MetaX

MetaX (沐曦; formally MetaX Integrated Circuits (Shanghai) Co., Ltd.) was founded in Shanghai on September 14, 2020, by Chen Weiliang, a thirteen-year AMD China executive, together with co-founders Peng Li and Yang Jian, who also came from AMD's GPU engineering ranks. The AMD pedigree of its founding team — echoed at rival Biren and, in part, at Enflame — has become a recurring pattern among China's most credentialed fabless GPU startups, whose engineers cut their teeth on commercial GPU architectures before returning to build domestic alternatives to the company that trained them. MetaX organizes its silicon into three product lines: N-series chips for AI inference, C-series chips for AI training and general-purpose GPU computing, and G-series chips for graphics rendering. Its highest-profile product, the C600, integrates HBM3e high-bandwidth memory with 144GB of on-package capacity and supports FP8 precision, which MetaX claims delivers two-to-three-times the training throughput of an FP16 baseline. First unveiled in mid-2025 and reaching risk production by the end of 2025 ahead of a first-half-2026 commercial ramp, MetaX describes the C600 as its first product built on what the company calls a "fully domestic" supply chain — an explicit self-sufficiency claim that distinguishes its marketing from peers who are more circumspect about supply-chain sourcing. That "fully domestic" framing deserves scrutiny. The C600 does appear to be fabricated on SMIC's 7-nanometer-class process rather than TSMC's, consistent with the rest of its peer group. But China's domestic high-bandwidth-memory manufacturing capability remains nascent: SK Hynix, Samsung, and Micron continue to dominate global HBM production, and Chinese memory makers such as CXMT are only beginning to bring domestic HBM output online at meaningful volume. Whether MetaX's "domestic supply chain" claim for the C600 reflects genuinely domestic-sourced HBM3e, as opposed to a narrower claim about domestic design, packaging, or non-memory components, is not fully verifiable from public disclosure — a gap emblematic of the broader aspirational quality of China's chip self-sufficiency messaging relative to the country's still-developing advanced-memory supply chain. MetaX's STAR Market debut on December 17, 2025 — twelve days after Moore Threads' listing — became the second entry in China's late-2025 wave of domestic GPU IPOs, and by some measures drew even more intense retail demand than its predecessor. The offering priced at 104.66 yuan per share, raising roughly 4.2 billion yuan (about $593-596 million), and was oversubscribed nearly 2,986 times. Shares surged as much as 755% intraday on debut and closed the first day up roughly 693%, pushing MetaX's implied market capitalization past 300 billion yuan (about $42.6 billion) and making founder Chen Weiliang a billionaire within hours of listing. MetaX has since filed for a secondary listing on the Hong Kong Stock Exchange, targeted for the second half of 2026, to fund C600 commercialization, C700 development, and cross-border software-ecosystem expansion. Unlike Moore Threads and Biren, MetaX has not, as of mid-2026, been placed on the US Entity List, leaving it nominally freer to pursue foreign supply relationships. In practice, though, broader US controls on advanced computing exports and packaging technology mean MetaX still cannot access TSMC's leading-edge nodes or NVIDIA-class CoWoS packaging, and the company's own disclosures acknowledge reliance on imported HBM, EDA software, and third-party IP — components that could become directly exposed to future export-control action given the trajectory of US-China chip policy. Fabrication instead runs through SMIC, with the same density and yield disadvantages relative to TSMC that constrain Cambricon, Biren, and Moore Threads, and MetaX depends on Cadence and Synopsys EDA tools whose licensing already carries geopolitical risk regardless of MetaX's own listing status. MetaX's customer base leans on the same domestic cloud and enterprise procurement dynamic that supports its peers: Chinese hyperscalers and government-linked data center operators substituting toward SMIC-fabbed domestic silicon under supply-chain-security mandates and export-control pressure, even where MetaX's raw performance trails NVIDIA's leading GPUs. That policy tailwind, combined with its N-series inference and G-series graphics lines rounding out the C-series training and general-purpose push, gives MetaX a broader product surface than narrower AI-only accelerator plays. Its near-term trajectory will be shaped by whether its capital-markets momentum — two record-breaking IPOs from the same Chinese GPU cohort within two weeks of each other — translates into durable commercial deployment beyond the sentiment-driven surge in its share price, and whether the C-series can hold share once Chinese buyers have more than one credible SMIC-fabbed alternative to choose from. Its "domestic supply chain" positioning gives it a distinct marketing angle in a crowded field of Chinese GPU challengers, but the underlying constraints of SMIC fabrication, nascent domestic HBM production, and the software gap with NVIDIA's CUDA ecosystem remain the same structural limits facing Cambricon, Biren, Moore Threads, and Enflame alike.