Moore Threads Falls 20% on Lock-up Expiry, Domestic GPUs Face Headwinds

Moore Threads, regarded as the leading domestic GPU stock, recorded its first-ever limit-down decline since its listing. On September 7, 2026, the lock-up expiry date, its share price hit the 20% daily limit down and remained locked there until the close. The direct trigger was the entry into tradable circulation of 25.7745 million shares allotted to offline subscribers at the time of the IPO. The sharp plunge, which occurred despite solid demand for AI computing, left challenges for both supply-demand dynamics and valuation. According to TMTPost, all of the top five selling brokerage branches were institutional-only seats.

Sharp Plunge on the First Day Triggered by

Large Lock-up Expiry

Trading data for the top sellers by transaction value on the first day of the expiry clearly showed concentrated selling pressure. The total sales by the top five selling seats reached 1.462 billion yuan. Converted at the closing price on that day, the sales by the five seats were equivalent to 3.5179 million shares. This accounted for about 43% of the total trading volume for the day. Selling by institutional investors amid limited free float amplified the price movement.

The shares unlocked this time account for only 5.48% of total issued shares. However, the free float before the expiry was only about 30.2255 million shares, or 6.4% of the total. The expiry expanded the free float by more than 80% at once, disrupting the supply-demand balance. The current free float is about 56 million shares, still no more than 12% of the total. In addition, about 180 million shares are scheduled to be unlocked on December 7, three months later. This portion is equivalent to about 39.55% of total issued shares. Regardless of whether selling actually occurs in the future, concerns remain that volatility in liquidity will recur.

Regarding the nature of the unlocked shares, there is a view that this was disciplined profit-taking. All of the shares unlocked this time originated from the offline allotment at the time of the IPO. In that allotment, the allocation ratio to Category A — which includes public funds, social security funds, pensions, insurance funds, and Qualified Foreign Institutional Investors (QFII) — reached 98.44%. Allocation to Category B was only 1.56%. Within Category A, strategic allocation products under public funds accounted for a large share. Selling at maturity can be interpreted as cashing in unrealized gains rather than a deterioration in the long-term outlook. On the other hand, doubts about whether fundamentals are truly solid have not disappeared.

Fading Scarcity Value of Domestic GPU Stocks

The previous market valuation was supported by positioning as a domestic alternative to NVIDIA. The scarcity of domestic GPUs themselves was compounded by computing demand in the AI era. Moore Threads attracted high expectations as the first domestic GPU stock. Now the competitive environment has changed completely, and the premium derived from scarcity is peeling away. In China, in addition to MetaX, Biren Technology, and Enflame, Iluvatar has also emerged. Each company has largely achieved its listing goals, broadening the options for investors.

In the field of AI computing, strong players such as Cambricon, Huawei Ascend, and Hygon compete. There is a view that Moore Threads’ former valuation was partially inflated. Share price trends also suggest a correction across the sector. In June this year, Iluvatar and Biren Technology both hit all-time highs since listing. In July, MetaX renewed its post-listing high at 1,033 yuan. Moore Threads also touched its year-to-date high of 769.5 yuan. As of the close on September 10, Biren Technology and MetaX had fallen more than 40% from their highs. Moore Threads has fallen to nearly half its high, and Iluvatar has dropped 60%. The domestic GPU sector is seen as having entered a cyclical adjustment phase.

Revenue Structure Skewed by Cloud Concentration

Amid intensifying competition, Moore Threads has a unique technological strength. It is positioned as the only full-function GPU company among the four. It is one of the few companies in China with operational experience running large-model training clusters with tens of thousands of cards. JD Cloud has announced a plan to build a full-function GPU cluster on the scale of 100,000 cards. Moore Threads’ full-function GPUs are said to form the core computing infrastructure for it. There remains significant room for differentiation on the technology front.

Financial results have not yet established an absolute advantage. In the first half of this year, Moore Threads’ revenue was about 1.736 billion yuan. MetaX posted about 1.324 billion yuan and Biren Technology about 1.236 billion yuan. While there are differences, the three companies are at a roughly similar stage of mass shipment. The core issue lies in the skewed revenue composition. Moore Threads’ revenue is highly concentrated in its cloud computing product line. The breakdown is centered on sales of intelligent computing boards, integrated machines, and intelligent computing clusters for cloud computing. In the first half of 2026, revenue from this product line was about 1.693 billion yuan, accounting for 97.52% of the total. Technological differentiation has not been sufficiently reflected in revenue. In reality, it is competing with MetaX and Huawei Ascend in the training and inference market.

A shift in business model is also underway at the same time. Moore Threads is transitioning from selling boards as a computing supplier to selling solutions as an AI infrastructure supplier. In the first half of this year, it sold and delivered KUAE intelligent computing clusters under day-to-day operating sales contracts with customers. The total contract value was 660 million yuan. If the future business model takes hold, it is seen as having greater industrial potential than simply selling boards. The trade-off is pressure on gross margin. Cluster deliveries include large quantities of third-party equipment such as externally procured memory devices, storage media, and switches procured from outside. The corresponding gross margin is relatively limited. A rising share of cluster deliveries pushes down the company’s overall gross margin. In the first half of this year, the gross margin was 56.95%, down 12.2 percentage points year-on-year. It was about 67.35% in the first quarter and 49.26% in the second quarter. As losses narrow at each company and valuation criteria shift, friction during the transition period is emerging.

Surging Inventory and the Reality of Ample

Cash on Hand

In addition to the growth potential of revenue and gross margin, attention to the balance sheet is increasing. In terms of cash reserves, Moore Threads’ performance is solid. At the end of the first half of this year, its cash on hand was 6.492 billion yuan. Financial assets held for trading were 1.44 billion yuan, with the two items totaling about 8 billion yuan. The net proceeds from the A-share issuance last year were about 7.576 billion yuan. Of the remaining amount after deducting about 2 billion yuan already used, about 2.725 billion yuan was kept in a dedicated account for the raised funds. In addition, 2.9 billion yuan was allocated to cash management products such as large-denomination time deposits and structured deposits. The company has also announced a plan to raise funds through a Hong Kong listing. If realized, it would expand room for trial and error and competitive leadership.

In contrast, the swelling scale of inventory is a point of attention. Inventory at the same time surged from 512 million yuan in the first half of last year to 3.55 billion yuan. By breakdown, the book value of raw materials was about 1.51 billion yuan. Work in progress was 1.171 billion yuan and finished goods inventory was 844 million yuan. Behind this is the continuing rise in prices of semiconductor raw materials. The shift to solution-based AI infrastructure supply has also prompted advance stockpiling. A certain level of reserves is necessary to ensure a stable supply chain. The new concern is the possibility of impairment due to declines in the value of finished goods inventory. Semiconductors are one of the products whose value declines rapidly. If a product with superior cost performance appears between production and sale, older models face the risk of impairment. Inventory turnover days reached 587.7 days. Amid competition with MetaX, Huawei Ascend, Alibaba T-Head, Baidu Kunlun and others, the potential risk cannot be ignored.

Production capacity constraints are also a common issue for the four domestic GPU companies. Whether the supply chain and foundry capacity can keep up with demand growth is the focus. TMTPost, citing information from external institutions, touched on supply-demand trends for 2025. The detailed analysis is cut off midway, and additional disclosure is needed to verify the full picture. At this point, it is essential to look at trends in inventory, cash, and order backlogs together.

The Crucial Test for Domestic AI Computing

Lies Ahead

From the perspective of the progress of the AI era, the development of domestic GPUs and AI semiconductors is still in its early stages. The road ahead remains long. The merits of companies cannot be judged by revenue scale alone. Gross margin, capital efficiency, and supply chain strength are added to the evaluation criteria. The case of Moore Threads illustrates the time lag between technological leadership and monetization. Operational experience on the scale of tens of thousands of cards provides technological backing. Resolving revenue concentration and stabilizing gross margin are the next challenges.

The Hong Kong listing plan will broaden funding options. Continued investment is required for R&D and ecosystem building. With a large-scale lock-up expiry approaching in December, the market’s liquidity management will also be tested. It is important to distinguish between short-term price movements and long-term fundamentals. Overcoming the hurdle for domestic computing requires both sustained demand and supply capacity.

Editorial Opinion

Looking at the short-term impact, we believe that over the next three to six months, price movements in domestic GPU stocks will be strongly influenced by the lock-up expiry schedule. Moore Threads’ large-scale expiry in December will prolong caution over supply and demand. If disciplined selling by institutional investors continues, volatility across the sector will remain elevated. It will be necessary to watch whether disclosures on orders and gross margins can act as a buffer against price swings.

From a long-term perspective, we believe that over one to three years, success or failure in the shift from board sales to AI infrastructure supply will determine winners and losers. Cluster orders such as KUAE will enhance industrial potential while pushing down gross margins. Each company will be forced to differentiate itself through its ecosystem and operational track record. We assess that companies that allocate their ample funds to R&D and strengthening their supply chains will be the ones to survive.

We pose questions from the editorial team. After the premium from scarcity peels away, what will determine the fair value of domestic GPUs? Will it be technological uniqueness, earnings quality, or the depth of operational track record? Does the figure of 587.7 inventory turnover days represent the result of advance procurement or a sign of excessive stockpiling? If the Hong Kong listing is realized, how should the use of the proceeds be judged?

References

• “摩尔线程解禁即跌停，一场国产GPU的“反厄尔尼诺”现象？”, by 互联网江湖 — 钛媒体, 2026-09-11T12:43:03.000Z (ARR)
• Source URL: https://www.tmtpost.com/8135737.html