Moore Threads

What Moore Threads is

Moore Threads is a Chinese chip company that designs graphics processing units (GPUs) and the systems built around them, from a single accelerator card up to complete computing clusters. Its catalogue covers hardware for artificial-intelligence workloads as well as graphics products, and the company also ships software meant to make it easier to move existing programs onto its chips^[1].

Origins

The company was founded in 2020 and formally began operations in October of that year^[4]. Xinhua Finance reported that Moore Threads was founded in October 2020 and that its founder, Zhang Jianzhong, had previously served as Nvidia's global vice president and China general manager^[5].

Products

Moore Threads describes its product line as spanning the MTT S5000 training-and-inference GPU card, MTT SGX5000 servers, MTT KUAE intelligent computing clusters, the MTT S3000 and MTT S80 graphics cards, and the MTT AIBOOK AI laptop^[1]. The division of labour is visible in that list: cards and clusters for data-centre AI work, stand-alone graphics cards, and a laptop.

The MTT S5000, the card positioned for both training and inference, is based on the PH100 chip, which is built on the company's fourth-generation MUSA architecture, known as Pinghu^[1]. Moore Threads said PH100 passed the China Information Technology Security Evaluation Centre's security and reliability assessment in 2026^[1].

Software: MUSA and MUSIFY

For any new GPU vendor the hardest practical obstacle is software. Most AI code in use is written for one dominant programming model, and rewriting it is expensive. Moore Threads says its MUSA architecture is highly compatible with CUDA syntax, so developers can reuse existing code when moving to its chips^[1]. It also ships MUSIFY, an automatic porting tool that the company says turns weeks of adaptation work into hours^[1].

Moore Threads also reports keeping pace with new model releases: it said it completed day-zero adaptation of the MiniMax M3.1 Flash model in October 2026^[1]. Such claims are about whether a newly released model can run on the hardware from the day it appears — a measure customers use when judging whether a chip vendor can keep up with the fast release cycle of AI models.

Listing and finances

Moore Threads listed on the STAR Market on 5 December 2025 under the code 688795, at an issue price of 114.28 yuan per share^[3].

The company's 2025 figures show rapid growth from a small base alongside continuing losses. Revenue for 2025 was 1.505 billion yuan, up 243.37% from 438 million yuan in 2024^[2]. The net loss attributable to shareholders was 1.024 billion yuan, narrowing 36.70% from a loss of 1.618 billion yuan the year before^[2]. Total assets stood at 15.308 billion yuan at the end of 2025, up 116.14% from the start of the year, and equity attributable to shareholders reached 11.435 billion yuan^[2]. Put simply, the company earned about 1.5 billion yuan in 2025 and reported a loss of about 1 billion yuan; revenue grew faster than the loss shrank.

Where it stands

Moore Threads' position rests on two things: a product line that runs from cards and servers to clusters and an end-user device^[1], and a software strategy built on compatibility with the CUDA ecosystem^[1]. In the year it listed, its total assets more than doubled^[2]; in 2026 its public statements emphasised security certification and fast support for new models^[1]. For a company that is still loss-making, how complete the hardware line is, how much work a software migration takes, and whether a new model runs on day one are the questions customers are most likely to ask before they buy.