
近日,英伟达与联发科联合官宣深化合作。英伟达斥资35亿美元认购联发科可转债,这是其在美境外最大的一笔直接投资。双方合作延伸至AI基础设施、本地AI计算、智能汽车三大板块,而核心落点在于:联发科的定制算力芯片业务将接入英伟达NVLink Fusion平台。当云厂商与大模型企业开发定制XPU时,芯片设计交由联发科完成,互联、内存、封装、机架等环节则由英伟达联合支撑,产出的芯片可无缝接入英伟达的AI基础设施体系。
分析观点认为,当前云厂商自研ASIC/XPU的浪潮愈演愈烈,定制芯片的设计主导权正被博通把持。英伟达通过入股联发科,采取了新的策略——不再阻止客户自研芯片,而是让「自研」本身发生在英伟达的生态之内。同时,联发科的SoC能力也补齐了英伟达在消费PC端对抗AMD、在车载端对抗高通的端侧短板。
放大到英伟达自2023年以来的整个投资体系来看,这早已不是单纯的「芯片公司投资」,而是一套围绕「算力主权」的系统性生态工程:投资OpenAI、Anthropic等大模型公司锁定下游需求;投资联发科、康宁、Marvell等中游供应链企业锁定互联与封装标准;投资CoreWeave、SK集团锁定算力分发。其本质,是用资本把「市场份额」升级为「生态锁定」。
本文来源:https://www.163.com/dy/article/L6P9L3SB0531B3OJ.html
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
NVIDIA recently announced a deepened partnership with MediaTek, investing $3.5 billion in convertible bonds — its largest direct investment outside the U.S. The collaboration spans AI infrastructure, local AI computing, and smart cars, with the core being MediaTek's custom compute chip business integrating into NVIDIA's NVLink Fusion platform. When cloud providers or model companies develop custom XPUs, MediaTek handles chip design while NVIDIA supports interconnects, memory, packaging, and racks — ensuring seamless integration into NVIDIA's AI ecosystem.
Analysts note that the surge in custom ASIC/XPU development by cloud providers has been dominated by Broadcom. NVIDIA’s investment in MediaTek represents a strategic shift: rather than blocking client customization, it allows “self-developed” chips within its own ecosystem. This also leverages MediaTek’s SoC expertise to counter AMD in consumer PCs and Qualcomm in automotive, strengthening NVIDIA’s end-side capabilities.
Since 2023, NVIDIA’s investments have evolved beyond mere “chip company funding” into a systematic ecosystem engineered around “compute sovereignty.” By investing in OpenAI, Anthropic, and others, it secures downstream demand; in MediaTek, Corning, Marvell, it locks in interconnect and packaging standards; and in CoreWeave, SK Group, it controls compute distribution. The goal is to transform market share into full ecosystem lock-in.