AI Infrastructure Becomes a Full-Stack Commitment AI基础设施转向全栈投入

Nvidia and SK Group link data-center expansion to memory supply, while AMD, Intel, and a growing field of challengers push their roadmaps from individual chips toward complete systems. Capital, packaging, foundry execution, and software are increasingly moving as one stack. Nvidia与SK Group把数据中心扩张同存储供应绑定，AMD、Intel及一批新势力则把路线图从单颗芯片推进到完整系统。资本、封装、晶圆代工执行力与软件正日益作为同一套技术栈协同推进。

AI & Accelerators AI与加速器

Nvidia and SK Group Unveil $500 Billion-Plus AI Infrastructure Initiative Nvidia与SK Group公布逾5000亿美元AI基础设施计划

Reuters · 14h ago Reuters · 14小时前

Nvidia and SK Group unveiled an AI data-center initiative described as exceeding $500 billion alongside a memory partnership. The combination ties accelerator deployment more directly to SK's memory and infrastructure capacity, highlighting that compute availability now depends on coordinated investment across the stack. Nvidia与SK Group公布一项规模据称超过5000亿美元的AI数据中心计划，并同步推进存储合作。该组合把加速器部署与SK的存储及基础设施产能更直接地绑定，表明算力供给越来越依赖整条技术栈的协同投资。

AMD Frames Its AI Roadmap at Rack Scale AMD把AI路线图推向机架级

The Next Platform · 20h ago The Next Platform · 20小时前

AMD is pursuing the AI infrastructure market through rack-scale system roadmaps rather than selling accelerators as isolated components. The strategy raises the importance of networking, memory, power, cooling, and software integration, the layers that determine whether GPU specifications translate into deployable cluster performance. AMD正通过机架级系统路线图争夺AI基础设施市场，而不再把加速器作为孤立部件销售。该策略提升了网络、存储、供电、散热和软件集成的重要性，因为这些层面决定GPU规格能否转化为可部署的集群性能。

Foundry & Manufacturing 晶圆代工与制造

Intel Foundry Execution Improves, but Outside Revenue Remains Small Intel Foundry执行力改善，但外部收入仍偏小

EE Times · 16h ago EE Times · 16小时前

Intel's fab execution is improving, but its foundry business generated only $293 million in external revenue. That figure leaves customer adoption, rather than internal process progress alone, as the central test of whether Intel can establish a credible alternative to TSMC. Intel晶圆厂的执行情况正在改善，但其代工业务外部收入仅为2.93亿美元。这意味着客户采用度，而非仅仅内部制程进展，仍是检验Intel能否成为TSMC可信替代者的核心指标。

TSMC CoPoS and Intel EMIB Target Different Packaging Constraints TSMC CoPoS与Intel EMIB瞄准不同封装约束

SemiWiki · 21h ago SemiWiki · 21小时前

A comparison of TSMC's emerging CoPoS panel-level approach and Intel's EMIB bridge technology shows two routes to integrating ever-larger heterogeneous chiplet systems. Both address the scaling limits of advanced packaging, but their different physical tradeoffs will shape substrate size, interconnect density, and manufacturing economics. 一项对比显示，TSMC新兴的CoPoS面板级方案与Intel的EMIB桥接技术代表了集成更大规模异构chiplet系统的两条路径。两者都在应对先进封装的扩展极限，但不同的物理权衡将影响基板尺寸、互连密度和制造经济性。

Memory 存储

Qualcomm Signals a Double-Digit Flagship Chip Price Increase Qualcomm预示旗舰芯片价格将上涨两位数

Wccftech · 17h ago Wccftech · 17小时前

Qualcomm told customers that higher supplier costs will require a double-digit price increase, with the Snapdragon 8 Elite Gen 6 family among the products expected to be affected. Rising 2nm wafer costs and the DRAM shortage are pushing memory and foundry inflation into handset silicon even as smartphone demand remains soft. Qualcomm已告知客户，供应商成本上升将迫使其进行两位数涨价，预计Snapdragon 8 Elite Gen 6系列也会受到影响。2nm晶圆成本上升与DRAM短缺正把存储和代工端的通胀传导至手机芯片，而智能手机需求仍然偏弱。

CXMT Explores Custom DRAM as HBM Tool Controls Bite HBM设备管制收紧，CXMT转向定制DRAM

Wccftech · 19h ago Wccftech · 19小时前

CXMT is exploring customized DRAM, 3D memory, and 3D integration as U.S. restrictions limit its access to equipment needed for mainstream HBM production. The approach could create an alternative path into AI memory, but it remains a strategic proposal rather than evidence that CXMT has closed the manufacturing and bandwidth gap with Korean suppliers. 在美国限制其获取主流HBM生产所需设备之际，CXMT正探索定制DRAM、3D存储和3D集成。该路线可能为进入AI存储市场提供替代路径，但目前仍是战略设想，尚不能证明CXMT已经缩小与韩国供应商之间的制造和带宽差距。

Policy & Geopolitics 政策与地缘政治

Nvidia and 24 Companies Oppose a Chinese Open-Model Ban Nvidia等25家公司反对禁止中国开放权重模型

Tom's Hardware · 20h ago Tom's Hardware · 20小时前

Nvidia and 24 other companies signed a letter defending open-weight AI as Washington considers restrictions on Chinese models. Chipmakers, server vendors, cloud operators, and software firms joined the effort, while OpenAI, Anthropic, and Google were absent, exposing a policy split over whether model openness supports or undermines U.S. technology interests. 在华盛顿考虑限制中国AI模型之际，Nvidia与另外24家公司联署支持开放权重AI。芯片商、服务器厂商、云运营商和软件公司参与其中，而OpenAI、Anthropic和Google缺席，显示业界对模型开放究竟有利还是不利于美国科技利益存在政策分歧。

Earnings & Markets 财报与市场

Intel Commits to a 2028 High-Volume Ramp for 14A Intel承诺14A于2028年进入大规模量产

Tom's Hardware · 20h ago Tom's Hardware · 20小时前

Intel says its 14A process remains on track for a high-volume production ramp in 2028 as company sales rose 25% year over year and earnings exceeded guidance. The financial improvement gives Intel more room to fund the roadmap, but 14A's foundry credibility will ultimately depend on external customer products reaching volume. Intel表示14A制程仍按计划于2028年进入大规模量产；公司销售额同比增长25%，盈利也超过业绩指引。财务改善为Intel投入路线图提供了更大空间，但14A的代工可信度最终仍取决于外部客户产品能否实现量产。

Challengers & New Silicon 新势力与新硅

AMD and Cerebras Join Forces on AI Inference AMD与Cerebras联手推进AI推理

Axios · 18h ago Axios · 18小时前

AMD and Cerebras have formed an AI inference collaboration, linking an established accelerator supplier with Cerebras' wafer-scale architecture. The announcement expands the competitive set for inference deployments, although its impact will depend on named workloads, customers, and production-scale installations. AMD与Cerebras建立AI推理合作，把成熟加速器供应商与Cerebras的晶圆级架构连接起来。该合作扩大了推理部署的竞争版图，但实际影响仍取决于明确的工作负载、客户以及生产规模的落地。

Microchip Agrees to Acquire Edge-AI Chipmaker Hailo Microchip同意收购边缘AI芯片公司Hailo

Microchip Technology · 18h ago Microchip Technology · 18小时前

Microchip Technology signed a definitive agreement to acquire Hailo, bringing the challenger's edge-AI accelerators into a broad embedded-chip portfolio. The deal gives Hailo a larger sales and support channel while giving Microchip a faster route into deployed neural-network inference at the edge; financial terms were not disclosed in the candidate item. Microchip Technology签署最终协议收购Hailo，把这家新势力的边缘AI加速器纳入其广泛的嵌入式芯片组合。该交易为Hailo带来更大的销售和支持渠道，也让Microchip更快进入边缘神经网络推理市场；候选信息未披露财务条款。

Etched Raises $300 Million at a $10.3 Billion Valuation Etched融资3亿美元，估值达103亿美元

MarketScale · 20h ago MarketScale · 20小时前