Anatrix
Category: AI Chips / Semiconductors
Beijing-based startup designing modern analog computing chips that solve matrix equations in one step, spun out of Peking University research. Anatrix was founded in 2025. The company is led by 向锐 (Xiang Rui). Based in Huairou District, Beijing, China. Latest round: Series A. Key investors include Matrix Partners China (经纬创投, led June 2026 round), Frees Fund (峰瑞资本), Alpha Commune (阿尔法公社), iFlytek Ventures (讯飞创投), Zhongying Venture (中赢创投), Galbot (银河通用机器人), Zhongheyongli (中泓永利), Yanyuan Capital (燕缘创投), Huarui Holdings (华睿控股), C&D Emerging Investment (建发新兴投资), Meihua Venture Capital (梅花创投).
- Founded
- 2025
- Headquarters
- Huairou District, Beijing, China
Value proposition
Maps matrix equations directly onto resistive-memory (RRAM) circuits so that Ohm's law and Kirchhoff's current law solve them in a single step. This avoids the iterative multiply-accumulate decomposition used by digital GPUs, and the chip is designed for mature process nodes (28nm and above).
Products and solutions
AX100: first fully integrated one-step matrix-equation-solving analog computing chip, taped out October 2026. Company-reported figures: solve time under 100ns and about 80mW per solve. Also a compiler toolchain that maps high-level algorithms to the hardware.
Unique value
A different compute paradigm (analog matrix solving) from GPU-style digital chips. It claims 24-bit fixed-point-class precision and does not depend on advanced process nodes.
Target customer
AI model training teams (second-order optimization), 6G/wireless communications equipment makers, embodied-intelligence/robotics companies, and scientific computing users
Industries served
AI training, 6G communications, embodied AI/robotics, scientific computing
Technology advantage
Built on the Peking University team's October 2025 Nature Electronics paper, which raised analog computing precision from about 1% to about 1e-7. The company reports 20-207x throughput gains on 128x128 MIMO matrix tasks, 92% and 56% lower training-update cost with second-order optimization, and 1.5-5x end-to-end speedups for embodied algorithms. All figures are self-reported and not independently verified. The chip has only been taped out, with no mass production, customer adoption or third-party benchmarks.
How they differentiate
Targets one-step solving of matrix equations (matrix inversion/linear systems) with analog RRAM arrays, rather than the compute-in-memory matrix multiply-accumulate approach of most domestic compute-in-memory players and analog inference startups.
Main competitors
Witmem (知存科技), Mythic, EnCharge AI
Key partnerships
Strategic investors Galbot (银河通用机器人) and iFlytek Ventures (讯飞创投), technology origin in Peking University's Institute for Artificial Intelligence
Major milestones
Nov 2025: founded, with iFlytek Ventures co-funding, Feb 2026: angel round, Jun 2026: several-hundred-million-RMB round led by Matrix Partners China, Oct 2026: AX100 taped out, Nov 2026: new hundred-million-RMB-scale round and AX100 announced
Growth metrics
Four funding rounds within about one year of founding, counting the November 2025 seed capital from iFlytek Ventures at incorporation
Market positioning
Early-stage analog computing chip startup positioned as an alternative compute paradigm for AI, 6G and robotics, in China's fast-growing compute-in-memory and analog AI chip space
Geographic focus
China