來源Bloomberg Technology•較早收集於 32m
MediaTek 因 AI 晶片市場轉型股價上漲
💡MediaTek 轉向 AI 晶片，為邊緣 AI 開發者提供了新的低成本硬體選擇。
⚡ 30 秒速覽
有什麼變化
MediaTek 股價創下季度歷史新高
為什麼重要
MediaTek 進入 AI 晶片市場加劇了與 Nvidia 和 Qualcomm 的競爭，可能降低邊緣 AI 部署的成本。
下一步行動
監控 MediaTek Dimensity 晶片組規格，以評估其作為邊緣 AI 部署項目的高性價比替代方案。
誰應關注：Developers & AI Engineers
關鍵要點
- •MediaTek 股價創下季度歷史新高
- •從傳統技術轉向 AI 晶片組的策略轉型
- •投資人信心受 AI 基礎設施需求推動
🧠 深度解析
背景與延伸：來自公開資料，非原文內容。引用 16 個來源。
🔑 增強重點摘要
- •MediaTek is strategically expanding its focus beyond smartphone processors to include custom AI chips for cloud infrastructure and data center solutions, aiming to diversify revenue streams and reduce reliance on the highly competitive smartphone sector.
- •The company is actively collaborating with Google on the development of Tensor Processing Units (TPUs), positioning itself as a key partner in Google's custom AI chip initiatives.
- •MediaTek is exploring advanced semiconductor packaging technologies, with reports indicating that some next-generation custom AI ASIC and data center chip projects may utilize Intel's EMIB-T technology as an alternative to TSMC's CoWoS for potential cost and flexibility advantages.
- •Its Dimensity 9500 SoC features a 9th-generation NPU 990 processor, which includes a dual NPU design and an industry-first Compute-in-Memory (CIM) NPU, enabling up to 100 TOPS for efficient, always-on generative AI models.
- •MediaTek has partnered with NVIDIA to develop the RTX Spark processor class for Windows 11 PCs, combining MediaTek's SoC and connectivity expertise with NVIDIA's AI platform to target thin laptops and small form factor desktops for agentic AI, content creation, and gaming.
📊 競品分析▸ Show
| Competitor | Key AI Chip Offerings | Primary Market Focus | Noteworthy AI Features / Benchmarks |
|---|---|---|---|
| MediaTek | Dimensity series (e.g., 9500, 9400), Genio (IoT), Pentonic (Smart TVs), Dimensity Auto, Custom AI ASICs | Smartphones (premium/flagship), IoT, Smart TVs, Automotive, AI PCs, Data Centers | Dimensity 9500 NPU 990: up to 100 TOPS, dual NPU, Compute-in-Memory (CIM) NPU for on-device LLM training, 27x greater power efficiency than typical CPU for AI tasks. |
| Qualcomm | Snapdragon series (e.g., Snapdragon 8 Elite Gen 6, Snapdragon X2 Elite) | Smartphones (premium/flagship), AI PCs, Automotive | Custom Oryon CPU cores; Snapdragon X2 Elite for AI PCs shows strong single-core performance (around 4,000 Geekbench 6). |
| Apple | A-series (e.g., A19 Bionic), M-series (e.g., M4 Pro) | iPhones, iPads, Macs | Vertically integrated design, strong raw chip performance (especially single-core speed), highly optimized for its ecosystem. |
| NVIDIA | GPUs (e.g., MI300, Blackwell RTX cores), RTX Spark (with MediaTek) | Data Centers (training/inference), AI PCs | Dominant in data center AI with CUDA software ecosystem; RTX Spark for AI PCs features 6,144 Blackwell RTX cores (equivalent to RTX 5070 GPU). |
| AMD | MI300 GPU | Data Centers (training/inference), Edge devices | MI300 GPU for data centers; aims to improve software ecosystem to challenge NVIDIA. |
| Tensor Processing Units (TPUs) | Cloud AI workloads (internal & partners) | Custom-designed ASICs for specific AI computing workloads, offers a cheaper alternative for internal AI workloads. |
🛠️ 技術深入
- MediaTek NeuroPilot SDK: A comprehensive software ecosystem supporting all MediaTek AI-capable hardware, enabling a 'write once, apply everywhere' strategy across smartphones, automotive, smart home, and IoT product lines. It includes a complete compiler, profiler, and application libraries for Android and Linux OS.
- MediaTek NPU (AI Processing Unit): Integrated into SoCs alongside CPUs and GPUs, providing heterogeneous computing capabilities. The NPU architecture is highly scalable and can include MDLA (Deep Learning Accelerator) and MVPU (Vision Processing Unit) cores in varying quantities.
- 7th Generation NPU (2023): Specifically engineered to accelerate generative AI based on transformer models. Its design allows for flexible scaling of performance with computation units, power usage, memory bandwidth, and memory capacity. It incorporates a hardware-based multicore scheduler and a dedicated DMA engine for deep layer fusion and data compression.
- Dimensity 9500 NPU 990 (9th-Gen): Achieves up to 100 TOPS (Tera Operations Per Second) and features a dual NPU design. This includes a powerful NPU for heavy workloads and an industry-first, super-efficient Compute-in-Memory (CIM) NPU optimized for light, always-on AI models. It supports BitNet 1.5–8-bit models on hardware, facilitating on-device LLM LoRA training and processing of up to 128K-token-long texts, and utilizes second-generation large-model hardware compression technology.
- Power Efficiency: MediaTek's Deep Learning Accelerator (DLA) demonstrates significant power efficiency, offering 27 times greater efficiency compared to a typical CPU and 15 times greater efficiency compared to a typical GPU when running five common neural networks.
- Advanced Packaging Exploration: MediaTek is reportedly considering Intel's Embedded Multi-die Interconnect Bridge (EMIB-T) technology for future custom AI ASIC and data center chip projects. This technology aims to connect multiple chiplets without a large silicon interposer, potentially offering lower manufacturing costs and greater packaging flexibility compared to TSMC's CoWoS.
🔮 前景展望基於引用來源的 AI 分析
MediaTek will significantly increase its market share in the data center AI chip segment by 2027.
Its strategic shift towards custom AI ASICs, partnerships with Google for TPUs, and exploration of advanced packaging like Intel's EMIB-T indicate a strong push into this high-growth area.
MediaTek's 'Agentic AI' strategy will lead to a new wave of highly personalized and proactive on-device AI experiences in smartphones and IoT devices.
The Dimensity 9500's advanced NPU with Compute-in-Memory and support for on-device LLM training is designed to enable AI that adapts and learns without constant cloud intervention.
MediaTek will become a significant player in the AI PC market by 2027, challenging Qualcomm and Apple.
Its partnership with NVIDIA on the RTX Spark platform for Windows 11 PCs, combining its SoC expertise with NVIDIA's GPU, positions it to capture a share of the growing AI PC silicon market.
⏳ 時間線
1997
MediaTek founded as a spin-off from UMC, initially focusing on optical drive and DVD player chipsets.
2019
Launched 'Dimensity' brand, entering high-performance computing and 5G smartphone market.
2021
Became the world's largest smartphone SoC vendor by volume.
2023
Launched 7th generation NPU for generative AI and began automotive pilot revenues with NVIDIA collaboration.
2025-09
Introduced Dimensity 9500 with 9th-gen NPU 990, achieving 100 TOPS and featuring Compute-in-Memory NPU.
2026-01
Stock surged due to increasing awareness of its collaboration with Google on Tensor Processing Units (TPUs).
📎 來源 (16)
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： Bloomberg Technology ↗
每週電子報
每週一封，可隨時退訂。