The archive · Hardware & Devices · Strategic decision · 2020–2026
MetaX's domestic-GPU bet: ¥426K revenue to a +693% STAR debut, ¥332B market cap
Ex-AMD engineers bet China's AI buildout needed homegrown CUDA-compatible GPUs; five-year-old MetaX's STAR debut closed +693% at a ¥332B market cap.
MetaX（沐曦科技 / 沐曦集成电路）
What the business is
MetaX (沐曦) designs high-performance general-purpose GPUs for AI inference (曦思 N), general computing (曦云 C) and graphics (曦彩 G), with its own MXMACA software stack that aims to be CUDA-compatible; products run in 10+ AI computing clusters across China.
Starting capital：Raised roughly ¥2B+ privately from Sequoia China, ZhenFund, Matrix Partners China, Lightspeed China, 国调基金 and others (2020-2024); IPO raised ~¥4.2B at ¥104.66 per share.
How it started
Chen Weiliang spent 17 years at AMD Shanghai (senior director; tape-out of 15 high-performance GPU products), then founded MetaX in Shanghai in October 2020 with a team averaging nearly 20 years of GPU experience, including two former AMD principal scientists. The bet: self-developed GPU IP with CUDA-compatible software could win China's fast-growing domestic AI-chip market.
What happened
Revenue went from ¥426K (2022) to ¥743M (2024) to ¥1.24B in 2025 Q1-Q3 (+453.5% YoY), as the 曦云 C500 reached volume production and inference products deployed in 10+ AI clusters; cumulative losses topped ¥3.2B (¥1.41B in 2024 alone), with R&D at 56% of revenue. MetaX raised roughly ¥2B+ privately, then listed on the STAR Market on 2025-12-17 at ¥104.66, the year's second-highest STAR issue price.
How it ended up
Still scaling: closed its first day +692.95% at ¥829.90 (market cap ~¥332B); orders on hand were ¥1.41B as of 2025-09-05 and management targeted breakeven in 2026 — while CUDA-migration friction, 7nm supply limits and rivals (Moore Threads, Biren, Huawei Ascend) remain.
Background
MetaX (沐曦) is a Shanghai GPU maker founded in October 2020 by Chen Weiliang, a 17-year AMD veteran who led tape-out and mass production of 15 high-performance GPU products, together with a team that includes two former AMD principal scientists. The bet: China's AI buildout would need domestic high-performance GPUs, and a CUDA-compatible software stack could let customers switch without rewriting code.
Revenue went from ¥426K in 2022 to ¥743M in 2024 and ¥1.24B in 2025 Q1-Q3 (+453.5% YoY), with products deployed in 10+ AI computing clusters; cumulative losses still topped ¥3.2B. On 2025-12-17 MetaX listed on Shanghai's STAR Market at ¥104.66 (the year's second-highest STAR issue price): 5.18M investor accounts subscribed, and the stock closed its first day +692.95% at ¥829.90, a market cap near ¥332B (~US$47B).
The company is still scaling: C600 is positioned between Nvidia's A100 and H100, C700 tape-out is planned for H2 2026, and breakeven was targeted for 2026 with ¥1.41B of orders on hand as of 2025-09-05. The open questions are whether CUDA compatibility costs real performance, whether 7nm supply holds, and whether it can avoid a price war with Moore Threads, Biren and Huawei Ascend.
What has to be true
- Hiring a 17-year AMD veteran and two former AMD principal scientists gave MetaX instant credibility and full tape-out-to-shipping skill, impossible to fake in a crowded market.
- CUDA compatibility via its own MXMACA stack lowered customers' switching cost — the classic wedge for breaking into a market dominated by a platform incumbent.
- Timing: 2024-2025 inference demand and the domestic-substitution tailwind turned a ¥426K-revenue company into ¥743M in two years and a ¥332B-market-cap stock.
- The IPO itself was the funding event: ¥4.2B raised at 0.033% lottery odds to fund next-gen silicon (C600/C700) and close the supply-chain loop.
What can be applied
Proven technical pedigree and software compatibility let a five-year-old chip startup ride a substitution wave to a mega-IPO — but the valuation holds only while the roadmap closes the gap.
Aftermath
As of 2026-09-02, MetaX is live and scaling: it is deploying the C600 (positioned between Nvidia A100 and H100), planning the C700 tape-out in H2 2026, and aiming for breakeven in 2026 with ¥1.41B of orders on hand (2025-09-05). It must keep closing the CUDA-performance gap and secure advanced-node supply (7nm) while competing with Moore Threads, Biren and Huawei Ascend.
Sources
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card