📰 DigiTimesDigiTimesCodex · 实盘不做把握 5/10
北京时间·🕒 发布 15:38:26时区推断·⟳ 抓取 16:45:02回填·⚡ 决策 07-06 17:28:40·113.1 天前
DigiTimes称Anthropic与三星代工芯片计划更偏低成本推理，但交易仍处早期且未获双方确认
这不是新的签约或订单公告，而是对7月2日前后已传播的Anthropic-Samsung早期谈判传闻的二次解读。
📄 原文DigiTimes↗ 原文
中文網 繁體 简体 SUBSCRIBE My account Sign out SUBSCRIBE Trending Anthropic HBM AI Search NTN Convergence Inference Era AI Adv…
不做 三星电子（005930.KS）
不做：该消息是已广泛传播的早期谈判传闻的二次解读，未获官方确认、无订单金额和量产节点，且三星电子市值体量下缺少可量化短线edge。
不做 Power Integrations（PWR.US）
不做：正文未提及Power Integrations，也没有可验证的同向业务传导链，属于detect层误映射或弱相关噪音。
展开详情(成交明细 / 研究档案 / 原文全文 / 决策记录链)
研究档案
消息传播链
- The Information率先报道Anthropic与三星讨论制造定制AI芯片，属于媒体独家传闻而非公司公告。
- TechCrunch跟进称Anthropic正讨论定制芯片，强调该消息来自此前报道，扩散到英文科技媒体。
- TrendForce转述称项目仍处早期，Anthropic仍在定义芯片功能、性能目标和服务器角色，削弱立即订单属性。
- MarketWatch称韩国股市此前一周已因Anthropic-Samsung传闻出现强波动，并提到三星股价受相关报道刺激上涨。市场反应:报道称三星股价曾因相关传闻上涨约8%。
- DigiTimes本条文章把事件解释为偏低成本推理芯片而非顶级性能追逐，但同时写明Anthropic与三星均未确认交易。
关键理由
- 这不是新的签约或订单公告，而是对7月2日前后已传播的Anthropic-Samsung早期谈判传闻的二次解读。
- DigiTimes正文明确写到Anthropic只表示希望尽量多元化算力来源、三星拒评，说明当前缺少可量化订单金额、流片时间、产能锁定或营收贡献。
- 三星电子体量极大，单一未确认ASIC代工意向若没有金额和量产节点，不足以形成相对市值级短线重定价。
主要风险
- 若后续出现Anthropic或三星一手确认、明确2nm/先进封装订单金额和量产时间，三星代工复苏逻辑可能重新进入交易评估。
- 市场已在7月2日至7月4日围绕The Information/Bloomberg转述和韩国股价反应交易过该主题，本条DigiTimes低成本推理角度更像补充解读。
原文全文
展开原文 · 5,393 字
中文網
繁體
简体
SUBSCRIBE
My account
Sign out
SUBSCRIBE
Trending Anthropic
HBM
AI Search
NTN Convergence
Inference Era AI
Advanced Packaging
Trending
SUBSCRIBE
CONNECT WITH US
Sign out
REALTIME NEWS
Samsung Foundry comeback builds as Meta, Anthropic weigh chip deals
Semiconductors
18min ago
Home
Tech
ICT
Anthropic's reported chip plans with Samsung could ease inference costs, not chase top-end performance
Jay Liu
, Taipei
Jul 6, 2026, 15:38
0
Credit: Claude
Anthropic's reported move into in-house chip development could matter well beyond Silicon Valley if it helps lower the cost of running AI services worldwide. By prioritizing cheaper inference rather than elite performance, the startup may be signaling a more pragmatic path that could influence how global AI systems are built and priced.
Anthropic is reportedly working on its own chips and has begun collaborating with Samsung Electronics' foundry process, according to industry observers. Neither company has confirmed the arrangement. Anthropic said only that it aims to diversify its compute sources as much as possible, while Samsung declined to comment.
The report has raised questions about whether Samsung is a practical foundry partner for a leading AI startup, given that cloud AI chips are often built on TSMC processes to secure stronger performance. However, chip industry sources say the project may be less about chasing the most advanced specification and more about improving inference efficiency, the part of AI operations that serves end users at scale.
Samsung is not new to ASIC work. Supplier sources say its foundry division already provides design services and has handled several notable projects, including Tesla's AI chip, Google's CPU, and Groq's LPU. Meta has also reportedly considered shifting some ASIC orders to Samsung. Against that backdrop, Anthropic's reported interest would not be unusual.
Anthropic already appears to have access to substantial compute from multiple suppliers, including Nvidia GPUs, Google TPUs, and AWS Trainium chips. That suggests its training needs are already broadly covered, leaving its custom chip effort more likely aimed at inference, where large user demand can drive high compute costs.
For inference chips, process requirements are generally less demanding than for the most advanced training accelerators, making Samsung's foundry technology a plausible fit. If TSMC's leading-edge capacity and advanced packaging remain tight and expensive, Anthropic's needs may align more naturally with Samsung's offerings.
Compared with OpenAI, Anthropic has taken a more cautious approach to chip strategy. ASIC industry sources say its in-house chip plans are arriving relatively late. Rather than replacing external suppliers, the company has preferred to use a broad mix of chips and focus its engineering resources on AI development itself.
That history suggests any Anthropic chip would likely be shaped by practicality, with cost-efficient inference as the main goal.
Article
translated by Jingyue Hsiao and edited by Jack Wu
AI Search
Related stories
Samsung eyes marquee AI win as Anthropic explores custom chip beyond Nvidia
Samsung Foundry eyes Anthropic as OpenAI chip project reportedly stalls
Anthropic-Microsoft deal could broaden ASIC demand across cloud supply chains
Potential Google-Marvell partnership escalates AI chip competition against Nvidia
Anthropic reportedly explores in-house chip design amid rapid revenue growth and evolving AI compute stack
Show more
Categories
AI Chips + components IC design, distribution IT + CE
Tags
AI inference Anthropic ASIC chips cost development Google performance Samsung Silicon Valley startup training worldwide
Share this article
Other links
Print
Related topics
Global supply chain: Key components
Global supply chain: Emerging tech, businesses
Innovations
Startups and innovations
BIZ FOCUS
Jun 1, 08:00
LITEON Showcases AI at COMPUTEX Panel Featuring NVIDIA, Infineon, GIGABYTE
Friday 29 May 2026
Datotek Introduces AI PSSD Sustainable Storage Solutions at COMPUTEX 2026
Thursday 28 May 2026
AGI Debuts DDR5 R-DIMM and Smart Industrial Solutions at COMPUTEX 2026
Thursday 28 May 2026
Retronix Launches Sparrow Hawk Edge AI Computing Platform Worldwide
MOST-READ
7 DAYS NEWS
Samsung Foundry turns to 5nm, 8nm orders as 2nm comeback takes shape
Samsung eyes marquee AI win as Anthropic explores custom chip beyond Nvidia
Intel deepens ties with Taiwan's chip supply chain as October talks expand
MediaTek issues official price hike notice, products facing supply constraints likely to increase first
Micron and GM strike long-term memory supply deal for vehicles
Weekly news roundup: Anthropic tightens controls; Innoscience-Infineon clash shows the weaponization of China's courts
India launches third semiconductor plant as CG Semi begins commercial production in Gujarat
Alibaba reportedly bans Claude Code for employees, citing security concerns as China shifts toward domestic AI coding tools
Micron breaks ground on Hiroshima expansion to boost AI memory chip production
Analysis: Taiwan's semiconductor sector enters 2026 helium crunch, with nearly 90% dependent on Qatar
Full list
Members only
×
Sorry, the page you are trying to open is available only for our paid subscribers.
Please login to read more
New users, please register first
Notification
×
繁體
简体
SUBSCRIBE
My account
Sign out
SUBSCRIBE
Trending Anthropic
HBM
AI Search
NTN Convergence
Inference Era AI
Advanced Packaging
Trending
SUBSCRIBE
CONNECT WITH US
Sign out
REALTIME NEWS
Samsung Foundry comeback builds as Meta, Anthropic weigh chip deals
Semiconductors
18min ago
Home
Tech
ICT
Anthropic's reported chip plans with Samsung could ease inference costs, not chase top-end performance
Jay Liu
, Taipei
Jul 6, 2026, 15:38
0
Credit: Claude
Anthropic's reported move into in-house chip development could matter well beyond Silicon Valley if it helps lower the cost of running AI services worldwide. By prioritizing cheaper inference rather than elite performance, the startup may be signaling a more pragmatic path that could influence how global AI systems are built and priced.
Anthropic is reportedly working on its own chips and has begun collaborating with Samsung Electronics' foundry process, according to industry observers. Neither company has confirmed the arrangement. Anthropic said only that it aims to diversify its compute sources as much as possible, while Samsung declined to comment.
The report has raised questions about whether Samsung is a practical foundry partner for a leading AI startup, given that cloud AI chips are often built on TSMC processes to secure stronger performance. However, chip industry sources say the project may be less about chasing the most advanced specification and more about improving inference efficiency, the part of AI operations that serves end users at scale.
Samsung is not new to ASIC work. Supplier sources say its foundry division already provides design services and has handled several notable projects, including Tesla's AI chip, Google's CPU, and Groq's LPU. Meta has also reportedly considered shifting some ASIC orders to Samsung. Against that backdrop, Anthropic's reported interest would not be unusual.
Anthropic already appears to have access to substantial compute from multiple suppliers, including Nvidia GPUs, Google TPUs, and AWS Trainium chips. That suggests its training needs are already broadly covered, leaving its custom chip effort more likely aimed at inference, where large user demand can drive high compute costs.
For inference chips, process requirements are generally less demanding than for the most advanced training accelerators, making Samsung's foundry technology a plausible fit. If TSMC's leading-edge capacity and advanced packaging remain tight and expensive, Anthropic's needs may align more naturally with Samsung's offerings.
Compared with OpenAI, Anthropic has taken a more cautious approach to chip strategy. ASIC industry sources say its in-house chip plans are arriving relatively late. Rather than replacing external suppliers, the company has preferred to use a broad mix of chips and focus its engineering resources on AI development itself.
That history suggests any Anthropic chip would likely be shaped by practicality, with cost-efficient inference as the main goal.
Article
translated by Jingyue Hsiao and edited by Jack Wu
AI Search
Related stories
Samsung eyes marquee AI win as Anthropic explores custom chip beyond Nvidia
Samsung Foundry eyes Anthropic as OpenAI chip project reportedly stalls
Anthropic-Microsoft deal could broaden ASIC demand across cloud supply chains
Potential Google-Marvell partnership escalates AI chip competition against Nvidia
Anthropic reportedly explores in-house chip design amid rapid revenue growth and evolving AI compute stack
Show more
Categories
AI Chips + components IC design, distribution IT + CE
Tags
AI inference Anthropic ASIC chips cost development Google performance Samsung Silicon Valley startup training worldwide
Share this article
Other links
Related topics
Global supply chain: Key components
Global supply chain: Emerging tech, businesses
Innovations
Startups and innovations
BIZ FOCUS
Jun 1, 08:00
LITEON Showcases AI at COMPUTEX Panel Featuring NVIDIA, Infineon, GIGABYTE
Friday 29 May 2026
Datotek Introduces AI PSSD Sustainable Storage Solutions at COMPUTEX 2026
Thursday 28 May 2026
AGI Debuts DDR5 R-DIMM and Smart Industrial Solutions at COMPUTEX 2026
Thursday 28 May 2026
Retronix Launches Sparrow Hawk Edge AI Computing Platform Worldwide
MOST-READ
7 DAYS NEWS
Samsung Foundry turns to 5nm, 8nm orders as 2nm comeback takes shape
Samsung eyes marquee AI win as Anthropic explores custom chip beyond Nvidia
Intel deepens ties with Taiwan's chip supply chain as October talks expand
MediaTek issues official price hike notice, products facing supply constraints likely to increase first
Micron and GM strike long-term memory supply deal for vehicles
Weekly news roundup: Anthropic tightens controls; Innoscience-Infineon clash shows the weaponization of China's courts
India launches third semiconductor plant as CG Semi begins commercial production in Gujarat
Alibaba reportedly bans Claude Code for employees, citing security concerns as China shifts toward domestic AI coding tools
Micron breaks ground on Hiroshima expansion to boost AI memory chip production
Analysis: Taiwan's semiconductor sector enters 2026 helium crunch, with nearly 90% dependent on Qatar
Full list
Members only
×
Sorry, the page you are trying to open is available only for our paid subscribers.
Please login to read more
New users, please register first
Notification
×
决策记录链 · 只追加
每次领取产生一条记录,重新决策追加一条并以 supersedes 指向被替代的记录(ADR 0014)。
这条决策早于决策账本(2026-10-08 起逐条追加),没有追加记录。