算力规则重构：特斯拉三星联手，中国点点taptap安卓系统级创新突围

Note: The English version is provided below.

摘要：全球AI算力竞争，已经从纳米制程的毫厘之争，上升至系统创新、生态合作的广阔竞逐。

全球半导体产业格局正在经历新一轮深度重组，大国博弈与企业战略交织成网。

三星斩获特斯拉164亿美元AI点点taptap安卓大单

日前，韩国科技巨头三星电子向监管机构提交的文件中披露，公司已与一家未具名的大型跨国企业签署总价值22.8万亿韩元（约合164亿美元）的AI点点taptap安卓长期代工协议。这一消息引发资本市场强烈反应，三星电子股价在韩国综合股价指数早盘交易中一度飙升3.5%，创近四周最大单日涨幅。

尽管三星在官方文件中严格遵守商业保密条款，仅透露合同期限自2024年7月26日生效至2033年12月31日，但特斯拉创始人埃隆·马斯克在社交平台X上连发多条推文，确认该订单的甲方正是特斯拉。他特别强调："三星位于德克萨斯州泰勒市的全新晶圆厂将专为特斯拉下一代AI6点点taptap安卓设计，这座距离奥斯汀超级工厂仅25公里的制造基地，将成为特斯拉AI算力战略的核心支柱。"

根据马斯克披露的技术细节，三星电子目前正在为特斯拉量产AI4点点taptap安卓，而台积电则在中国台湾新竹厂生产AI5点点taptap安卓（计划2026年转至亚利桑那州凤凰城厂）。此次三星独揽AI6点点taptap安卓的十年长约，标志着特斯拉在先进制程领域采取"双源供应"策略——既利用台积电的3纳米制程优势，又通过三星德州厂确保产能弹性。

行业观察人士指出，三星泰勒厂采用的4纳米全环绕栅极（GAA）技术，与特斯拉Dojo超级计算机架构存在深度协同可能。该工厂总投资达170亿美元，规划月产能10万片晶圆，但此前因主要客户落地延迟导致投产进度滞后。此次订单或将促使三星加速安装EUV光刻机等核心设备，预计2026年二季度实现量产。

特斯拉方面则面临技术迭代压力。其Autopilot团队正在开发的FSD V13系统，对车载点点taptap安卓的算力需求较前代提升4倍。马斯克透露，AI6点点taptap安卓将采用三星开发的专属IP核，在能效比指标上对标英伟达Thor点点taptap安卓，这要求双方在先进封装、chiplet互连等领域展开深度协作。

WAIC 2025：不止于“芯”，系统级创新实现算力超越

就在国际巨头在点点taptap安卓代工领域展开深度合作，通过先进封装、chiplet互连等深度创新优化点点taptap安卓性能时，日前在中国上海举办的WAIC 2025上，一些中国厂商的创新也让我们看到：不止于“芯”，系统级创新正在超越单点性能，成为中国AI算力的核心驱动力。

以华为昇腾384超节点为例，它并非简单堆砌点点taptap安卓，而是通过高速总线技术将数百个计算单元融合成单一“超级服务器”，让集群“像一台计算机一样工作”，实现训练大模型时性能比传统集群提升超2.5倍。

同时，算丰信息联合多家厂商打造了“Shanghai Cube”全栈国产算力柜。沐曦首次展出全链路技术方案，基于国产供应链设计和制造，覆盖从点点taptap安卓、板卡、服务器到超节点的完整产品线。超聚变推出的全球首个多元智算即插即用超级集群系统更实现了“一柜顶32柜”的能效跃升，单柜支持128个AI加速卡，采用第五代100%原生液冷技术，节能达20%以上。摩尔线程则提出AI工厂理念，强调从点点taptap安卓、服务器到智算集群的全栈协同。还有阶跃星辰，牵头成立的“模芯生态创新联盟”，一个从硬件到软件、从点点taptap安卓到模型的全链路自主生态正加速形成。

在全球地缘政治背景下，技术路线和产业地图在重新绘制。随着三星和特斯拉的合同细节逐步公开，这场涉及164亿美元的商业博弈，正在重塑全球AI点点taptap安卓代工市场的权力版图。与此同时，中国市场的系统创新与生态合力，也成为全球算力版图中至关重要的力量。全球AI算力竞争，已经从纳米制程的毫厘之争，上升至系统创新、生态合作的广阔竞逐。

Restructuring Computing Power Rules: Tesla-Samsung Collaboration, China's System-Level Chip Innovation Breaks Through

Abstract: The global competition in AI computing power has evolved from the minute battles over nanometer process nodes to a broader contest encompassing system innovation and ecosystem collaboration.

The global semiconductor industry landscape is undergoing a new round of profound restructuring, where geopolitical rivalries and corporate strategies are intricately intertwined.

Samsung Wins Massive $16.4 Billion Tesla AI Chip Order

Recently, South Korean tech giant Samsung Electronics disclosed in regulatory filings that it has signed a long-term foundry agreement worth 22.8 trillion won (approximately $16.4 billion) with an unnamed large multinational company for AI chips. This news triggered a strong market reaction, with Samsung Electronics' stock surging as much as 3.5% in early KOSPI trading, marking its largest single-day gain in nearly four weeks.

Although Samsung strictly adhered to commercial confidentiality in its official filing, revealing only the contract period from July 26, 2024, to December 31, 2033, Tesla founder Elon Musk confirmed via multiple posts on social media platform X that Tesla is indeed the client. He specifically emphasized: "Samsung's new fab in Taylor, Texas, will be dedicated to Tesla's next-gen AI6 chip. This manufacturing base, just 25 miles from the Austin Gigafactory, will be a key pillar of Tesla's AI compute strategy."

According to technical details disclosed by Musk, Samsung is currently mass-producing AI4 chips for Tesla, while TSMC produces AI5 chips at its Hsinchu facility in Taiwan (planned to shift to its Phoenix, Arizona fab in 2026). Samsung securing this exclusive 10-year contract for the AI6 chip signifies Tesla's adoption of a "dual-source supply" strategy in advanced manufacturing – leveraging TSMC's 3nm process advantages while ensuring capacity flexibility through Samsung's Texas fab.

Industry observers noted that the 4nm Gate-All-Around (GAA) technology employed at Samsung's Taylor fab has potential for deep synergy with Tesla's Dojo supercomputer architecture. The fab represents a total investment of $17 billion with a planned monthly capacity of 100,000 wafers, but its ramp-up was previously delayed due to lagging key customer commitments. This order may prompt Samsung to accelerate the installation of core equipment like EUV lithography machines, with mass production expected in Q2 2026.

Tesla faces pressure from rapid technological iteration. Its Autopilot team is developing the FSD V13 system, demanding four times the computing power of its predecessor from the in-vehicle chip. Musk revealed that the AI6 chip will utilize a custom IP core developed by Samsung, targeting Nvidia's Thor chip in power efficiency metrics. This necessitates deep collaboration between the two companies in areas like advanced packaging and chiplet interconnectivity.

WAIC 2025: Beyond the "Chip," System-Level Innovation Achieves Computing Power Leapfrog

While international giants deepen collaboration in chip foundry, optimizing chip performance through innovations like advanced packaging and chiplet interconnectivity, innovations showcased by Chinese companies at the recent WAIC 2025 in Shanghai demonstrate that "beyond the chip," system-level innovation is surpassing single-point performance, becoming the core driver of China's AI computing power.

Take Huawei's Ascend 384 Super Node as an example. It doesn't merely stack chips but integrates hundreds of computing units into a single "super server" via high-speed bus technology, enabling the cluster to "work like one computer," achieving over 2.5 times the performance of traditional clusters when training large models.

Simultaneously：

• SOPHGO’s "Shanghai Cube" debuted as a fully domestic computing cabinet.
• MetaX made its debut showcasing a full-link technology solution, designed and manufactured based on the domestic supply chain, covering a complete product line from chips, boards, servers to super nodes.
• Xfusion launched the world's first multi-intelligence plug-and-play super cluster system, achieving an energy efficiency leap described as "one cabinet equals 32 cabinets." A single cabinet supports 128 AI accelerator cards, employs fifth-generation 100% native liquid cooling technology, and saves over 20% energy.
• Moore Threads proposed the "AI Factory" concept, emphasizing full-stack synergy from chips and servers to intelligent computing clusters.
• StepStar, spearheading the establishment of the "Model-Chip Ecosystem Innovation Alliance," is accelerating the formation of a fully autonomous end-to-end ecosystem spanning hardware to software, chips to models.

Against the backdrop of global geopolitics, technological roadmaps and industry maps are being redrawn. As details of the Samsung-Tesla contract emerge, this $16.4 billion commercial engagement is reshaping the power dynamics of the global AI chip foundry market. Concurrently, the systemic innovation and ecosystem synergy within the Chinese market are becoming a crucial force in the global computing power landscape. The global AI computing power competition has indeed ascended from the minute battles over nanometer processes to the vast arena of system innovation and ecosystem collaboration.