due to the structure, that's okay. But the structure itself doesn't come out of thin air.

With only 4% of tradable shares, a monthly catalytic rhythm, and starting A - share financing before the lock - up period ends, when you put these things together, it's hard not to think that there is a precisely - operating machine behind it.

Investors believe that the models will continue to improve, the tokens will continue to be sold, and the next round of financing will be successful. As long as the confidence remains, this loop will keep running.

......

I checked. On July 8th, Zhipu's first batch of restricted shares will be unlocked.

This batch of shares belongs to the cornerstone investors, with a cost of HK$116.2. Based on the recent price, the floating profit is more than 10 times. Whether they will sell or not, I can't say. With a 10 - fold floating profit in front of you, you'd also think about it.

After the unlocking, the number of tradable shares in the market will increase from 17.35 million to about 43 million. The proportion of the tradable shares will increase from 4% to nearly 10%.

But this is not the biggest change.

The next day, on July 9th, MiniMax's restricted shares will be unlocked. Calculated based on the total share capital, this time it's not 5%, but nearly 50%. 44% of old shareholders and 5% of cornerstone investors will unlock their shares, with a total unlocked market value of about HK$65.8 billion. Zhipu's unlocked shares account for 5.76%, while MiniMax's is nearly 50%. The magnitudes are completely different.

This is not just about these two companies. According to Phoenix Network data, the total unlocking scale of Hong Kong stocks in 2026 is expected to be between HK$1.55 trillion and HK$1.7 trillion, nearly three times higher than last year. AI companies are the main force in this round of unlocking wave.

From July 7th to 9th, within three days, there will be a concentrated release. The entire AI sector will face this challenge at the same time.

In the past five months, Zhipu's stock price has been in a rather special state. There are few tradable shares, there is a stimulating news