Moore Threads

Moore Threads Technology Co., Ltd. is a Chinese fabless chip designer focused on GPUs. It was founded in October 2020 and is headquartered in Beijing. The company was started by Zhang Jianzhong, a former Nvidia executive who led Nvidia’s China operations.

‍

Moore Threads designs “universal GPU” silicon and sells products that span consumer graphics and data center compute. Its lineup includes add-in cards such as the MTT S80 for graphics workloads and the MTT S4000 aimed at large model training and inference, plus rack and cluster solutions under the KUAE platform. A core part of the strategy is the in-house MUSA software stack, including tooling like MUSIFY that targets CUDA code migration.

‍

The company has operated under tighter external constraints since it was added to the U.S. Entity List in October 2023. In December 2025, Moore Threads listed on the Shanghai Stock Exchange STAR Market under ticker 688795, moving from venture-backed growth to public-market funding for continued GPU and software development.

Moore Threads is a fabless GPU company that earns revenue by selling GPU-based hardware and scaling deployments with its own software stack and data center solution bundles.

‍

How Moore Threads makes money

‍

1. GPU IP and chip road map
   The company develops successive GPU architectures and then commercializes them through board-level and system products. Its filings describe four internally developed GPU architecture generations and a product family that spans AI compute, professional graphics, desktop graphics, and AI SoC products.
   ‍
2. AI compute cards and cluster deployments
   The center of gravity has shifted to data center workloads. The MTT S4000 targets large model training and inference and pairs with MTLink for multi-GPU scaling. The KUAE offering packages GPUs plus an 8-GPU server platform as a “full-stack” AI data center solution.
   ‍
3. Graphics acceleration products
   Moore Threads also sells desktop and professional graphics products, such as the MTT S80 for PC graphics workloads, alongside server-side rendering and visualization use cases described in its filings.
   ‍
4. Software stack and ecosystem lock-in
   MUSA is the core software layer, with MUSIFY positioned for CUDA code migration. The company also maintains a PyTorch extension, torch_musa, which mirrors PyTorch patterns and routes workloads to the MUSA backend.
   ‍
5. Go-to-market
   Filings describe sales into Chinese operators, cloud service providers, AI enterprises, and research institutions, with an order pattern that includes framework agreements and short delivery cycles once purchase orders land.

‍

Market position

‍

• Revenue mix shows a fast pivot toward AI compute. One report based on the prospectus states AI compute products made up 77.63% of revenue in 2024 and 94.85% in H1 2025, reflecting a move from graphics-first into data center demand.
  ‍
• Scale is rising, losses remain large. The company disclosed 2024 revenue of RMB 438.46 million and guided 2025 revenue to RMB 1.45 to 1.52 billion, with a 2025 net loss expected at RMB 0.95 to 1.06 billion.
  ‍
• Share remains small versus incumbents. A September 2025 report citing filings says Moore Threads’ share in several domestic submarkets was below 1% in 2024, while Nvidia and AMD remain dominant reference points in the GPU market.
  ‍
• Strategic tailwinds and constraints come from geopolitics. China’s domestic substitution push supports demand, while U.S. export controls and the U.S. Entity List add supply-chain and ecosystem friction.

China is Moore Threads’ core market and the main source of commercial traction. In its STAR Market filings, the company says it works with telecom operators, cloud service providers, AI enterprises, and research institutes. The same filing shows a sharp shift toward data center demand, with AI intelligent computing products at 77.63% of 2024 revenue, and cluster products at 42.42%.

‍

In December 2025, Moore Threads raised RMB 8 billion in its Shanghai STAR Market listing, giving it local-market funding for R&D and capacity ramp.  Demand is tied to China’s domestic substitution push as U.S. restrictions limit access to Nvidia-class AI GPUs. Moore Threads forecasts 2025 revenue of RMB 1.45 to 1.52 billion, up from RMB 438 million in 2024.  Reuters also reported China Unicom’s Qinghai data center project plans to add Moore Threads chips as part of a broader domestic supply mix.

Moore Threads’ growth story is tied to China’s demand for domestic AI compute and to its ability to ship a competitive GPU platform, including drivers, compilers, and framework support. After listing on Shanghai’s STAR Market in December 2025 and raising RMB 8 billion, the company published its first post-listing guidance, projecting 2025 revenue of RMB 1.45 to 1.52 billion with a net loss of RMB 1.04 to 1.15 billion. It also said its S5000 GPU entered mass production in 2025.

‍

Key growth drivers include

‍

1. Domestic AI infrastructure build-out
   Large state-linked projects are expanding local GPU demand. Reuters reported China Unicom’s Qinghai data center plans to add Moore Threads chips alongside other domestic suppliers.
   ‍
2. Product cadence moving up the stack
   At its MUSA Developer Conference, Moore Threads presented the next “Huagang” architecture, with “Lushan” (consumer GPU) and “Huashan” (AI GPU) positioned as the next major steps in performance and scaling via MTLink.
   ‍
3. Software stack maturation
   Ecosystem work is a gating factor for GPU adoption. The public torch_musa project shows active releases into January 2026, including updates aimed at operator coverage and distributed training performance.
   ‍

Challenges ahead

‍

• Manufacturing and supply-chain constraints linked to U.S. export controls and the Entity List designation since October 2023.
  ‍
• Ecosystem depth versus CUDA, including tooling stability, framework coverage, and performance consistency at scale.
  ‍
• Crowded domestic field, with multiple China-based accelerators competing for the same cloud and state-led budgets.
  ‍

What to watch in 2026: shipment ramp of S5000-based systems, progress of KUAE-style cluster deployments, and tangible adoption signals for the Huagang generation across both data center and graphics.