來源SCMP Technology•較早收集於 18m
MetaX 擬赴港上市，加速 AI 晶片研發與擴張
💡中國 Nvidia 挑戰者尋求全球資本，以擴大 AI 基礎設施所需的 GPU 生產規模。
⚡ 30 秒速覽
有什麼變化
MetaX 計劃發行不超過擴大後股本 5% 的 H 股。
為什麼重要
MetaX 的舉動顯示中國半導體企業正積極尋求進入全球資本市場，以維持在 AI 硬體領域與 Nvidia 競爭所需的高額研發投入。
下一步行動
持續關注 MetaX 的硬體基準測試與軟體堆疊相容性，評估其作為受限市場中 AI 基礎設施替代方案的可能性。
誰應關注：Founders & Product Leaders
關鍵要點
- •MetaX 計劃發行不超過擴大後股本 5% 的 H 股。
- •募集資金將用於支持國際業務擴張並強化公司治理。
- •該公司目標是挑戰 Nvidia 在 GPU 市場的統治地位。
- •此舉緊隨其近期在上海科創板的上市步伐。
🧠 深度解析
背景與延伸：來自公開資料，非原文內容。引用 17 個來源。
🔑 增強重點摘要
- •MetaX was founded in September 2020 by former AMD employees, bringing significant industry experience to its GPU development efforts.
- •The company successfully debuted on the Shanghai Stock Exchange's STAR Market on December 17, 2025, raising RMB 4.2 billion (USD 596 million) with shares surging up to 693% on its first trading day.
- •MetaX has developed a comprehensive GPU product portfolio, including MXN for AI inference, MXC for AI training and general computing, MXG for graphics rendering, and the recently launched X-Series for AI-for-Science applications.
- •Despite reporting a net loss of 1.41 billion yuan in 2024, MetaX's revenue more than doubled in 2025 to 1.6 billion yuan, driven by strong domestic demand for its GPUs amidst US export controls.
- •To comply with US restrictions, MetaX submitted downgraded chip designs, such as the C280, to TSMC in late 2023, indicating the impact of geopolitical factors on its supply chain and product strategy.
📊 競品分析▸ Show
| Feature/Metric | MetaX C500 (AI Training) | MetaX N100 (AI Inference) | Nvidia A100 | Nvidia A30 |
|---|---|---|---|---|
| FP32 Performance | 15 TFLOPS | N/A | ~20 TFLOPS | N/A |
| FP16 Performance | N/A | 80 TFLOPS | N/A | 165 TFLOPS (with sparsity) |
| INT8 Performance | N/A | 160 TOPS | 1248 TOPS (with sparsity) | 330 TOPS (with sparsity) |
| Process Node | 7nm | Undisclosed (advanced packaging with HBM2E) | 7nm | 7nm |
| Software Stack | MXMACA 2.0 (CUDA-compatible) | MXMACA (compatible with various frameworks) | CUDA | CUDA |
| Video Capabilities | N/A | 128-ch encoding, 96-ch decoding for 8K (HEVC, H.264, AV1, AVS2) | N/A | N/A |
| Market Share (China AI chip, 2024) | ~1% | ~1% | Dominant | Dominant |
🛠️ 技術深入
- GPU Architecture: MetaX GPUs are built on a 7nm process and GPGPU architecture.
- Software Stack: The company utilizes its self-developed MXMACA software stack, designed for hardware-software co-design and compatible with major international GPU ecosystems and scientific computing frameworks.
- C500 (AI Training): This GPU delivers 15 TFLOPS FP32 performance and is CUDA-compatible. It supports multi-GPU interconnects for scalable performance.
- N100 (AI Inference): Features a heterogeneous N100 core with HBM2E memory, offering 160 TOPS INT8 and 80 TFLOPS FP16. It supports 128-channel encoding and 96-channel decoding for 8K video in formats such as HEVC, H.264, AV1, and AVS2.
- X206 (AI-for-Science): The first product in the X-Series, it supports full-precision hybrid computing, multiple compute precisions, expanded memory capacity and bandwidth, and multi-card interconnect technology for cluster communication. It is available in modular and server formats.
- US Restrictions Impact: In late 2023, MetaX developed and submitted downgraded chip designs, including the C280, to TSMC to comply with US export restrictions, having run out of stock of the C500.
🔮 前景展望基於引用來源的 AI 分析
MetaX's Hong Kong IPO will significantly boost its international expansion and R&D capabilities.
The capital raised from the H-share offering is explicitly intended to fund international growth, strengthen corporate governance, and support next-generation GPU development and commercialization efforts.
US export restrictions will continue to shape MetaX's product development and market strategy.
MetaX has already adapted its chip designs (e.g., C280) to comply with US restrictions, indicating an ongoing need to navigate geopolitical factors in its technology roadmap and supply chain.
MetaX will intensify competition in the domestic Chinese AI chip market, potentially reducing reliance on foreign suppliers.
The company's rapid revenue growth and focus on domestic substitution, coupled with government support and its comprehensive product portfolio, position it to capture a larger share of the Chinese market.
⏳ 時間線
2020-09
MetaX founded by former AMD employees.
2021-08
Raised RMB 1 billion ($155 million) in Series A funding.
2023-06
Successfully tested its first AI training GPU, C500, and launched its first AI inference GPU accelerator, N100.
2023-Late
Submitted downgraded chip designs (C280) to TSMC to comply with U.S. restrictions.
2025-12-17
Listed on the Shanghai Stock Exchange's STAR Market, raising RMB 4.2 billion (USD 596 million).
2026-01
Launched X-Series GPUs for AI-for-Science, with the X206 beginning mass production.
📎 來源 (17)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： SCMP Technology ↗
每週電子報
每週一封，可隨時退訂。