AI智能总结
为何科技巨头选择与博通联⼿以摆脱英伟达桎梏 Broadcom has establisheditself as the premier design partner for custom AI chips(ASICs), carving out a lucrative role at the center of the industry’s “shadow war” againstNvidia dominance. The company’s strategy is to enable major tech giants, includingGoogle, Apple, and OpenAI, to reduce their costly reliance on proprietary hardware bycreating their own bespoke silicon. We asked Deep Research about what lies ahead for thecompany. Here is the response. 博通已成为定制AI芯⽚(ASIC)领域的顶级设计合作伙伴,在这个⾏业反抗英伟达主导地位的"影⼦战争"中占据了利润丰厚的核⼼位置。该公司的战略是通过为科技巨头打造专属芯⽚,帮助包括⾕歌、苹果和OpenAI在内的⼤型科技企业降低对专利硬件的⾼成本依赖。我们向深度研究咨询了该公司的未来前景。以下是回应内容。 微信公众号 404K According to The Information’s reporting, Broadcom has carved out a distinct and highlylucrative role in the AI chip race by serving as the premier design partner for tech giantsseeking to create their own custom silicon. Rather than competing head-to-head withNvidia or AMD in the market for general-purpose GPUs, Broadcom collaborates with largetechnology companies to design and manufacture bespoke chips, known as application-specific integrated circuits (ASICs), tailored to their specific AI workloads. 据The Information报道,博通在AI芯⽚竞赛中开辟了⼀个独特且利润极⾼的⾓⾊定位——作为科技巨头打造定制芯⽚的⾸选设计合作伙伴。不同于英伟达或AMD在通⽤GPU市场的正⾯竞争,博通与⼤型科技公司合作,专门针对其特定AI⼯作负载需求来设计和制造定制芯⽚,即专⽤集成电路(ASIC)。 This strategy places Broadcom at the center of a critical industry trend: the push by thelargest tech companies—including Google, Apple, and OpenAI—to reduce their costlydependence on Nvidia.微信公众号 404K微信公众号 404K 这⼀战略使博通站在了关键⾏业趋势的中⼼:包括⾕歌、苹果和OpenAI在内的科技巨头正努⼒降低对英伟达的昂贵依赖。 The Custom Silicon Strategy 定制芯⽚战略 Broadcom’s core business in AI involves deep collaboration with its customers to translatetheir chip blueprints into manufacturable designs. This is a capital-intensive, high-expertise business that few can master. Broadcom’s CEO, Hock Tan, has noted that thebiggest cloud providers and tech platform companies areall moving toward creating theirown custom chips, a trend that directly benefits hisfirm.1 博通在⼈⼯智能领域的核⼼业务是与客户深度合作,将其芯⽚蓝图转化为可制造的⽅案。这是⼀个资⾦密集、⾼度专业化的领域,鲜有公司能掌握。博通⾸席执⾏官陈福阳指出,⼤型云服务提供商和科技平台公司都在转向定制⾃研芯⽚,这⼀趋势直接使他的公司受益。1 Broadcom’s AI business is substantial and growing at a blistering pace. In its fourthfiscalquarter, which ended on November 2, 2025, Broadcom reported a 74% jump in AI chiprevenue, contributing to a total revenue of $18 billion, up 28% overall. Looking ahead,CEO Hock Tan projected that AI chip revenue would double in the company’sfirstfiscalquarter, through January, with total revenue expected to rise 28% to $19.1 billion. Thecompany’s free cashflow also saw significant growth, reaching $7.466 billion in the fourthquarter,a 36% increase year-over-year.2微信公众号 404K微信公众号 404K 博通的⼈⼯智能业务体量庞⼤且增速惊⼈。在截⾄2025年11⽉2⽇的第四财季,博通AI芯⽚营收激增74%,推动总营收达到180亿美元,整体增长28%。公司⾸席执⾏官陈福阳预计,在截⾄次年1⽉的第⼀财季,AI芯⽚收⼊还将翻倍,总营收有望增长28%⾄191亿美元。该公司⾃由现⾦流也实现⼤幅增长,第四季度达到74.66亿美元,同⽐增长36%。2 Key Partnerships and Power Dynamics 核⼼合作伙伴与权⼒格局 Broadcom’s influence is best understood through its deep, albeit sometimes fraught,relationships with the biggest names in tech: 博通的影响⼒,从其与科技巨擘们深厚却时⽽紧张的合作关系中可见⼀斑: Google:For nearly a decade, Google has been Broadcom’sflagshipcustom AI chip customer. Broadcom has beenthe exclusive partnerhelping to design and produce Google’s Tensor Processing Units, orTPUs.This partnership has been a cornerstone of Google’s ability tocompete in AI by lowering its reliance on Nvidia. However, ourreporting indicates this relationship has faced challenges. Google hasclashed with Broadcom over pricing, prompting it to explore working3微信公众号 404K微信公众号 404K withotherfirms like MediaTekfor future TPUs to lower costs.Thishighlights a key vulnerability for Broadcom: its reliance on a smallnumber of massive, powerful customers who can and will seekalternatives.3 ⾕歌:近⼗年来,⾕歌⼀直是博通定制AI芯⽚的核⼼客户。博通作为独家合作伙伴,协助设计⽣产⾕歌的张量处理单元(TPU)。这项合作通过降低对英伟达的依赖,成为⾕歌AI竞争⼒的基⽯。但我们的调查显⽰,双⽅关系已出现裂痕。因定价争议,⾕歌正考虑与联发科等企业合作开发下⼀代TPU以降低成本。这暴露出博通的关键弱点:过度依赖少数具有议价权的巨头客户,这些客户随时可能转向替代⽅案。33微信公众号 404K微信公众号 404K Apple:In a significant move to control its own AI infrastructure, Appleis developing itsfirst server chip for AI, code-named “Baltra,” andisworking with Broadcomon the chip’s crucial networking technology.This project, expected to be ready for mass production by 2026,underscores Apple’s long-standing preference to avoid buying fromNvidia. Broadcom is providing a more limited scope of design serviceshere compared to its Google deal, focusing on its strength innetworking technology that links chips together to work in unison—acritical component for large-scale AI.44 苹果公司:为了掌控⾃⾝AI基础设施,苹果正采取⼀项重要举措——开发⾸款代号为"巴尔特拉"的AI服务器芯⽚,并与博通合作研发该芯⽚的关键⽹络技术。这项预计2026年实现量产的项⽬,突显出苹果⼀贯倾向于不从英伟达采购的长期策略。与⾕歌的合作协议相⽐,博通在此项⽬中提供的设计服务范围较为有限,主要专注于其连接多芯⽚协同⼯作的⽹络技术优势,这正是⼤规模AI系统的关键组件。44微信公众号 404K微信公众号 404K OpenAI:As part of its aggressive strategy to diversify its chip supplybeyond Nvidia,OpenAI has struck a deal with Broadcomto co-designits own AI chips.This move is part of a broader multi-vendorstrategy for OpenAI, which