moore threads
on this page
moore threads intelligence technology is a chinese gpu design company specializing in graphics and ai computing processors.
Update (September 2026). Moore Threads began trading on Shanghai’s STAR Market on December 5, 2025 under code 688795.13 In 2026 it added the MTT S5000 data-center GPU and consumer AI systems to its product list. The company also disclosed that it was considering an H-share issue and Hong Kong listing.14 The pre-IPO figures below are retained as historical context, not current company facts.
company overview
founded in october 2020, the company achieved unicorn status (>$1b valuation) within 100 days.1
company facts
- founder/ceo: zhang jianzhong (former nvidia global vp)
- headquarters: beijing, china
- listing: shanghai star market, code 688795
- products: consumer, workstation, and data-center gpus
- 44% owned by founder
products
consumer gpus
| model | memory | performance | price | target |
|---|---|---|---|---|
| mtt s90 | 16gb gddr6 | outperforms rtx 4060 | tbd | gaming/compute |
| mtt s80 | 16gb gddr6 | 14.4 tflops fp32 | ~$170 | gaming/ai inference |
| mtt s70 | 8gb gddr6 | similar to s80 | entry-level | general compute |
the mtt s80 has 16 gb of memory and supports the company’s MUSA software stack. Price and value depend on local availability, driver support, and the workload; the source formerly cited here did not provide a reproducible value comparison.
datacenter and professional gpus
| model | memory | bandwidth | compute | target |
|---|---|---|---|---|
| mtt s4000 | 48gb gddr6 | 768 gb/s | 200 tflops fp16 | datacenter ai |
| mtt s3000e | 24gb gddr6 | energy efficient | server/cloud focused | cloud computing |
| mtt x300 | 16gb gddr6 | same hardware as s80 | workstation optimized | professional/domestic market |
| mtt s3000 | 24gb gddr6 | 384 gb/s | 100 tflops fp16 | cloud/virtualization |
the s3000e was revealed through gpu-z 2.68.0 in august 2025, targeting energy efficiency for server deployments.10
notably, the latest quyuan architecture lacks fp8 support despite being standard in 2024 ai accelerators.2
software ecosystem
musa platform
cuda-compatible computing platform with:3
- opengl, vulkan, directx support
- opencl and sycl compatibility
- version 4.0.1 (2024)
musify tool
zero-cost cuda migration tool enabling code translation from nvidia’s ecosystem.2
mtlink interconnect
gpu interconnection technology with 240 gb/s bandwidth, supporting up to 10,000 gpu clusters.4
financing
before its public offering, Moore Threads raised private capital from corporate and state-backed investors. Historical reports estimated $800 million across six rounds.5 That number is not a current measure of company value or available cash.
major investors:
- bytedance (tiktok parent)
- tencent holdings
- hongshan (formerly sequoia china)
- beijing zhongguancun science city (state fund)
- china mobile capital
partnerships
deepseek deployment
february 2025: announced successful deployment of deepseek v3 and r1 models on moore threads gpus.6
enterprise customers
- china mobile (telecom infrastructure)
- china unicom (cloud services)
- lenovo (hardware integration)
- china energy engineering corp (computing clusters)
regulatory status
the United States added Moore Threads to the Entity List on October 17, 2023.7 The listing imposes license requirements on covered exports, reexports, and transfers. It does not, by itself, prove the absolute supplier bans or the fixed 7 nm manufacturing limit previously stated here.
public listing
the Shanghai Stock Exchange approved trading in Moore Threads shares under code 688795 beginning December 5, 2025.13 In August 2026, the company disclosed that it was considering an H-share issue and a listing in Hong Kong. That disclosure described a plan under study, not a completed secondary listing.14
technical performance
competitive positioning
moore threads gpus target different market segments than nvidia flagship products:4
- datacenter: mtt s4000 (200 tops int8) vs nvidia a100 (624 tops int8)
- consumer: mtt s80 targets graphics and local compute
- inference: the company reports support for DeepSeek models
- evidence limit: this page has no reproducible cross-vendor benchmark for the s80 or s90
manufacturing constraints
trade controls restrict access to some foundries, equipment, and software. Public sources used here do not establish Moore Threads’ current foundry, process node, or production allocation. The company’s stated product strategy emphasizes:
- cost-effectiveness over peak performance
- inference optimization rather than training workloads
- software ecosystem development (musa platform)
- multi-gpu clustering capabilities (mtlink scaling to 10,000 processors)12
driver improvements
early products faced optimization challenges, but 2024-2025 reviews note significant driver improvements and better compatibility with ai frameworks.
references
[2] videocardz. (2023, december). moore threads introduces mtt s4000 48gb ai gpu.
[3] tom’s hardware. (2024). china’s moore threads polishes homegrown cuda alternative.
[5] yahoo finance. (2023, february). china’s moore threads valued at $3.4 billion.
[7] tom’s hardware. (2023, october). us sanctions on biren and moore threads.
[10] chinese gpu maker reveals new workstation gpu: moore threads mtt x300 | tom’s hardware.
[12] chinese gpu maker moore threads’ mtlink fabric tech challenges nvidia’s nvlink | reddit hardware.
[13] shanghai stock exchange. (2025, december 4). notice of moore threads listing and trading.
[14] moore threads. investor relations and 2026 company disclosures.