Shares of the domestic GPU champion plunge 20% as first batch of restricted shares hits market, exposing tensions between valuation and fundamentals.
By Su Yang
A wave of expiring share lock-ups is emerging as a major test for Moore Threads (688795.SH), one of China’s leading domestic GPU makers.
The company’s shares plunged 20% on Sept. 7 to close at 415.49 yuan, pushing its market capitalization down to 195.3 billion yuan ($29.1 billion), the lowest since listing in December 2025. The immediate trigger was the release of 25.77 million shares issued in its IPO and public offering to institutional investors, equivalent to 5.48% of the total shares outstanding.
The number of freely tradable shares rose to about 56 million, an increase of 85%.
From market darling to reality check
Moore Threads became the first major domestic GPU maker to list on Shanghai’s tech-heavy STAR Market. Its market capitalization briefly topped 440 billion yuan, while investors who won IPO allocations made a paper profit of more than 280,000 yuan on a single allocation. The stock quickly earned the nickname “Mo Wang,” or “GPU king.”
The enthusiasm reflected both confidence in China’s computing-power ambitions and investors’ willingness to price in a future that had yet to arrive.
The company itself tried to cool the frenzy soon after its listing, warning that its stock had risen sharply in a short period and highlighting risks including continued losses and the lack of meaningful revenue from its new products.
Now, faced with the market impact of its first major share unlock, Moore Threads has again acknowledged the issue.
“The company has noted today’s stock-price volatility. The relevant lock-up expiry procedures were disclosed in accordance with regulations, and the company’s production and operations continue to improve,” it said.
Moore Threads is not alone. MiniMax (0100.HK), one of China’s two leading AI model developers, also suffered a sharp decline on the first trading day after its first lock-up expiry. Other Chinese hard-tech companies that listed after Moore Threads, including Muxi (MetaX) (688802.SH), Biren Technology (6082.HK) and Tianshu Zhixin (9903.HK), will face similar tests.
But the impact is not always negative.
On July 8, shares representing 5.76% of Z.AI (Zhipu AI) (2513.HK) equity became tradable. Its stock rose almost 20% intraday and closed at 13.35 yuan. The market attributed the gain to long-term commitments from cornerstone investors and the relatively small proportion of shares becoming freely tradable after the expiry.
Why some investors are selling
Optimists argue that a sell-off on the day of a lock-up expiry is a normal consequence of a sudden increase in the number of shares available to trade. It does not necessarily signal a deterioration in a company’s fundamentals.
“Institutions will sell their placement shares if they can. At current prices, they are still making a lot of money,” a private-equity investor told Tencent Technology.
The first batch of shares to unlock generally has the shortest holding period, the investor said, meaning institutional placements are designed in part to generate relatively quick returns. Selling therefore does not necessarily reflect a judgment on the company’s operating performance.
Moore Threads’ financial results have in fact improved substantially. In the first half of 2026, revenue jumped 147.4% year on year to 1.736 billion yuan, already exceeding its full-year 2025 revenue of 1.506 billion yuan. Its net loss narrowed from 271 million yuan a year earlier to just 12 million yuan, a 95.7% reduction, bringing the company close to breakeven.
But rising costs, particularly for high-bandwidth memory, or HBM, and the economics of delivering computing clusters have squeezed margins. Gross margin fell from 67.35% in the first quarter to 49.26% in the second. First-half gross margin was 56.95%, down 12.19 percentage points from a year earlier.
A bigger test to come
The much larger challenge is expected in December.
On Dec. 7, about 186 million Moore Threads shares are scheduled to become tradable, representing 39.55% of the company’s total equity. That is roughly seven times the size of this month’s batch of shares becoming tradable and, more importantly, involves original shareholders from the company’s pre-IPO financing rounds.
“By December, the first-tier investors will be fully unlocked, and these are major early-stage institutional players,” a veteran semiconductor industry executive said.
The more fundamental concern, however, is whether domestic GPUs can justify their prices.
A private-equity investor who has followed China’s semiconductor industry for years argued that the recent sell-off was reasonable because the performance of domestic chips does not yet match their price. In other words, domestic GPUs may simply be too expensive.
Fabless chip companies are pricing products based on expected orders in the next one or two years, the investor said. But HBM prices have risen sharply, creating a difficult equation.
“A domestic card costs around 100,000 yuan this year. If it costs 200,000 to 300,000 yuan next year while delivering only H200-level performance, the investment case is extremely poor,” the investor said, referring to Nvidia’s older-generation high-end AI accelerator.
That could eventually limit demand from major technology companies and AI labs. Government-backed intelligent computing centers, however, could provide a floor for demand for domestic computing products.
Cost of computing matters
Additional demand from China’s intelligent computing infrastructure may be coming. China’s Ministry of Industry and Information Technology on Sept. 7 issued its 15th Five-Year Plan for the information and communications sector, which includes a target of 9,800 EFLOPS of intelligent computing capacity by 2030 and large-scale adoption of domestic chips.
But policy support and incremental orders are passive backstops. For China’s domestic computing industry to develop sustainably, the more important variable is the cost of generating each token.
China’s technology markets have repeatedly given new business models room to grow before time determines which companies survive, from social platforms and group-buying services to bike sharing. The same logic now applies to computing chips.
Source:
Tencent Technology
