来源：金色note 在2026年日本NEPCON展会上，英特尔正式发布了一款突破性的原型产品，有望重新定义人工智能硬件 的未来：一块尺寸为78mm x 77mm的巨型玻璃芯基板，并集成了英特尔的嵌入式多芯片互连桥 （EMIB）技术。这一突破标志着芯片封装技术的重大变革，正朝着更大、更强大、更复杂的设计方向 发展，以满足下一代人工智能的需求。 作为背景，EMIB 就像一条嵌入基板内的高速通道。它专门用于连接相邻的芯片，使它们之间的数据传 输如同它们属于单个单片芯片一样。 为什么人工智能芯片必须从塑料材质转向玻璃材质？ 随着人工智能芯片尺寸不断增大，不断挑战光刻技术的极限（光刻胶尺寸极限），传统的有机衬底正面 临严峻的物理挑战。有机材料在高温下容易因热胀冷缩而发生形变，导致芯片与衬底之间的连接不良。 相比之下，玻璃的热膨胀系数（CTE）与硅非常接近，能够确保芯片在高温下具有卓越的尺寸稳定性， 从而提供了一种解决方案。 此外，与有机基底相比，玻璃超光滑的表面可以蚀刻出更精细的电路图案。这使其成为支撑下一代人工 智能加速器强大计算能力和复杂布线的理想基底。 在规格方面，英特尔的这款原型封装尺寸巨大，达到 78mm x ...

2月5日消息，香港科技大学与英特尔宣布成立"香港科技大学-英特尔联合实验室"。该实验室将开展为 期三年的研究计划，重点探索高能效近存计算架构，以应对 人工智能应用在性能与能效方面的挑战。 通过软硬件协同设计创新，双方旨在为智能设备与可持续人工智能系统的未来发展提供技术参考。 ...

在存储持续短缺之际，英特尔计划重返存储市场，已和软银子公司Saimemory签署新合作协议，将推动 开发和生产被称为Z-Angle Memory（ZAM）的新型垂直堆叠存储，与用于最新AI资料中心的高频宽存 储（HBM）竞争。 ZAM提供的容量大于HBM，频宽也较大，耗电量则较低。 据软银声明，Saimemory和英特尔目标是在2028年3月31日止的年度打造出原型，2029年度商业化。 存储价格飙涨已促使数个消费者电子品牌涨价，或调整产品计划。任天堂公司社长古川俊太郎表示，如 果存储价格高涨持续时间比预期长，该公司明年度起获利能力可能承压；任天堂股价4日应声重挫近 11%。 英特尔执行长陈立武表示，由于人工智能（AI）热潮瓜分存储供给，电脑业存储芯片短缺可能要到 2028年才会舒缓，也就是至少还会持续两年。 思科3日举行AI峰会，陈立武转述两家大型存储业者的看法。他说：「据我所知，（存储短缺）完全没 有改善…2028年前都不会缓解。」 AI基础设施大规模扩张，已为存储芯片计划需求增添柴火，分给传统电脑与智慧手机的供给随之减 少。这个现象已导致存储短缺并带动价格大涨，可能降低消费者购买电脑和智慧手机的欲望。陈 ...

我真是做梦都没想到，这辈子还能看到英特尔主动跑去当小弟的事情。 就在前几天的财报会议上，谈到公司持续简化服务器产品路线图的举措时，英特尔新掌门陈立武透露："我们在与英伟达紧密合作，打造一款完全集成其 NVLink技术的定制版至强处理器，为AI主机节点带来一流的x86性能。" 消息一出，让整个科技圈都炸锅了：英特尔要和英伟达联手，开发一款专门为AI设计的定制版至强Xeon处理器？ （图源：wccftech） 掐指一算，这也是英伟达去年投资50亿美元，收购英特尔的部分股份后，双方在硬件上的第一次实质性合作。 这事儿要是放在五年前，甚至三年前，绝对是天方夜谭。 但现在，陈立武这波操作，可以说英特尔已经不准备跟英伟达硬刚谁才是AI老大了，而是退了一步说：老黄，既然大家都要买你的显卡，那不如我专门做 个CPU来配合你，咱们搭伙过日子吧。 要知道，当年那颗E3-1230 v2可是被捧上神坛的垃圾佬神U，花着不到i3的钱就能享受i7的性能。再往后，那些从国外服务器机房淘汰下来的E5处理器漂洋 过海，配合着华强北魔改的X79、X99主板，让多少预算吃紧的学生党和图吧老哥望眼欲穿。 而这次陈立武狠心找英伟达借力，肯定是希望能够 ...

"据我所知，目前没有任何缓解措施，"陈立武透露，自己近期与两位业内关键人士进行了深入交流，对 方均明确表示，"2028年之前都不会有任何实质性缓解"。 这一严峻预测的背后，是人工智能（AI）基础设施的迅猛扩张对高性能内存芯片的空前需求。而当前 内存市场的供需失衡已引发价格上涨，可能削弱消费者对电脑和手机等终端设备的购买意愿，进而影响 整个消费电子生态。 来源：环球网 【环球网科技综合报道】2月4日消息，据彭博社报道称，英特尔首席执行官陈立武（Lip-Bu Tan）近日 在一场行业会议上表示，当前全球计算机行业面临的内存芯片短缺问题短期内难以缓解，预计将持续至 2028年。 陈立武特别提到，作为AI处理器供应商，英伟达即将推出的Rubin平台及后续产品将进一步加剧内存消 耗。"人工智能将会消耗大量内存。"他强调。 资料显示，作为全球最大的PC处理器制造商，英特尔高度依赖稳定的内存供应链以确保其CPU产品的 性能与交付能力。全球主要内存芯片制造商包括三星电子、SK海力士和美光科技，这些企业正全力扩 产以应对AI驱动的需求激增。然而，从产能建设到实际产出仍需时间，短期内难以填补缺口。（青 云） ...

Core Viewpoint - Intel is entering the GPU market, traditionally dominated by Nvidia, under the leadership of new CEO Lip-Bu Tan, aiming to target the lucrative AI chip business in data centers [1] Group 1: GPU Strategy - Intel plans to manufacture GPUs specifically for data center applications, focusing on customer collaboration to define product needs rather than pre-determined product lines [2][4] - The GPU initiative is seen as a strategic expansion for Intel, moving beyond its traditional CPU dominance to address specific workloads like gaming and AI model training [2] Group 2: Key Personnel - Eric Demmers has been appointed as the Chief GPU Architect, bringing over 13 years of experience from Qualcomm, which is expected to strengthen Intel's GPU project [3] Group 3: Customer-Centric Approach - The GPU project is still in the early stages, with Intel prioritizing customer needs to shape its product strategy and production capacity [4] - Lip-Bu Tan emphasized the importance of understanding customer demand to effectively plan production and capacity [4] Group 4: Foundry Business Development - Intel Foundry is also progressing, with several clients showing interest in the 14A manufacturing process, with mass production expected to ramp up later this year [5] - The dual focus on GPU development and foundry services aims to capture incremental demand in data centers while attracting foundry clients [5]

银河证券表示，模拟芯片行业供需关系或将迎来变化。数字芯片方面，以AI算力、国产CPU为代表的先 进设计领域表现强势，国产数字芯片自主可控仍将是长期方向。 （文章来源：21世纪经济报道） 科创芯片设计ETF天弘（589070）具备20%的涨跌幅空间，该产品紧密跟踪上证科创板芯片设计主题指 数，指数聚焦于芯片设计核心环节，相关行业含量接近95%，行业集中度较高、"含芯量"突出，其成分 股涵盖50家科创板芯片领域龙头企业。 消息面上，据第一财经，英特尔首席执行官陈立武周二在思科人工智能峰会上表示，公司已任命新的首 席架构师全面推进图形处理器（GPU）研发，以把握人工智能数据中心需求快速增长带来的机遇。他表 示，人工智能算力需求激增，正显著推高对GPU和存储芯片的依赖，而当前席卷行业的存储芯片短缺已 成为AI发展面临的"最大挑战"，相关供需失衡预计要到2028年才可能缓解。 2月4日，三大指数涨跌不一，芯片方向走弱。上证科创板芯片设计主题指数(950162.SH)下跌2.91%，该 指数成分股中，国芯科技上涨超3%，盛科通信与灿芯股份上涨超1%，睿创微纳上涨近1%。 相关ETF方面，科创芯片设计ETF天弘（58907 ...

Group 1 - Intel has appointed a new Chief Architect to lead its GPU business, amidst a significant shortage of memory chips expected to last at least two more years [1][2] - The demand for GPUs, essential for large language models, has surged as companies invest in AI infrastructure and data centers [1] - Intel has lagged behind major semiconductor firms like TSMC, but has seen a stock price rebound due to significant investments from the U.S. government, SoftBank, and Nvidia [1] Group 2 - CEO Chen stated that there are no signs of relief from the ongoing memory chip shortage, with expectations that it will not ease until at least 2028 [2] - UBS analysts noted a structural differentiation in the global memory industry, predicting meaningful supply relief will not occur until around 2028 [2] - The expansion of AI infrastructure is driving up demand for memory chips, which is constraining supply for traditional computers and smartphones, potentially affecting consumer purchasing behavior [2] Group 3 - Nvidia's latest Rubin platform and next-generation products are expected to further increase demand for memory chips, with AI anticipated to "consume" vast amounts of storage resources [3]

Summary of Conference Call on Global Semiconductors Industry Overview - The focus is on the semiconductor industry, specifically the competition between Intel and TSMC regarding advanced packaging technologies for AI chips, particularly the Embedded Multi-die Interconnect Bridge-T (EMIB-T) technology offered by Intel as an alternative to TSMC's CoWoS packaging method [2][12]. Core Points and Arguments 1. **EMIB-T Technology**: - EMIB-T is an enhanced version of Intel's existing EMIB technology, designed to support larger reticle sizes and provide a more cost-effective solution for AI chip packaging compared to TSMC's CoWoS [3][34]. - Intel claims EMIB-T can support reticle sizes of 8-12x by 2026-2027, compared to TSMC's current capabilities of 3.3x and future plans for 5.5x and 9.5x [3][34]. 2. **Financial Impact**: - If 1 million chips shift from CoWoS to EMIB-T, TSMC could face a revenue loss of approximately $1 billion, which is about 5-10% of its advanced packaging revenue in 2027, but only 0.5% of its total revenue [4][52]. - Conversely, Intel could see a revenue increase of high triple-digit millions, representing 1-2% of its revenue [4][51]. 3. **Market Positioning**: - Ibiden is highlighted as a strong player in the EMIB-T market, with expectations of increased revenue and margins due to the complexity of EMIB-T substrates [5][54]. - The substrate value for EMIB-T is projected to rise to approximately $300, significantly higher than previous generations [5][49]. 4. **Geopolitical Considerations**: - Intel's existing advanced packaging capacity in the U.S. provides a competitive edge, especially as TSMC plans to build new packaging fabs in Arizona, which may not be operational until 2028 [13][35]. 5. **Challenges for EMIB-T**: - The main challenges for EMIB-T include a lack of proven track record and potential lower production yields due to the complexity of embedding silicon bridges in the substrate [3][34]. Additional Important Content - **Customer Engagement**: Intel has indicated potential early customer engagements worth "north of a billion dollars" each, although this remains uncertain [4][51]. - **Future Developments**: Intel is also exploring advancements in 3.5D packaging, which combines EMIB with Foveros Direct technology, aiming to compete more effectively with TSMC's offerings [42][43]. - **Investment Ratings**: - Ibiden, TSMC, and MediaTek are rated as "Outperform," while Intel is rated as "Market-Perform" [8][9][10][11]. Conclusion - The semiconductor industry is witnessing a significant shift with Intel's EMIB-T technology potentially challenging TSMC's dominance in advanced packaging for AI chips. The financial implications for both companies are substantial, with Ibiden positioned to benefit from this transition. However, challenges remain regarding production yields and the need for proven technology.

他表示，人工智能算力需求激增，正显著推高对GPU和存储芯片的依赖，而当前席卷行业的存储芯片短 缺已成为AI发展面临的"最大挑战"，相关供需失衡预计要到2028年才可能缓解。 编辑：七三 陈立武还透露，多家客户正与英特尔晶圆代工业务展开深入接洽，兴趣集中在14A制程技术，相关量产 有望在今年晚些时候加速。 英特尔首席执行官陈立武周二在思科人工智能峰会上表示，公司已任命新的首席架构师全面推进图形处 理器（GPU）研发，以把握人工智能数据中心需求快速增长带来的机遇。 ...