Moore Threads Crashes 20% as IPO Lockup Expires: China’s Nvidia Challenger, GPU Demand, Software Constraints, and the Domestic AI Chip Race

Moore Threads entered September 7, 2026 with a very different market structure from the one investors had been trading since its December IPO. A block of 25,774,510 previously restricted IPO shares became eligible for trading, equal to 5.48% of the company’s total equity and nearly as large as the entire unrestricted share base that existed before the unlock.

The Shanghai STAR Market reaction was immediate: Moore Threads opened at RMB 498 and then fell to the 20% daily limit at roughly RMB 415.48 during morning trading, pushing its market capitalization below RMB 200 billion. The price move is therefore not just another volatile session in an AI stock; it is a direct test of how a thin post-IPO float absorbs a sudden expansion in tradable supply.

The selloff also lands at a strategically important point for China’s domestic GPU industry. Moore Threads is trying to compete not only through silicon but through a vertically integrated software and systems stack built around MUSA, AI training and inference products, data-center systems, migration tools and cluster software. That makes the September unlock useful as a market event and as a broader stress test of whether investors are willing to continue assigning premium valuations to domestic AI-compute platforms before their software ecosystems reach the depth and maturity of Nvidia’s CUDA-centered platform.

··········

⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣

THE IPO LOCKUP EXPIRY CHANGED THE FLOAT IN ONE SESSION.

The September 7 unlock released a large block of institutional IPO shares into a stock that had previously traded with a relatively narrow unrestricted base.

Moore Threads listed on the Shanghai STAR Market on December 5, 2025 after issuing 70 million A-shares. Immediately after the IPO, the company had 470,028,217 total shares, but only 29,382,386 were unrestricted. The 25,774,510 shares released on September 7 came from the IPO’s offline placement tranche and had been subject to a nine-month lockup.

That distinction matters because the unlocked block represented only 5.48% of total equity but an exceptionally large percentage of the previously tradable float. Adding 25.77 million shares to a pre-unlock unrestricted base of 29.38 million increases the potential tradable supply to roughly 55.16 million shares, an increase of about 87.7%. Not every newly eligible share must be sold, but the market suddenly has to price the possibility that a much larger pool of institutional holders can reduce positions.

........

LOCKUP / MARKET ITEM	VALUE	WHY IT MATTERS
STAR Market listing	December 5, 2025	Starts the nine-month lockup clock for the relevant offline-placement shares.
Unlocked shares	25,774,510	Large new block becomes eligible for trading in a single session.
Share of total company equity	5.48%	Moderate relative to total shares, but large relative to the prior float.
Pre-unlock unrestricted shares	29,382,386	Shows how narrow the tradable base was before September 7.
Approx. post-unlock unrestricted shares	55,156,896	Potential tradable supply rises sharply if all newly eligible shares are considered.
Increase vs. prior unrestricted base	~87.7%	Explains why a 5.48% equity unlock can create disproportionate liquidity pressure.
Lockup term	9 months	Standardizes the timing of when this institutional block could become tradable.
September 7 market reaction	Intraday 20% limit down	Price discovery immediately reflected the larger available float and valuation sensitivity.

........

The practical takeaway from the table is that the denominator matters. A 5.48% unlock sounds manageable when compared with total shares outstanding, but it is much more significant when compared with the shares that investors could actually trade the day before. For a high-valuation technology stock, that change can compress scarcity premiums very quickly.

The unlocked shares also came from institutions that received stock through the IPO’s offline placement process, not from a newly issued financing round. The event therefore does not dilute existing shareholders by increasing total shares outstanding; instead, it changes liquidity, possible selling supply and the balance between locked and freely tradable ownership.

··········

⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁣⁤

THE 20% DROP IS A LIQUIDITY SHOCK, NOT A NEW OPERATING LOSS.

The stock move should be separated from the company’s operating performance because the trigger is a capital-market structure event rather than a newly disclosed deterioration in the business.

Moore Threads’ shares opened at RMB 498 on September 7 and then reached the STAR Market’s 20% daily downside limit near RMB 415.48. The market value fell below RMB 200 billion during the session. The company’s securities representatives publicly characterized the day as a major release of restricted offline-placement shares and said operating fundamentals remained normal.

That does not mean the market reaction is irrelevant to fundamentals. Lockup expiries force investors to re-evaluate whether the valuation can be supported when scarcity decreases. Before an unlock, a limited float can amplify upside because relatively little stock is available. After an unlock, the same valuation has to be sustained with more shares potentially offered by institutions whose cost basis may be far below the prevailing market price.

The distinction is important: a lockup expiry changes who can sell and how much stock can circulate; it does not itself change revenue, product performance, customer contracts or the number of total shares outstanding.

For Moore Threads, this is especially relevant because the company came to market as one of the most visible listed proxies for China’s domestic GPU build-out. Early post-IPO enthusiasm can support valuations that imply years of rapid execution. Once the float expands, investors tend to focus more aggressively on measurable progress: revenue conversion, gross margin, operating leverage, product cadence, software compatibility, data-center deployments and the cost of serving large-scale training and inference workloads.

The company’s first-half 2026 numbers provide a more constructive operating backdrop than the share-price move alone suggests. Moore Threads reported revenue of approximately RMB 1.736 billion for the first six months of the year, up about 147.42% year over year, while its net loss narrowed to roughly RMB 11.56 million from a much larger loss in the prior-year period. Those figures do not settle the valuation question, but they show why the market is debating execution quality rather than treating the company as a pre-revenue semiconductor concept.

A second capital-markets layer is also developing: Moore Threads has been moving toward a potential Hong Kong listing. An additional venue could broaden the investor base and improve financing flexibility, but it would also reinforce the same core requirement exposed by the September 7 selloff: a premium AI-chip valuation ultimately has to be justified by sustained product demand and an ecosystem that customers can deploy at scale.

··········

MUSA IS THE STRATEGIC LAYER THAT DETERMINES WHETHER THE HARDWARE CAN SCALE.

Moore Threads is not competing only with GPU specifications; it is building the software, libraries, migration tools and cluster layer needed to make its hardware usable for production AI workloads.

Nvidia’s strongest competitive advantage is not reducible to raw accelerator performance. CUDA, libraries, compilers, communications software, profiling tools, framework integration and years of developer familiarity collectively reduce the cost of moving a model from research code into reliable production. A domestic Chinese GPU vendor therefore needs a credible software path for existing workloads, not merely a chip that can execute matrix operations.

Moore Threads’ answer is MUSA, its Meta-computing Unified System Architecture. The current MUSA SDK includes a compiler and runtime, accelerated math libraries, deep-learning libraries, multi-GPU communication software, CUDA migration tooling, profiling utilities and compiler integrations. The company also markets AI training and inference suites, cloud-native deployment tools and data-center systems around the same platform.

........

STACK LAYER	MOORE THREADS COMPONENT	COMPETITIVE ROLE
GPU compute platform	MUSA architecture and MTT GPU family	Provides the execution model tying Moore Threads hardware to parallel-compute workloads.
Developer SDK	MUSA SDK 5.x	Packages compiler, runtime, libraries, migration and optimization tooling into one development environment.
CUDA migration	musify and CUDA-compatibility tooling	Reduces the engineering burden of moving existing CUDA-oriented code toward MUSA.
Deep-learning acceleration	muDNN	Targets optimized neural-network operators used in training and inference.
Multi-GPU communication	MCCL	Supports single-node and multi-node GPU communication, critical for distributed AI workloads.
Compiler ecosystem	Triton-MUSA and TileLang-MUSA	Connects modern AI kernel-development workflows to the Moore Threads platform.
Inference / training products	AI training suite, AI inference suite, MTT S5000	Extends the platform from SDK components into production-oriented AI systems.
Cloud-native operations	KUAE cloud-native toolkit	Targets containerized and Kubernetes-based GPU deployment at cluster scale.
Performance analysis	Moore Perf and muPTI	Provides profiling and tracing needed to diagnose kernels and optimize production workloads.

........

This stack explains why the software question is more consequential than a simple benchmark comparison. A production customer needs framework support, stable drivers, distributed communication, kernel libraries, debugging tools, container integration and repeatable performance across model versions. Missing any one layer can erase an apparent hardware advantage by increasing engineering time or reducing utilization.

Moore Threads has also been pushing the MTT S5000 as a universal GPU for AI training and inference and pairing it with full-stack data-center offerings. That is strategically logical because the fastest route to revenue for domestic accelerators may not be winning every benchmark against Nvidia, but delivering a sufficiently integrated Chinese supply chain for customers that prioritize availability, local deployment, procurement certainty and compatibility with domestic infrastructure.

The remaining constraint is ecosystem depth. Migration tools can lower switching costs, but a mature CUDA workload often depends on third-party libraries, custom kernels, monitoring systems, orchestration logic and developer practices accumulated over years. Moore Threads has to keep closing those gaps while simultaneously improving hardware performance, memory behavior, interconnect efficiency and inference economics.

··········

THE SELL-OFF SHIFTS THE TEST FROM SCARCITY TO EXECUTION.

After the unlock, Moore Threads has to support its valuation with a larger float while proving that domestic GPU adoption can translate into durable software usage, customer expansion and economically competitive AI compute.

China’s demand for domestic AI accelerators is structurally supported by rapid model deployment, data-center construction, supply-chain localization and continuing restrictions on access to the most advanced foreign chips. Those conditions create a large addressable market for Moore Threads and other domestic vendors, but they do not remove competition between them. Buyers will still compare throughput, memory capacity, cluster scaling, power efficiency, inference cost, software stability and the amount of engineering work required to port existing applications.

That competition becomes more intense as Chinese suppliers move from isolated accelerator cards toward complete platforms. The winning architecture is unlikely to be determined by one generation of silicon. It will depend on whether developers can train and serve models reliably, whether operators can run large clusters at high utilization, whether migration from CUDA is practical, and whether hardware availability can compensate for gaps in ecosystem maturity.

The September 7 lockup expiry therefore acts as a useful dividing line. Before the unlock, Moore Threads traded with a much tighter public float and a scarcity component embedded in the stock. After the unlock, the available share base is materially larger, making valuation more sensitive to actual execution and less dependent on limited supply.

If revenue growth remains strong, the S5000 and newer architectures gain production customers, MUSA compatibility improves and the company demonstrates credible economics for large-model inference and training, the larger float can eventually support a more liquid and fundamentally anchored market. If software friction, hardware performance gaps or customer concentration slow adoption, the same larger float can make those weaknesses visible faster.

The 20% limit-down session is therefore best read as two events at once: a mechanical repricing caused by an 87.7% expansion in potential unrestricted share supply, and a broader valuation test for one of China’s most closely watched domestic GPU platforms.

For Moore Threads, the next phase is less about proving that investors want a Chinese Nvidia alternative and more about proving that customers can deploy the company’s hardware and software at scale. The stock unlock changes the market structure in a day; closing the gap in AI-compute ecosystems will take sustained product execution across silicon, systems, compilers, libraries and production infrastructure.

·····

FOLLOW US FOR MORE.

·····

·····

DATA STUDIOS

·····

[datastudios.org]