Semiconductors / 半導體
-
Industry ENOne Hundred Trillion Won: Samsung's First $80 Billion Quarter Redefines the AI Boom's Pecking Order
Samsung guides to 107.4 trillion won (~$80B) in Q3 operating profit — up 783% YoY and the first 100-trillion-won quarter in its 56-year history — as AI memory scarcity turns chips into the boom's real kingmakers.
-
Industry 中一百兆韓元的一季：三星首度單季賺進 800 億美元，改寫 AI 熱潮的權力版圖
三星電子公布第三季初步財測：營業利益 107.4 兆韓元（約 800 億美元）、年增 783%，創公司 56 年來首見的百兆韓元季度，AI 記憶體短缺讓晶片廠成為這波熱潮真正的贏家。
-
Industry ENFinancing Is the Product: Broadcom Lines Up $50 Billion So OpenAI Can Buy Its Own Chips
Broadcom is arranging over $50 billion in private credit for OpenAI's Nexus custom-silicon program, its third vendor-financing package since June — and the clearest sign yet that in the AI buildout, the money is now part of the machine.
-
Industry 中融資即產品：Broadcom 為 OpenAI 籌措 500 億美元，讓它買下自家晶片
Broadcom 正為 OpenAI 的 Nexus 自訂晶片計畫安排超過 500 億美元私募信貸，這是它 6 月以來第三筆供應商融資方案——也清楚說明在 AI 建設浪潮中，資金本身已成為機器的一部分。
-
Industry ENNT$500 Billion a Month: TSMC's Record Q3 Shows AI Demand Still Has No Ceiling
TSMC's September revenue hit NT$511.86B, up 54.6% YoY, sealing a record NT$1.49T quarter that beat every forecast — the clearest signal yet that the AI buildout is still accelerating.
-
Industry 中月營收站上五千億：台積電第三季寫下新紀錄，AI 需求依然不見天花板
台積電 9 月營收達 5,118.6 億新台幣、年增 54.6%，第三季以 1.494 兆元的歷史新高作收、優於所有預估——這是 AI 建置熱潮仍在加速的最清晰訊號。
-
Tools ENThe Chip That Wrote Itself: openTPU Runs Qwen3 on an FPGA Its Agents Designed
A single GitHub committer claims AI agents designed a full-stack inference accelerator — RTL, ISA, compiler and all — and it now runs ten open models bit-exactly on a $200 Kintex-7 FPGA card.
-
Tools 中自己設計自己的晶片：openTPU 讓 AI Agent 打造的 FPGA 加速器跑起 Qwen3
一位 GitHub 開發者聲稱，AI agents 設計了整套推論加速器——從 RTL、指令集到編譯器——如今它在一張約 200 美元的 Kintex-7 FPGA 卡上，位元級精確地跑著十個開源模型。
-
Industry ENOne Petaflop in Your Lap: Microsoft Bets the PC's Next Chapter on Local AI
At its San Francisco event today, Microsoft pairs Satya Nadella with Jensen Huang to launch the RTX Spark-powered Surface Laptop Ultra and an agentic vision for Windows — the biggest PC bet since Copilot+.
-
Industry 中一兆次運算放進筆電：微軟把 PC 的下一章押在本地 AI 上
微軟今天在舊金山舉行發表會，由 Nadella 與黃仁勳同台，推出搭載 RTX Spark 的 Surface Laptop Ultra 與代理式 Windows 願景——這是 Copilot+ 之後微軟最大的一次 PC 豪賭。
-
Industry ENThe $60 Billion Test: Wall Street Starts Syndicating the Largest Chip-Debt Deal Ever Built
Bank of America, Citigroup, and Morgan Stanley have launched syndication of a record $60 billion debt package backing Anthropic's TPU leases — the largest chip-financing deal ever structured, and a test of whether the financial system can absorb AI-scale compute debt.
-
Industry 中600 億美元的大考驗：華爾街開始分銷史上最大晶片債務交易
美國銀行、花旗與摩根士丹利已啟動一筆 600 億美元債務方案的分銷，背後是 Anthropic 的 TPU 租賃——這是史上結構規模最大的晶片融資交易，也是對金融體系能否吸收 AI 級運算債務的一場測試。
-
Industry ENThe $100 Phone Is Going Extinct: AI Data Centers Are Eating the World's Memory Chips
Sub-$100 smartphone shipments collapsed nearly 60% in a year as Samsung, SK Hynix, and Micron divert memory to AI data centers — and the digital divide is the first casualty.
-
Industry 中百美元手機正在絕跡：AI 資料中心正在吃掉全世界的記憶體晶片
100 美元以下智慧型手機出貨量一年內暴跌近 60%，因為三星、SK 海力士與美光把記憶體產能全數轉向 AI 資料中心——而數位落差的擴大，成為第一個犧牲品。
-
Industry ENAsk Not for Whom the Fab Bells Toll: Lisa Su Asks AMD's Rivals to Share Taiwan's Packaging Capacity
With 14GW of AI deals in hand, AMD's CEO is in Taiwan asking competitors to stop outbidding each other for advanced-packaging capacity — the boldest supply-chain diplomacy yet in the compute arms race.
-
Industry 中先進封裝不夠用了：蘇姿丰親赴台灣，開口請對手「別再互相搶產能」
手上握有 14GW AI 訂單的 AMD 執行長蘇姿丰本週走訪台灣，罕見地請競爭對手協調共享先進封裝與載板產能——算力軍備競賽至今最大膽的供應鏈外交。
-
Industry ENDoubling Down Twice: Etched Fields Funding Bids at $40–50 Billion, Just Weeks After Its $21 Billion Round
TechCrunch reports that AI inference chip startup Etched is reviewing investment offers at $40–50 billion valuations — more than double the $21 billion it was worth in September — as investors bet its purpose-built transformers can dent Nvidia's grip on inference.
-
Industry 中六週身價再翻倍：AI 推論晶片新創 Etched 傳獲 400–500 億美元投資報價
據 TechCrunch 報導，九月底才以 210 億美元估值募得 7 億美元的 AI 推論晶片新創 Etched，目前正在評估 400 億至 500 億美元的投資報價——投資人押注其專為 transformer 打造的專用晶片能撼動 Nvidia 在推論市場的霸主地位。
-
Industry ENThe License That Ate the Company: Ex-Groq Engineers Sue Over Nvidia's $20 Billion Deal
Two former Groq engineers told Delaware's Court of Chancery that Nvidia's $20 billion 'licensing' deal was an acqui-hire in disguise that paid insiders billions and left common stockholders behind.
-
Industry 中吃掉整家公司的「授權」：前 Groq 工程師就 Nvidia 200 億美元交易提告
兩位前 Groq 工程師向德拉瓦州衡平法院指控，Nvidia 的 200 億美元「授權」協議實為變相收購，內部人士獲利數十億美元，普通股股東卻被拋在後頭。
-
Industry ENThe Check Reverses: Qualcomm Becomes a Net Patent Payer to Huawei in a Landmark Cross-License
Huawei and Qualcomm signed a multi-year patent cross-license spanning 5G, compute, AI, and networking — with Qualcomm paying Huawei for the first time and buying a slice of Huawei's US patent portfolio.
-
Industry 中二十五年来首次反向付款：高通與華為簽署歷史性專利交叉授權，AI 與運算專利成核心
華為與高通宣布多年期專利交叉授權，涵蓋 5G、運算、AI 與網路技術，高通首次成為淨付款方，並將收購華為部分美國專利。
-
Meta ENSpinning Rust Strikes Back: Toshiba Will Double HDD Capacity for AI Data Centers by 2027
Toshiba is spending ¥60B ($380M) to double hard-drive output for AI data centers by FY2027, targeting 30% capacity share — and the news instantly wiped 10-12% off Seagate and Western Digital.
-
Meta 中旋轉磁碟的逆襲：東芝將在 2027 年前把 AI 資料中心硬碟產能翻倍
東芝投入 600 億日圓（約 3.8 億美元），要在 2027 財年前將 AI 資料中心用硬碟產能翻倍、搶下 30% 容量市占——消息一出，Seagate 與 Western Digital 股價應聲重挫 10%。
-
Research ENFour TPUs in Orbit: Google's Suncatcher Prototype Is Live and Phoning Home
Google's first Project Suncatcher satellite — carrying four Trillium TPUs — reached orbit October 1 aboard SpaceX's Transporter-18 and is operating as expected, opening the first in-orbit test of AI compute hardware.
-
Research 中四顆 TPU 進駐軌道：Google Suncatcher 原型衛星已上線並成功回傳訊號
Google 首顆 Project Suncatcher 衛星搭載四顆 Trillium TPU，於 10 月 1 日隨 SpaceX Transporter-18 任務進入軌道且運作正常，開啟 AI 運算硬體的首次軌道實測。
-
Policy EN343 Machines That Print AI Chips: New Report Puts Hard Numbers on China's ASML Stockpile
A Center for Technology & Statecraft report estimates Chinese fabs now hold 343 immersion DUV lithography systems — enough to keep printing 7nm-class Huawei Ascend AI chips — and calls for a China-wide export ban.
-
Policy 中343 台印製 AI 晶片的機器：新報告量化中國的 ASML 微影設備庫存
Center for Technology & Statecraft 報告估計，中國晶圓廠已累積 343 台浸潤式 DUV 微影設備——足以持續量產 7 奈米級華為昇騰 AI 晶片——並呼籲全面對中禁運。
-
Industry ENOne Customer, $20B a Fab: TSMC Weighs a Texas Campus With Musk's Terafab as Anchor
TSMC is exploring a multi-fab Texas campus with Elon Musk's Terafab as anchor client — a pivotal geography shift if the US 35% manufacturing tax credit gets extended past year-end.
-
Industry 中一座廠 200 億美元：台積電評估以馬斯克 Terafab 為錨定客戶的德州多廠園區
台積電正評估與馬斯克 Terafab 合作、在德州打造多座晶圓廠園區——前提是美國 35% 先進製造稅收抵免能在年底前獲得延長。
-
Tools ENSix Repos Against CUDA: DeepSeek and Huawei Open-Source the Ascend Software Stack
DeepSeek has open-sourced its Ascend software stack — TileLang, DeepGEMM, DeepEP, FlashMLA, TileKernels, and DeepSelect — giving Huawei's chips a CUDA-alternative toolkit that already powers most operators in DeepSeek V4 training.
-
Tools 中六個開源儲存庫對決 CUDA：DeepSeek 與華為開源 Ascend 軟體堆疊
DeepSeek 開源了完整 Ascend 軟體堆疊——TileLang、DeepGEMM、DeepEP、FlashMLA、TileKernels 與 DeepSelect——為華為晶片提供 CUDA 之外的完整工具鏈，且已在 DeepSeek V4 訓練中承擔多數算子。
-
Industry ENTwelve Racks a Month: Inside Bull's Doubled Angers Factory, Europe's Only Supercomputer Plant
Bull, France's state-owned supercomputer maker, has reopened its expanded Angers factory after an €80 million rebuild — doubling output to 12 racks a month, assembling Alice Recoque and LUMI-AI in parallel, and scaling toward 24 racks as Europe races to close its AI compute gap with the US and China.
-
Industry 中每月十二座機櫃：Bull 倍增生產的 Angers 工廠，歐洲唯一的超級電腦製造廠
法國國營超級電腦製造商 Bull 在投入 8,000 萬歐元擴建後重新啟用 Angers 工廠，月產能從六座機櫃倍增至十二座，可平行組裝 Alice Recoque 與 LUMI-AI 兩大系統，並視需求於明年提升至 24 座——這是歐洲急起直追、縮小與美中 AI 算力差距的最新一步。
-
Tools ENHalf the Memory, Two-Thirds the Price: DGX Spark 64GB Brings Local Agents Down to $4,999
NVIDIA's new 64GB DGX Spark cuts the entry price of a Grace Blackwell desktop to $4,999, runs 100B-parameter models on device, and clusters two units to 128GB via the new Sync Cluster Assistant.
-
Tools 中記憶體減半、價格省三分之二：DGX Spark 64GB 把本地 AI 代理的入門價降到 4,999 美元
NVIDIA 推出 64GB 版 DGX Spark，把 Grace Blackwell 桌上型 AI 主機的入門價砍到 4,999 美元，單機可跑 100B 參數模型，兩台透過新版 Sync Cluster Assistant 叢集可擴充到 128GB。
-
Industry ENThe Chips Stay, the Debt Leaves: Amazon's $8 Billion SPV Would Let It Compute on Hardware It No Longer Owns
Amazon is in talks to move ~$8B of installed Nvidia Grace Blackwell chips into an off-balance-sheet SPV and lease them back, easing capex pressure as annual spending heads toward $220B — the latest, largest sign that AI compute is becoming a financial product.
-
Industry 中晶片留在機房，債務離開帳本：Amazon 80 億美元 SPV 交易將讓它用「不屬於自己的」硬體做運算
Amazon 正洽商將約 80 億美元已部署的 Nvidia Grace Blackwell 晶片移入表外 SPV 再租回使用，在年度資本支出邁向 2,200 億美元之際減輕帳面壓力——這是 AI 算力正成為金融商品的最新、也是最大訊號。
-
Industry ENOne Sentence, Two Flagships: Qualcomm Renews With Apple and Bets the Phone on Agentic AI
Qualcomm extended its global patent license with Apple on April 1, 2027 and shipped two 2nm Snapdragon flagships built for on-device agents — the hybrid-AI future now has a silicon business plan.
-
Industry 中一句話續約、兩款旗艦晶片：Qualcomm 與 Apple 达成協議，把手機的未來押在 Agentic AI 上
Qualcomm 確認與 Apple 的全球專利授權自 2027 年 4 月 1 日續約，同時推出兩款 2nm 旗艦晶片，為裝置端 AI 代理而生的混合 AI 架構，現在有了完整的晶片商業計畫。
-
Industry ENThe Lender Is Also the Supplier: Broadcom's $42 Billion Convertible Loan Anchors Anthropic's $125 Billion TPU Bet
Anthropic's IPO prospectus reveals Broadcom has agreed to lend the AI lab up to $42 billion via convertible notes to finance its $125.2 billion, five-year TPU lease — a supplier-as-financier structure that mirrors Nvidia's playbook and concentrates risk in two companies.
-
Industry 中貸方即供應商：博通 420 億美元可轉債，押注 Anthropic 1,252 億美元 TPU 大計
Anthropic 的 IPO 招股書揭露，博通已同意透過可轉換票據借給這家 AI 實驗室最多 420 億美元，資助其為期五年、總額 1,252 億美元的 TPU 租約——這種「供應商兼金主」結構仿效輝達打法，也把風險集中在兩家公司身上。
-
Tools ENThe Model That Designs the Chips: OpenAI and Synopsys Launch GPT-Synopsys
OpenAI and Synopsys signed a multi-year deal to build GPT-Synopsys, a frontier model trained to run EDA tools like an expert engineer — with agents closing PPA, timing, and verification loops on their way to first-time-right silicon.
-
Tools 中設計晶片的模型：OpenAI 與新思科技聯手推出 GPT-Synopsys
OpenAI 與 Synopsys 簽署多年協議，共同打造 GPT-Synopsys——一個像資深工程師一樣操作 EDA 工具的前沿模型，讓 agent 逐步閉合 PPA、時序與驗證迴圈，朝一次流片成功邁進。
-
Industry EN379% Growth and an 87% Margin: Micron's Record Quarter Shows AI Now Runs on Memory
Micron closed fiscal 2026 with $54.23B in quarterly revenue, an ~87% gross margin, and Q1 FY27 guidance of up to $63B — the clearest evidence yet that the AI boom's bottleneck has shifted from GPUs to memory.
-
Industry 中營收暴增 379%、毛利近 87%：美光創紀錄財報證明 AI 的瓶頸已從 GPU 轉向記憶體
美光（Micron）以 542.3 億美元季度營收、近 87% 毛利率結束 2026 財年，並給出最高 630 億美元的下季指引——這是 AI 榮景的瓶頸已從 GPU 移向記憶體最明確的證據。
-
Industry ENThe $8.2 Billion Godmother: AMD Buys Fei-Fei Li's World Labs to Build AI Silicon That Understands Reality
AMD is acquiring Fei-Fei Li's spatial-intelligence startup World Labs in an $8.2 billion all-stock deal — its second-largest ever — bringing the 'Godmother of AI' in as chief scientist to design future accelerators around world models.
-
Industry EN82 億美元請來「AI 教母」：AMD 收購李飛創辦的 World Labs，要打造懂真實世界的 AI 晶片
AMD 以約 82 億美元的全股票交易收購李飛飛創辦的空間智慧新創 World Labs，這是該公司史上第二大收購案。李飛飛將出任首席科學家，直接向執行長蘇姿丰匯報，未來加速器將圍繞世界模型的運算需求來設計。
-
Tools ENThe CUDA Killer Is Free: DeepSeek Open-Sources Its Entire Huawei Ascend Toolkit
DeepSeek publicly releases TileLang, DeepGEMM, DeepEP and its full Huawei Ascend software stack — the clearest attempt yet to erase the switching cost that has locked the AI industry into Nvidia's CUDA for over a decade.
-
Tools 中CUDA 殺手免費開放：DeepSeek 開源整套華為 Ascend 工具鏈
DeepSeek 公開釋出 TileLang、DeepGEMM、DeepEP 與整套華為 Ascend 軟體棧——這是迄今為止最直接的一次嘗試，要抹掉十年來把 AI 產業鎖在 Nvidia CUDA 上的轉換成本。
-
Industry ENSix Samsungs, One Check: Samsung Group Commits $1 Billion to KKR's Helix AI Infrastructure Venture
Samsung Electronics and five affiliates will pour a combined $1 billion into Helix Digital Infrastructure, the KKR-launched, Nvidia- and Vistra-backed AI buildout platform led by former AWS CEO Adam Selipsky — the conglomerate's most coordinated capital move into the AI infrastructure race yet.
-
Industry 中六個三星，一張支票：三星集團向 KKR 旗下 Helix AI 基礎設施公司注資 10 億美元
三星電子與五家關係企業將共同向 Helix Digital Infrastructure 投資 10 億美元。這家由 KKR 創立、獲 Nvidia 與 Vistra 支持、由前 AWS 執行長 Adam Selipsky 領軍的 AI 基礎設施平台，如今迎來第一個加入創始投資人行列的亞洲工業集團。
-
Industry ENThe Largest Buyback in Market History: Nvidia's $150 Billion Wager on Itself
Nvidia's board added $150 billion to its share repurchase program — the biggest authorization increase ever — lifting remaining capacity to $235 billion through fiscal 2028, eclipsing Apple's $110 billion record and doubling down on the AI boom just as chip competition intensifies.
-
Industry 中史上最大規模庫藏股：輝達對自己的 1,500 億美元豪賭
輝達董事會為庫藏股計畫追加 1,500 億美元授權——史上最大的單次增幅——使剩餘回購額度達 2,350 億美元、預計執行至 2028 財年，一舉超越蘋果 2024 年創下的 1,100 億美元紀錄，在 AI 晶片競爭升溫之際加倍押注自身前景。
-
Industry ENStarship Reaches Orbit at Last: Flight 14 Clears the Path for Space-Based AI Compute
SpaceX's Starship completed its first-ever orbital insertion on Flight 14, deploying 26 Starlink V3 satellites and unlocking the launch capacity behind orbital AI data centers.
-
Industry 中星艦終於抵達軌道：Flight 14 為太空 AI 算力時代掃清最後障礙
SpaceX 星艦在第 14 次試飛中首次成功進入地球軌道，部署 26 顆 Starlink V3 衛星，為軌道 AI 資料中心時代打通了運力關鍵瓶頸。
-
Industry ENThe Biggest Buyback in Corporate History: Nvidia Bets $150 Billion on Itself
Nvidia's board added a record $150 billion to its share repurchase program, lifting total authorization to $235 billion — the largest increase ever announced by a US company.
-
Industry 中企業史上最大規模回購：輝達豪賭 1,500 億美元押注自己
輝達董事會為股票回購計畫追加創紀錄的 1,500 億美元，總授權額升至 2,350 億美元——這是美國企業史上最大的單次回購授權增幅。
-
Industry ENA Million Chips for ByteDance: China Moves to Clear Nvidia's RTX Pro 5500
China's MIIT has told ByteDance, Alibaba and other domestic giants it intends to approve purchases of Nvidia's new RTX Pro 5500 — with ByteDance alone reportedly planning a one-million-unit order.
-
Industry 中字節跳動百萬張訂單：中國擬放行輝達 RTX Pro 5500
中國工信部已向字節跳動、阿里巴巴等國內大廠表明有意核准採購輝達新款 RTX Pro 5500 晶片，據報字節跳動單獨規劃了下單一百萬張的採購計畫。
-
Industry ENFiber, Not Chips: Musk's Colossus 2 Timetable Aims at 1.21 Million GPUs by New Year
Musk's most detailed Colossus 2 timetable yet: 220,000 GB300s online next week, another 220,000 in November, and — 'if we get lucky' — a third batch by late December, taking the Memphis cluster past 1.2 million GPUs.
-
Industry 中瓶頸不再是晶片：馬斯克公布 Colossus 2 時程表，年底衝刺 121 萬顆 GPU
馬斯克迄今最詳細的 Colossus 2 擴建時程：22 萬顆 GB300 下週上線、11 月再一批，年底「如果夠幸運」再加一批，讓曼菲斯集群突破 120 萬顆 GPU。
-
Industry ENThe Brand That Recall Broke: Microsoft Retires 'Copilot+ PC' Two Years In
Microsoft's Surface CVP confirms new Surface PCs 'are not called Copilot+ PCs' — the AI PC category label born in 2024 is quietly dead, killed by Recall's security backlash and consumer confusion.
-
Industry 中被 Recall 拖垮的品牌：Microsoft 悄悄淘汰兩年前的「Copilot+ PC」
Microsoft Surface 高層證實新款 Surface「不再稱為 Copilot+ PC」——2024 年誕生的 AI PC 品類標籤悄然退場，始作俑者是 Recall 的資安風暴與消費者的困惑。
-
Industry ENThe Jet That Never Spun: Crusoe Walks Away From Its $1.25 Billion Boom Turbine Deal
Crusoe has scrapped its plan to power AI data centers with Boom Supersonic's jet-derived gas turbines, ending a $1.25 billion launch partnership and resetting the playbook for behind-the-meter AI power.
-
Industry 中永不旋轉的噴射引擎：Crusoe 放棄 12.5 億美元 Boom 渦輪機交易
Crusoe 確定終止以 Boom Supersonic 噴射引擎改裝燃氣渦輪機供電 AI 資料中心的計畫，12.5 億美元的先導合作告吹，也重寫了 AI 電力採購的遊戲規則。
-
Meta EN1.44 Million GPUs and a 1.2 GW Power Plant: SpaceXAI's Colossus Nears the Finish Line
SpaceXAI is adding 660,000 GB300 GPUs this year to reach 1.44 million in operation, backed by its own 1.2-gigawatt power plant in Mississippi — the endgame of Musk's two-year million-GPU bet.
-
Meta 中144 萬張 GPU 與 1.2 GW 電廠：SpaceXAI 的 Colossus 抵達終線前最後一哩
SpaceXAI 今年將再新增 66 萬張 GB300 GPU，使在運總數達 144 萬張，並由自建的 12 億瓦密西西比電廠供電——馬斯克兩年百萬 GPU 豪賭的終局之戰。
-
Industry ENZero Lithography Machines, Zero Percent: ASML's Home Continent Just Stopped Buying
Europe accounted for 0% of ASML's system sales in the first half of 2026 — down from 1% in 2025 and 5% in 2024 — and the Dutch lithography giant is now asking Brussels to manufacture demand, not just subsidize fabs.
-
Industry 中零台微影機、零營收：ASML 的娘家歐洲，已經半年不買單
2026 上半年，歐洲在 ASML 系統銷售的區域佔比為 0%——低於 2025 年的 1% 與 2024 年的 5%——這家荷蘭微影設備龍頭如今呼籲布魯塞爾：與其補貼蓋廠，不如創造需求。
-
Industry ENFrom Dojo's Ashes to a $10 Billion Valuation: DensityAI Nears a Mega-Round With an AWS Deal Attached
DensityAI, the AI inference chip startup founded by the leaders of Tesla's shuttered Dojo supercomputer team, is in advanced talks to raise hundreds of millions of dollars at a $10 billion valuation — with a conditional chip procurement agreement from Amazon Web Services reportedly already in hand.
-
Industry 中從 Dojo 廢墟到百億美元估值：DensityAI 可望完成鉅額融資，AWS 條件式採購協議成最大籌碼
由特斯拉已解散的 Dojo 超級電腦團隊核心成員創立的 AI 推論晶片新創 DensityAI，據報正在進行 advanced talks，以約 100 億美元估值募集數億美元資金，並已與 Amazon Web Services 簽下條件式晶片採購協議——效能達標，AWS 就買單。
-
Tools ENUpstream or Bust: Qualcomm Ships a Linux Developer Preview for Snapdragon X2 Laptops
At Snapdragon Summit, Qualcomm released an early Linux developer preview for Snapdragon X2 laptops — a custom kernel with Debian 13, upstreamed drivers, Mesa graphics, and FastRPC access to the 80 TOPS Hexagon NPU, with Ubuntu certification due in 2027.
-
Tools 中上游優先：Qualcomm 為 Snapdragon X2 筆電推出 Linux 開發者預覽
Qualcomm 在 Snapdragon Summit 發布 Snapdragon X2 的 Linux 早期開發者預覽：自訂核心搭配 Debian 13、上游化驅動、Mesa 繪圖堆疊，並透過 FastRPC 開放 80 TOPS Hexagon NPU，Ubuntu 認證預計 2027 年到位。
-
Industry EN128 GOPS in Your Earbuds: Qualcomm's Snapdragon Sound Elite Gen 2 Makes Hearables the New AI Frontier
Qualcomm's new hearables platform doubles on-device AI performance, adds micro-power Wi-Fi 6E for direct-to-cloud agents, and launches a health alliance with Optum and Scripps — turning earbuds into the quiet vanguard of personal AI.
-
Industry 中128 GOPS 塞進耳機裡：Qualcomm Snapdragon Sound Elite Gen 2 讓穿戴裝置成為個人 AI 新戰場
Qualcomm 全新聽穿戴平台將裝置端 AI 效能翻倍、內建微功耗 Wi-Fi 6E 直連雲端代理，並與 Optum、Scripps 組成健康聯盟——耳機正悄悄成為個人 AI 的先鋒載體。
-
Industry ENThe Fourth Member: AMD Joins the Trillion-Dollar Club on the Back of Helios
AMD closed above a $1 trillion market cap for the first time on September 21, powered by a 24% weekly rally, data-center revenue up 107% to $6.7B, and the first Helios racks shipping to AI labs.
-
Industry 中第四位成員：AMD 搭上 Helios 東風，正式加入兆美元俱樂部
AMD 於 9 月 21 日首次收在 1 兆美元市值之上，靠的是單週 24% 的漲勢、年增 107% 達 67 億美元的資料中心營收，以及開始出貨給 AI 實驗室的首批 Helios 機架。
-
Policy ENNinety Percent Foreign Silicon: China's SASAC Surveys Broadcom's Grip on State Data Centers
An FT-reported SASAC survey found Broadcom switches may run up to 90% of networking gear at some Chinese state-owned data centers — and informal guidance to cut back could follow.
-
Policy 中九成仰賴洋晶片：中國國資委普查博通交換機在國有數據中心的滲透率
英國《金融時報》報導，中國國資委調查發現部分國有企業數據中心的網通設備高達九成來自博通，後續可能以「非正式指導」要求逐步替換為國產方案。
-
Industry ENChina's Most Powerful AI Chip and a 20-Gigawatt Bet: Inside Alibaba's Full-Stack Apsara Announcement
At Apsara 2026 in Hangzhou, Alibaba unveiled the Zhenwu V900 AI accelerator, targeted a 5-10 trillion parameter model, and pledged over 20GW of cloud capacity by 2032.
-
Industry 中中國最強 AI 晶片與 20GW 豪賭：阿里巴巴雲棲大會全棧戰略解析
阿里巴巴在杭州雲棲大會發布 Zhenwu V900 AI 加速器，宣布挑戰 5 至 10 兆參數模型，並承諾 2032 年前營運超過 20GW 全球資料中心容量。
-
Industry ENA Model Four Times Bigger and a Chip to Match: Alibaba Bets 10 Trillion Parameters on the Machine Intelligence Era
At Apsara 2026 Eddie Wu unveiled a 5–10T-parameter Qwen successor, the Zhenwu V900 'China's most powerful' AI chip, and a 20GW data-centre buildout — Alibaba's full-stack answer to the post-Nvidia era.
-
Industry 中十兆參數的豪賭：阿里雲棲大會揭曉 Zhenwu V900 晶片與 20GW 資料中心藍圖
吳泳銘在杭州雲棲大會宣布將訓練 5–10 兆參數的新一代 Qwen 模型、發表中國最強 AI 晶片「鎮武 V900」，並承諾 2032 年前全球資料中心容量突破 20GW——一場橫跨模型、晶片與電力的全方位下注。
-
Industry ENTwo Chips for the Agentic Age: Snapdragon Summit 2026 Opens in Maui With 2nm Flagships and AI in the GPU
Qualcomm's ninth Snapdragon Summit kicks off September 22 in Maui under the banner 'Snapdragon for the Agentic Age' — two Snapdragon 8 Elite Gen 6 flagships built on TSMC 2nm, a 5GHz Oryon CPU with FlexCache, and Adreno Neural Fusion putting dedicated AI matrix cores inside the GPU.
-
Industry 中兩顆晶片迎接代理時代：Snapdragon Summit 2026 於茂宜島登場，2nm 旗艦與 GPU 內建 AI 運算
Qualcomm 第九屆 Snapdragon Summit 於 9 月 22 日在茂宜島開幕，主題「Snapdragon for the Agentic Age」——兩款台積電 2nm 製程的 Snapdragon 8 Elite Gen 6 旗艦、時脈突破 5GHz 並搭載 FlexCache 的 Oryon CPU，以及把 AI 矩陣運算核心直接放進 GPU 的 Adreno Neural Fusion。
-
Industry ENNVIDIA Launches DSX Ready: A Seal of Approval for the Batteries and Cooling Gear That Keep AI Factories Alive
NVIDIA's new DSX Ready program qualifies third-party power and cooling products against its AI factory reference designs, starting with battery energy storage from Tesla, Hitachi Energy and LG Energy Solution and liquid-cooling units from Vertiv, LiquidStack and LG Electronics.
-
Industry 中NVIDIA 推出 DSX Ready：為支撐 AI 工廠運作的電池與散熱設備掛上認證標章
NVIDIA 新推出的 DSX Ready 認證計畫，依其 AI 工廠參考設計檢驗第三方電力與散熱產品，首批通過名單包括 Tesla、Hitachi Energy 與 LG Energy Solution 的電池儲能系統，以及 Vertiv、LiquidStack 與 LG Electronics 的冷卻液分配單元。
-
Industry ENAMD Joins the Trillion-Dollar Club — and Just Crossed the $600 Bar That Unlocks Its Biggest Deals
AMD became the fourth US chipmaker to top $1 trillion in market cap, riding a 185% rally, OpenAI and Meta warrant deals, and a broad AI trade revival.
-
Industry 中AMD 加入兆美元俱樂部——並一舉跨過解鎖最大交易的 600 美元門檻
AMD 成為美國第四家市值突破 1 兆美元的晶片商，受惠於全年大漲 185%、OpenAI 與 Meta 的認股權證交易，以及 AI 交易熱潮的全面回歸。
-
Industry ENNo GPUs Required: Meta's Muse Agent Sparks a 13% Arm Rally and Reprices the Entire CPU Trade
Meta's Muse personal agent lit a fire under processor stocks on September 21, with Arm up 13%, Intel up 12% and AMD up 9% as investors bet agentic AI shifts compute demand back to CPUs.
-
Industry 中不只靠 GPU：Meta 的 Muse 智慧代理引爆 Arm 13% 大漲，CPU 類股全面重新定價
Meta 的個人 AI 代理 Muse 在 9 月 21 日點燃處理器類股行情，Arm 大漲 13%、Intel 上漲 12%、AMD 攀升 9%，投資人押注代理式 AI 將讓運算需求回流 CPU。
-
Tools ENEditor's Choice, With Caveats: Tom's Hardware's M5 Ultra Mac Studio Review Beats DGX Spark at Local LLMs
Independent benchmarks are in: Apple's $12,299 M5 Ultra Mac Studio posts prompt-processing speeds faster than Nvidia's DGX Spark and nearly 4x its token throughput — but the win comes with soldered RAM and a 16-18 week backlog.
-
Tools 中編輯推薦但有但書：Tom's Hardware 實測 M5 Ultra Mac Studio，本地 LLM 推論擊敗 DGX Spark
獨立評測出爐：Apple 要價 12,299 美元的 M5 Ultra Mac Studio，prompt 處理速度快過 Nvidia DGX Spark，token 產出吞吐量接近其 4 倍——但代價是記憶體焊死、出貨得等 16-18 週。
-
Industry ENFrom Nvidia to Ascend: DeepSeek's Founder Makes Training on Huawei Chips a Strategic Bet
The Information reports DeepSeek founder Liang Wenfeng told investors that training on Huawei chips is one of the lab's biggest strategic bets, with Huawei expected to begin delivering training chips as early as Q4 2026 — a decisive shift for a lab whose frontier models were built on Nvidia silicon.
-
Industry 中從輝達到昇騰：DeepSeek 創辦人把「用華為晶片訓練模型」變成一場戰略豪賭
The Information 報導，DeepSeek 創辦人梁文鋒向投資人表示，改用華為晶片訓練模型是公司最大的戰略押注之一，華為最快將於 2026 年第四季開始交付訓練晶片——對一個此前完全建立在輝達硬體之上的實驗室而言，這是一次決定性的轉向。
-
Industry EN11.95 Nanometers Without EUV: CXMT's G5 DRAM Node Enters Mass Production
At the World Manufacturing Convention in Hefei, CXMT announced its fifth-generation G5 DRAM platform is in mass production — an 11.95nm half-pitch achieved with SAQP quadruple patterning instead of EUV, 24Gb LPDDR5X parts, 45:1 capacitors, and 50% more dies per wafer.
-
Industry 中沒有 EUV 的 11.95 奈米：CXMT G5 DRAM 平台進入量產
在合肥世界製造業大會上，中國記憶體大廠長鑫存儲（CXMT）宣布第五世代 G5 DRAM 平台進入量產——以 SAQP 四重曝活在 DUV 微影下達成 11.95 奈米半間距、24Gb LPDDR5X 產品、45:1 電容深寬比，以及每片晶圓多出 50% 的裸晶產出。
-
Industry ENFrom HBM Kings to Kingmakers: SK hynix Launches 'SK hynix Ventures' in Silicon Valley
SK hynix formally unveiled its corporate venture capital brand 'SK hynix Ventures' on September 18, kicking off with a CEO-led 'Ventures Day' in Silicon Valley and widening its investment lens from semiconductors to the whole AI stack.
-
Industry 中從 HBM 霸主到造王者：SK hynix 在矽谷成立「SK hynix Ventures」
SK hynix 於 9 月 18 日正式發布企業創投品牌「SK hynix Ventures」，以執行長郭魯正領軍的矽谷 Ventures Day 揭開序幕，投資視野從半導體一口氣擴大到整個 AI 技術棧。
-
Industry ENTwo Chips or Four: Apple Weighs a 2029 Return to the Server Market with M8 Ultra and NVIDIA's NVLink
Apple is reportedly designing enterprise AI servers with two or four M8 Ultra chips linked by NVIDIA's NVLink Fusion — its first server product since Xserve, aimed squarely at AI inference and EU data sovereignty.
-
Industry 中兩顆還是四顆：Apple 傳以 M8 Ultra 搭配 NVIDIA NVLink，評估 2029 年重返伺服器市場
Apple 據報正在設計搭載兩顆或四顆 M8 Ultra 晶片、以 NVIDIA NVLink Fusion 互連的企業級 AI 伺服器——這將是 Xserve 之後 Apple 首款伺服器產品，瞄準 AI 推論與歐盟資料主權需求。
-
Industry ENGolden Opportunity: Modi Opens SEMICON India 2026 as ISM 2.0 Doubles to $13.5B — Applied Materials Pledges $5B, Lam Builds India's First Component Plant
At SEMICON India 2026 in New Delhi, PM Modi pitched India as a trusted chip hub amid weaponized supply chains: the Semiconductor Mission's Phase 2 raises the outlay to $13.5B, Applied Materials committed $5B over a decade with a 140-acre research park, and Lam Research will invest ₹10,000 crore in India's first silicon component plant.
-
Industry 中黃金機遇：莫迪開幕 SEMICON India 2026，半導體任務 2.0 加碼至 135 億美元——應用材料承諾 50 億美元、Lam Research 建印度首座零組件廠
在新德里開幕的 SEMICON India 2026 上，印度總理莫迪將印度定位為「可信賴」的晶片樞紐：半導體任務第二階段預算提高至 135 億美元，應用材料承諾十年投資 50 億美元並興建 140 英畝研究園區，Lam Research 將投入 1,000 億盧比建立印度首座矽零組件製造廠。
-
Industry ENFifteen Thousand Chips, One Machine: Huawei Unveils the Atlas 960 SuperPoD and Pulls Its Ascend Roadmap Forward
At Huawei Connect 2026 in Shanghai, Huawei shipped the Atlas 960 SuperPoD — 15,488 Ascend NPUs across 220 cabinets fused by UnifiedBus and new near-packaged optics — and announced that its Ascend 960DT training chip will arrive three quarters early, in Q1 2027.
-
Industry 中一萬五千顆晶片、一台機器：華為發表 Atlas 960 SuperPoD，Ascend 路線圖提前三季
在上海舉行的 Huawei Connect 2026 上，華為發表了搭載 15,488 顆 Ascend NPU、橫跨 220 個機櫃的 Atlas 960 SuperPoD，首度導入近封裝光學（NPO）與升級版 UnifiedBus 互連技術，並宣布訓練晶片 Ascend 960DT 提前三季、將於 2027 年第一季就緒。
-
Models ENThe Benchmark Grows Teeth: MLPerf Inference v6.1 Debuts Vera Rubin, a 512-GPU AMD Cluster, and the First Agentic Tests
MLCommons' biggest inference round ever: 30 submitters, 486 results, 5.7x per-accelerator gains in a year, NVIDIA Vera Rubin NVL72's first peer-reviewed numbers at up to 3.7x GB300, and new End-to-End RAG and Edge Agentic benchmarks.
-
Models 中基準測試長出牙齒：MLPerf Inference v6.1 發表 Vera Rubin 首批受審數據、史上最大 512 GPU 叢集與首度登場的 Agentic 測試
MLCommons 史上最大規模推理基準測試回合：30 家機構提交 486 項結果、單加速器效能一年提升 5.7 倍、NVIDIA Vera Rubin NVL72 首批受同儕審查數據最高達 GB300 的 3.7 倍，並新增 End-to-End RAG 與 Edge Agentic 兩項新測試。
-
Industry ENArke in H1 2027, Astrid by Year-End: Meta Locks In Its Custom-Silicon Escape From Nvidia's Bill
Bloomberg's interview inside Meta's custom silicon program reveals MTIA 450 'Arke' enters data centers in H1 2027, MTIA 500 'Astrid' follows by year-end, over a gigawatt is already committed — and the cancelled Olympus training chip shows exactly where the strategy bends.
-
Industry 中Arke 上半年度進場、Astrid 年底接棒：Meta 把自研晶片逃離 Nvidia 帳單的時程表定案了
Bloomberg 專訪揭露 Meta 自研矽晶片計畫時程：第三代 MTIA 450「Arke」將於 2027 上半年進入資料中心量產，第四代 MTIA 500「Astrid」年底接棒，承諾部署量已超過 1 GW——而被砍掉的 Olympus 訓練晶片，正好說明這套策略的底線在哪裡。
-
Industry ENMemory Comes Home: SK Hynix in Exploratory Talks to Make Chips Inside Intel's Ohio Fabs
Reuters reports SK Hynix is in early-stage talks to lease capacity at Intel's delayed Ohio megafab — a deal that would bring advanced memory manufacturing to US soil for the first time and mark the first time Hynix outsources memory production to a rival.
-
Industry EN記憶體回家了：SK 海力士洽談進駐英特爾俄亥俄廠，美國本土先進記憶體製造可望寫下首例
路透社報導，SK 海力士正與英特爾進行早期洽談，考慮租用俄亥俄州大型晶圓廠產能——這將是美國本土首次生產先進記憶體，也是海力士首度把記憶體生產外包給競爭對手。
-
Tools ENThe Phone Becomes the Agent: Nubia's NaviX Ultra Goes on Sale Today With Doubao at Its Core
ZTE's Nubia launches the NaviX Ultra today — the world's first mass-produced AI-agent smartphone, with ByteDance's Doubao assistant woven into the OS, on-device inference, MCP/A2A app integration, and 380,000+ pre-orders.
-
Tools 中手機本身就是智能體：努比亞 NaviX Ultra 今日開賣，豆包成為它的核心
中興努比亞今日發表 NaviX Ultra——全球首款量產 AI 智能體手機，字節跳動豆包助手深度整合系統層，支援端側運算、MCP/A2A 應用串接，預約量已突破 38 萬台。
-
Policy ENBoxing Out Beijing: Washington Wants a Hard Cap on Foreign Parts in AI Hardware Made in Mexico
A US proposal in USMCA talks would cap how much of an AI chip or server can be built from non-North American components — taking direct aim at the Mexico assembly route that made AI hardware Mexico's top export to the US.
-
Policy 中把北京擋在門外：華府擬對墨西哥製 AI 硬體設下零組件含量硬上限
美國在 USMCA 談判中提案，限制 AI 晶片與伺服器可使用多少非北美零組件——直接衝著讓 AI 硬體成為墨西哥對美最大出口項目的組裝路線而來。
-
Industry ENThe GPU IPO That Refuses to Stop Raising: Biren Weighs Another $1 Billion Share Sale
Fresh off a $900M July placement, Hong Kong-listed Biren Technology is sounding out investors for another ~$1B raise — even as Zhipu's $5B sale and a 22% Hang Seng Tech slump test China's appetite for AI chip paper.
-
Industry 中不停募資的 GPU 第一股：壁仞科技傳將再融資 10 億美元
7 月才完成 9 億美元配售的壁仞科技，據傳正接觸投資人規劃新一輪約 10 億美元融資——此時智證 50 億美元融資引發拋售，恒生科技指數今年已下跌 22%。
-
Meta ENFrom Megawatts to Tokens: Nvidia's DSX Posts First Production Numbers — 24% More Throughput on the Same Power
At AI Infra Summit, Nvidia published DSX's first real-world results: Lambda squeezed 24% more token throughput from a fixed power budget, and the Eos AI factory has answered 200+ utility demand signals without interrupting a single critical workload.
-
Industry ENEurope's Inference Play: Axelera Launches Europa, Signs AI Factory Deals Worth Tens of Millions
Dutch chipmaker Axelera AI launched its second-generation Europa chip, signed AI factory supply contracts worth tens of millions, and is chasing a $1.5 billion pipeline across Dell and Supermicro systems.
-
Industry 中歐洲的推論晶片突圍：Axelera 發表 Europa、簽下數千萬美元 AI 工廠供貨合約
荷蘭晶片新創 Axelera AI 發表第二代推論晶片 Europa，簽下價值數千萬美元的 AI 工廠供貨合約，並透過 Dell 與 Supermicro 系統搶進 15 億美元商機。
-
Industry ENMediaTek Claims the 2nm Crown: Dimensity 9600 Pro Runs 30B-Parameter Models On Your Phone
MediaTek announced the Dimensity 9600 Pro, the first flagship smartphone SoC built on TSMC's 2nm N2P process. With an all-big-core CPU, dual-NPU architecture, LPDDR6 and UFS 5.0, it promises 61% lower multi-core power and on-device models up to 30B parameters — a bet that agentic AI, not camera megapixels, is the next smartphone battleground.
-
Industry 中聯發科搶下 2nm 頭香：Dimensity 9600 Pro 讓 300 億參數模型直接跑在手機上
聯發科發布 Dimensity 9600 Pro，業界首款採用台積電 2nm N2P 製程的旗艦手機 SoC。全大核 CPU、雙 NPU 架構、LPDDR6 與 UFS 5.0，多核功耗降低 61%，並支援最高 300 億參數的端側模型——這是一場豪賭：賭的是下一輪手機戰場的主角是 Agentic AI，而不是相機畫素。
-
Industry ENFour Weeks to One: Empyrean's AI Agents Slash Chip Design Time and Open a New Front in China's EDA Push
China's top EDA firm Empyrean says agentic AI cut a circuit-layout task from four weeks to one, as Beijing's MIIT action plan and the emerging Tau Law hand homegrown chip-design tools a rare opening.
-
Industry 中從四週到一週：華大九天 AI 代理改寫晶片設計時程，中國 EDA 迎來雙重機會之窗
中國最大 EDA 廠華大九天宣稱，AI 代理將電路布局任務從四週壓縮到一週；配合工信部 AI 軟體行動計畫與 Tau 定律開啟的新需求，本土晶片設計工具正迎來難得的追赶窗口。
-
Tools ENEight Devices, Five OEMs, One New OS: Google's Googlebook Showcase Takes Manhattan Today
Today in New York, Google finally pulls back the curtain on Googlebook — the Android-17-native laptop platform with a dedicated Gemini key, a mandatory Glowbar, and roughly eight devices from Acer, ASUS, Dell, HP and Lenovo. Here's everything confirmed before the embargo lifts.
-
Tools 中八款裝置、五家 OEM、一套新作業系統：Google Googlebook 發表會今天登陸紐約
Google 今天在紐約正式揭開 Googlebook 的面紗——以 Android 17 原生打造的筆電平台，配備專屬 Gemini 實體鍵、強制規格 Glowbar 燈條，以及來自 Acer、ASUS、Dell、HP 與 Lenovo 的約八款新機。在媒體禁令解除前，這是所有已確認資訊的總整理。
-
Models EN7x Per Megawatt: SemiAnalysis Catches Nvidia Sandbagging Vera Rubin's Real Inference Power
SemiAnalysis's first verified AgentX results for Vera Rubin NVL72 show up to 7x better token throughput per megawatt than Blackwell on DeepSeek V4 Pro — more than double Jensen Huang's 3x GTC claim — worth ~$149.9B in modeled annual profit per gigawatt.
-
Models 中每百萬瓦 7 倍：SemiAnalysis 實測揪出 Nvidia 低報 Vera Rubin 真實推論效能
SemiAnalysis 首批經驗證的 Vera Rubin NVL72 AgentX 實測出爐：在 DeepSeek V4 Pro 上，每百萬瓦 token 產量較 Blackwell 高達 7 倍——遠超黃仁勳 GTC 宣稱的 3 倍——換算每吉瓦年利潤約 1,499 億美元。
-
Industry ENThe Fabric Learns to Compute: Cornelis Unveils Active Compute Fabric, a $205M Raise and a Qualcomm Alliance
Cornelis Networks enters scale-up networking with an open, programmable fabric backed by $205M in new funding and a Qualcomm alliance — aiming at the ~50% of GPU hours giant AI clusters lose to waiting for data.
-
Industry 中讓網路織網學會運算：Cornelis 發表 Active Compute Fabric、2.05 億美元融資與 Qualcomm 結盟
Cornelis Networks 以開放、可編程的網路架構進軍 scale-up 連網市場，搭配 2.05 億美元融資與 Qualcomm 合作——瞄準大型 AI 叢集中約半數 GPU 時數浪費在等待資料的痛點。
-
Industry ENFive Years of Power Bills Up Front? Samsung and SK Hynix Just Told KEPCO No
Korea's debt-laden state utility asked its two biggest chipmakers to prepay 25 trillion won ($18.4B) of electricity bills to fund grid upgrades for the Yongin and Honam chip clusters. Both said no — leaving KEPCO's 210 trillion won debt mountain and Korea's AI power crunch unresolved.
-
Industry 中預繳五年電費？三星與 SK 海力士正式拒絕 KEPCO 的 25 兆韓元提案
負債累累的韓國電力公社要求兩大晶片廠預繳 25 兆韓元（約 184 億美元）電費，為龍仁與湖南晶片聚落的電網升級籌資。兩家都說不——KEPCO 的 210 兆韓元債務大山與韓國的 AI 電力缺口，依然無解。
-
Industry EN172 Million Shares at HK$32.96: Inside Ligent's $727M Hong Kong IPO and the Optical Land Grab Fueling AI Data Centers
Chinese optical-module maker Ligent Technologies prices its Hong Kong IPO at HK$5.7B (~$727M) with a Sept 22 debut, betting that GPU-to-GPU interconnect demand — 70% of its revenue — keeps compounding as AI clusters scale.
-
Industry EN1.72 億股、每股 32.96 港元：Ligent 7.27 億美元香港 IPO 與 AI 資料中心背後的光學軍備競賽
中國光學模組廠 Ligent Technologies 以 57 億港元（約 7.27 億美元）在香港 IPO 定價，預計 9 月 22 日掛牌，押注 GPU 互連需求持續爆發——這塊業務占其營收七成。
-
Industry ENSelling First, Asking Later: Asia's AI Stocks Crash as the Industry's Own CEOs Call for a Slowdown
SoftBank fell 13.2% and Kioxia 9.8% at Monday's open in Tokyo after Amodei, Altman and Musk backed a frontier development pause — the first time AI's biggest bulls have moved the market against themselves.
-
Industry 中先賣再說：AI 巨頭自己喊煞車，亞洲 AI 概念股應聲重挫
Amodei、Altman 與 Musk 表態支持放緩前沿模型開發後，SoftBank 週一開盤暴跌 13.2%、Kioxia 重挫 9.8%——這是 AI 多頭首度親手把自己的行情賣下去。
-
Industry ENFolding Silicon Instead of Shrinking It: Huawei's Kirin 9050 Pro Puts LogicFolding on Sale
Huawei's Kirin 9050 Pro is the first mass-market chip built on LogicFolding — stacking logic in vertical tiers to lift transistor density 55% without EUV lithography. It shipped inside the Mate XT 2 tri-fold this weekend.
-
Industry 中不微縮、改折疊：華為 Kirin 9050 Pro 讓 LogicFolding 晶片首次量產上街
華為 Kirin 9050 Pro 是首款採用 LogicFolding 架構的量產晶片——透過垂直堆疊邏輯層，在完全不依賴 EUV 微影技術下將電晶體密度提升 55%，本週末隨 Mate XT 2 三折機正式開賣。
-
Policy ENThirty Years for a Stolen Blueprint: South Korea's Rewritten Espionage Law Takes Effect Tomorrow
On September 13, South Korea's first espionage-law rewrite in 73 years takes effect — extending spying charges beyond North Korea to any foreign beneficiary, with courts empowered to impose up to 30 years for leaking chip technology. The target is China's recruitment of Samsung and SK Hynix engineers.
-
Policy 中竊取一張藍圖，坐牢三十年：南韓 73 年來首次改寫的間諜法明天生效
9 月 13 日，南韓 73 年來首次大幅改寫的刑法第 98 條之 2 正式生效——間諜罪不再限於北韓受益，擴及任何外國或同等組織，法院最高可判 30 年。目標直指中國對三星與 SK 海力士工程師的挖角行動。
-
Industry ENNo HBM, No Problem: Positron Raises $875M to Scale Its Memory-First Inference Bet
Positron's $875M Series C at a $5B valuation fully funds its Asimov chip tapeout on TSMC N3P and the Titan system targeting 16-trillion-parameter models — all without scarce HBM.
-
Industry 中不用 HBM 也行：Positron 融資 8.75 億美元，全力押注「記憶體優先」推理架構
Positron 以 50 億美元估值完成 8.75 億美元 C 輪融資，資金將全額支應 Asimov 晶片以台積電 N3P 製程投片，以及可服務 16 兆參數模型的 Titan 系統——全程不依賴稀缺的 HBM。
-
Industry ENNT$514.8 Billion in One Month: TSMC's August Revenue Record Shows the AI Boom Is Still Accelerating
TSMC booked NT$514.8 billion (US$16.35 billion) in August revenue, up 53.3% year-over-year — its first month ever above NT$500 billion, with AI demand keeping every advanced node sold out.
-
Industry 中單月營收 5,148 億元：台積電八月營收破紀錄，證明 AI 熱潮仍在加速
台積電八月營收達 5,148.06 億新台幣（約 163.5 億美元），年增 53.3%，史上首度單月突破 5,000 億——AI 需求讓所有先進製程產能持續滿載。
-
Industry ENFrom 142 to 410 Yuan in One Bell: Enflame's Debut Triples the Last AI Dragon's valuation
Enflame opened 188% above its IPO price on the Shanghai STAR Market, instantly valuing Tencent's captive AI-chip maker at 176.4 billion yuan — and completing the public-market coronation of China's four AI-chip dragons.
-
Industry 中一聲鐘響從 142 到 410 元：燧原科技掛牌暴漲，最後一條 AI 晶片小龍完成三倍市值加冕
燧原科技在上海科創板掛牌首日以 410 元開盤、較發行價大漲 188%，騰訊加持的這家 AI 晶片商市值一舉衝上 1,764 億人民幣 — 中國四大 AI 晶片「小龍」的資本市場加冕至此全數完成。
-
Industry ENOne Company, 26 New Gigawatts: Microsoft's 38GW Data-Center Roadmap Would Out-Power New York State
Bloomberg reports Microsoft plans to grow its owned-and-leased data-center capacity from ~12GW today to more than 38GW by 2032 — a single-company buildout that would eclipse New York state's peak electricity consumption, with a third of it on AI-specific silicon.
-
Industry 中一家公司，26 座新百萬瓩：微軟 38GW 資料中心藍圖，用電量將超越紐約州
彭博報導，微軟計畫將自有與租賃資料中心容量從今日約 12GW 擴增至 2032 年的 38GW 以上——這場單一公司的建設狂潮，總用電規模將超越紐約州夏季用電峰值，其中三分之一是 AI 專用晶片叢集。
-
Industry ENCopper Is the New Bottleneck: Ayar Labs Adds $150M to Hit $650M for 2026 as Co-Packaged Optics Go Manufacturing-Ready
The San Jose photonics company's co-packaged optics replace power-hungry copper interconnects between AI racks; Wiwynn is now a strategic investor and a new Bengaluru design center targets high-volume production of the copper-replacement technology.
-
Industry ENRaptor Meets the AI Factory: d-Matrix Plugs Its 3D-DRAM Inference Chips Into NVIDIA's Racks
d-Matrix will ship its Raptor XPUs inside NVIDIA MGX racks via NVLink Fusion, pairing memory-centric inference silicon with Vera CPUs and Spectrum-X networking for premium low-latency token services from Q4 2027.
-
Industry 中Raptor 住進 AI 工廠：d-Matrix 的 3D DRAM 推論晶片正式插進 NVIDIA 機架
d-Matrix 將透過 NVLink Fusion 把 Raptor XPU 整合進 NVIDIA MGX 機架，與 Vera CPU、Spectrum-X 網路搭配，自 2027 年第四季起提供高階低延遲 token 服務。
-
Industry ENNo HBM, No Problem: Positron Closes $875M at $5B to Ship Its Memory-First Asimov Chip
The Reno AI-inference startup's Series C was co-led by NEA, Atreides and Valor; its Asimov accelerator swaps scarce HBM for cheap DDR5 and claims Nvidia-class inference at a fraction of the power.
-
Industry 中不用 HBM 也行：Positron 以 50 億美元估值完成 8.75 億美元 C 輪，量產記憶體優先的 Asimov 晶片
這家位於美國雷諾的 AI 推論新創由 NEA、Atreides 與 Valor 共同領投；其 Asimov 加速器捨棄稀缺的 HBM、改用廉價 DDR5，宣稱以遠低於 Nvidia 的功耗提供同等推論效能。
-
Industry ENTwo Gigawatts Down Under: NVIDIA and Eight Australian Partners Bet Big on Sovereign AI
NVIDIA has enlisted eight Australian cloud and data center operators — Firmus, Sharon AI, NEXTDC, AirTrunk, CDC, IREN, Megaport and ResetData — to build up to 2 GW of AI factory capacity by 2027, even as grid queues threaten to leave some partners waiting for power.
-
Industry 中南半球的兩 GW 藍圖：NVIDIA 攜手八家澳洲夥伴豪賭主權 AI
NVIDIA 宣布與 Firmus、Sharon AI、NEXTDC、AirTrunk、CDC、IREN、Megaport、ResetData 等八家澳洲雲端與資料中心業者合作，目標 2027 年前建成最高 2 GW 的 AI 工廠產能，但電網排隊問題可能讓部分夥伴等不到電。
-
Industry EN20-50% Price Hikes Across the Board: HBM Shortage Squeezes China's AI Chipmakers
Reuters reports Huawei, Cambricon, MetaX and Iluvatar CoreX raised AI accelerator prices 20-50% in two months as the HBM shortage bites — Huawei's Ascend 950DT now quotes above 250,000 yuan.
-
Industry 中全系漲價 20-50%：HBM 短缺重擊中國 AI 晶片商
路透社報導，華為、寒武紀、摩爾線程與壁仞等中國 AI 晶片商在兩個月內將加速卡價格調漲 20-50%，華為 Ascend 950DT 報價已突破 25 萬人民幣。
-
Policy ENAcquisition by Contract: DOJ Probes Whether Nvidia's $20B Groq Deal Was Built to Dodge Antitrust Review
The Justice Department is formally investigating whether Nvidia structured its $17-20B Groq licensing deal as a license-plus-acquihire to escape merger review — a test case for how AI giants absorb rivals without ever 'acquiring' them.
-
Policy 中契約併購：美國司法部調查 Nvidia 200 億美元 Groq 授權案是否為規避反壟斷審查而設計
美國司法部正式調查 Nvidia 是否將 170-200 億美元的 Groq 授權案結構設計為「授權＋團隊挖角」，以規避併購審查——這是 AI 巨頭「不併購的併購」手法所面臨最嚴峻的一次法律考驗。
- Models EN
No EUV Required: Kepler Computing Exits Stealth With $468M and a Ferroelectric Fix for the AI Memory Crisis
The San Jose startup says 3D stacking plus a proprietary ferroelectric composite can lift HBM and SRAM density on legacy 28nm fabs — no EUV lithography, no $20B megafab — with first HBM samples due this year.
- Models EN
不需要 EUV：Kepler Computing 帶著 4.68 億美元走出隱身模式，用鐵電材料破解 AI 記憶體危機
這家聖荷西新創宣稱，3D 堆疊加上專有鐵電複合材料，能在既有的 28nm 晶圓廠提升 HBM 與 SRAM 密度——不需要 EUV 微影設備，也不需要 200 億美元的超級工廠，首批 HBM 樣品今年出貨。
-
Industry ENCheaper Than Open Source: OpenAI Pitches AI for Chip Design and Claims It Undercuts China on Price
At Goldman Sachs' Communacopia conference, CFO Sarah Friar said OpenAI's 80% Luna price cut drove 10x usage — and that its API can now beat Chinese open-source models on total cost in specialized verticals like chip design and life sciences.
-
Industry 中比開源還便宜？OpenAI 把 AI 推進晶片設計，宣稱在價格上擊敗中國對手
在高盛 Communacopia 科技會議上，OpenAI 財務長 Sarah Friar 表示 Luna 降價 80% 帶來十倍用量成長，並宣稱在晶片設計、生命科學等專業領域，其 API 的總持有成本已低於中國開源模型。
-
Meta EN100,000 Homegrown GPUs: JD Cloud and Moore Threads to Build China's Largest Domestic-Chip AI Cluster
At its Global Technology Explorers Conference, JD Cloud committed to a 100,000-GPU intelligent-computing cluster built entirely on Moore Threads' domestic full-function GPUs — the first time Chinese silicon reaches 100K scale inside a top AI cloud provider.
-
Meta EN十萬張國產 GPU：京東雲攜手摩爾線程，打造中國最大全國產晶片 AI 叢集
在 2026 京東全球科技探索者大會上，京東雲宣布擬建設十萬卡全功能 GPU 智算叢集，算力底座全面採用摩爾線程國產 GPU——這是國產晶片首次進入頭部 AI 雲廠商的十萬卡級核心叢集。
-
Tools ENIt's Official: Apple Ships the iPhone Duo — a $1,999 Titanium Foldable With a Near-Invisible Crease
At its 'Surprise and Shine' event, Apple finally unveiled the iPhone Duo: a 7.6-inch book-style foldable with an A20 Pro chip, a titanium-and-ceramic hinge, and a $1,999 price tag — the most expensive iPhone ever, shipping October 23.
-
Tools 中正式登場：Apple 發表 iPhone Duo — 1,999 美元的鈦合金摺疊機，摺痕近乎隱形
在「Surprise and Shine」發表會上，Apple 終於揭曉 iPhone Duo：7.6 吋書本式摺疊機，搭載 A20 Pro 晶片、鈦合金與陶瓷纖維鉸鏈，售價 1,999 美元起 — 這是史上最貴的 iPhone，10 月 23 日上市。
-
Industry ENFrom Memory Buyer to Co-Designer: OpenAI Confirms It Is Building Next-Generation Chips With Samsung
At a Seoul press conference, OpenAI Korea chief Harrison Kim said joint research and production of next-generation chips with Samsung is one of the lab's biggest areas of progress — signaling OpenAI's custom silicon roadmap may soon run on two foundries, not one.
-
Industry 中從記憶體買家到共同設計者：OpenAI 證實與三星共同研發下一代晶片
OpenAI 韓國負責人 Harrison Kim 在首爾記者會上表示，與三星就下一代晶片的共同研究與生產已取得重大進展——意味著 OpenAI 的自研晶片路線圖，未來可能同時由兩座晶圓廠代工。
-
Industry EN$1.35 Billion for Milliwatts: Analog Devices Buys Alif Semiconductor to Put AI on a Battery
ADI will pay $1.35B in cash (up to $1.55B with earnouts) for Alif Semiconductor, whose AI-native microcontrollers and fusion processors run neural networks on milliwatts — the edge-computing counterweight to gigawatt data centers.
-
Industry 中13.5 億美元買毫瓦：Analog Devices 收購 Alif Semiconductor，要把 AI 裝進電池裝置
ADI 以 13.5 億美元現金（含業績獎金最高 15.5 億美元）收購 Alif Semiconductor，其 AI 原生微控制器與融合處理器能用毫瓦級功耗跑神經網路——這是 GW 級資料中心之外的另一端 AI 經濟。
-
Industry EN128 Cores, 3.8 GHz, LPDDR6: Arm's Neoverse CSS N4 Is the Semi-Custom Blueprint for the Agentic AI Data Center
Arm's Neoverse CSS N4 doubles the per-die core ceiling to 128, adds LPDDR6 and PCIe Gen 7, and claims 2x socket performance and 1.25x performance per watt over CSS N3 — a semi-custom on-ramp aimed squarely at agentic AI's CPU-hungry workloads.
-
Industry 中128 核心、3.8 GHz、LPDDR6：Arm Neoverse CSS N4 為 Agentic AI 資料中心獻上半客製化藍圖
Arm 的 Neoverse CSS N4 將單晶片核心數上限翻倍至 128，導入 LPDDR6 與 PCIe Gen 7，並宣稱較 CSS N3 提供最高 2 倍插槽效能與 1.25 倍每瓦效能——一套直接瞄準 Agentic AI 吃重 CPU 工作負載的半客製化捷徑。
-
Industry ENSamsung Puts Mistral AI Inside the Fab: On-Premises LLMs for Every Chip Plant
Announced during the Korea–France summit in Paris, Samsung's strategic partnership with Mistral AI brings Mistral Large and customized on-premises models into semiconductor design and manufacturing — defect detection, equipment optimization, and yield stabilization across memory, logic, and foundry.
-
Industry 中三星把 Mistral AI 裝進晶圓廠：全面導入在地化 LLM 的智慧製造布局
在巴黎舉行的韓法高峰會上，三星電子宣布與 Mistral AI 建立策略夥伴關係，將 Mistral Large 與客製化在地部署模型導入半導體設計與製造——涵蓋記憶體、邏輯與晶圓代工的缺陷檢測、設備優化與良率穩定。
-
Policy EN9,800 EFLOPS by 2030: China's ¥3.8 Trillion Compute Plan Turns Ambition Into a Procurement Curve
MIIT's 15th Five-Year Plan for ICT quadruples China's intelligent computing target to 9,800 EFLOPS by 2030, backed by ¥3.8 trillion in cumulative information infrastructure investment and 100,000-card clusters.
-
Policy 中2030 年 9,800 EFLOPS：中國 3.8 兆元算力藍圖，把雄心變成一條採購曲線
工信部「十五五」資通訊規劃將中國智慧算力目標定為 2030 年 9,800 EFLOPS（現況四倍以上），配套 3.8 兆人民幣資訊基礎設施累計投資與十萬卡叢集部署。
-
Industry EN40,000 Square Feet of Cleanroom for the Post-Silicon Era: HCLTech Opens a ₹185 Crore Advanced Semiconductor Lab in Bengaluru
HCLTech launched a 40,000 sq ft Advanced Semiconductor Lab in Bengaluru with ₹185 crore of investment, betting that India's next chip opportunity is not fabs but the unglamorous engineering that happens after silicon comes back from the foundry.
-
Industry 中為後矽晶世代而建的 4 萬平方英尺無塵室：HCLTech 於班加羅爾啟用 185 億盧比先進半導體實驗室
HCLTech 於班加羅爾啟用佔地 4 萬平方英尺、投資額 185 億盧比的先進半導體實驗室，押注印度下一波晶片機會不在晶圓廠，而在晶片從代工廠回來之後那些不起眼卻關鍵的工程環節。
-
Policy EN9,800 EFLOPS by 2030: China's Five-Year Plan Commits $532 Billion to Quadrupling AI Compute
China's MIIT has set a target of 9,800 EFLOPS of intelligent computing capacity by 2030 — more than four times today's level — backed by 3.8 trillion yuan in planned information infrastructure investment.
-
Policy 中2030 年 9,800 EFLOPS：中國五年計劃投入 5,320 億美元，要將 AI 算力翻四倍
中國工信部發布「十五五」資訊通信業發展規劃，設定 2030 年智能算力規模達 9,800 EFLOPS 的目標——超過目前水準的四倍——並規劃 3.8 兆元人民幣的資訊基礎設施累計投資。
-
Industry ENToo Complex for Humans, Solvable for Machines: Arm's Rene Haas Says AI Will Cure Cancer
Arm's CEO tells the BBC that AI will find cancer cures humans cannot, humanoid robots arrive within five years, and the real bottleneck is chips — with Meta's $2bn Arm AGI demand 'off the charts'.
-
Industry 中人類解不開、機器能解開：Arm 執行長 Haas 宣稱 AI 將治癒癌症
Arm 執行長向 BBC 表示，AI 將找出人類無法發現的癌症療法、五年內人形機器人將普及，而真正的瓶頸是晶片——Meta 的 Arm AGI 晶片需求「爆表」，已超過 20 億美元。
-
Industry ENAI Will Cure Cancer, Says Arm's CEO — If the Chip Supply Holds
Arm CEO Rene Haas tells the BBC that AI will crack cancer's complexity in our lifetimes and put humanoid robots to work within five years — but warns chip shortages are the real bottleneck.
-
Industry 中Arm 執行長：AI 將治癒癌症——前提是晶片供應跟得上
Arm 執行長 Rene Haas 接受 BBC 專訪時表示，AI 有生之年將破解癌症的複雜性、五年內人形機器人將廣泛應用，但晶片短缺才是真正的瓶頸。
-
Industry EN25 Million Shares, $60 Billion in Chips: Qualcomm and Amazon Lock In a Multi-Generation AI Data Center Pact
Qualcomm will co-design custom AI inference silicon and 1.6T optical connectivity for AWS, backed by a 25M-share warrant tied to up to $60B in purchases — QCOM jumped as much as 10%.
-
Industry 中2,500 萬股認股權證、600 億美元晶片採購：Qualcomm 與 Amazon 簽下多世代 AI 資料中心合作
Qualcomm 將為 AWS 共同設計客製化 AI 推論晶片與 1.6T 光互連，並向 Amazon 發行與最高 600 億美元採購掛鉤的 2,500 萬股認股權證，QCOM 股價大漲近 10%。
-
Industry ENOne Mask, Four Times the Silicon: ASML and TSMC Put the Whole Chip Industry on a 12-Inch Roadmap
Announced ahead of SPIE BACUS, ASML and TSMC will lead the semiconductor industry from 6-inch to 12-inch photomasks for High-NA EUV — a 2031 pilot line, 2033 system readiness, Samsung's 2028 DRAM debut, and the eventual end of stitching for giant AI dies.
-
Tools EN1,024 INT8 MACs Inside the Shader Cores: Arm's Mali G2-Ultra NX Is the First AI-Native Mobile GPU
Arm's CSS for Mobile 2 embeds neural accelerators directly inside the Mali G2-Ultra NX shader cores — 1,024 INT8 MACs per clock, a seven-generation ISA overhaul, and DLSS-style neural upscaling promising 4x efficiency, targeting 2027 flagships like Samsung's Galaxy S27 and Xiaomi's XRING O3.
-
Tools 中每時脈 1,024 次 INT8 運算塞進著色器核心：Arm Mali G2-Ultra NX 是首款 AI 原生行動 GPU
Arm 的 CSS for Mobile 2 把神經網路加速器直接嵌入 Mali G2-Ultra NX 著色器核心——每時脈 1,024 次 INT8 MAC、七個世代以來最大規模的 ISA 改版，加上類 DLSS 的神經升頻技術帶來 4 倍效率，瞄準 2027 年旗艦機如 Samsung Galaxy S27 與 Xiaomi XRING O3。
-
Tools EN128 Cores on N3P and a Fifth of a Watt Saved per MIPS: Arm's Neoverse CSS N4 'Ranger' Rewrites the Semi-Custom Server Playbook
Arm's Neoverse CSS N4 'Ranger' compute subsystem scales from 8 to 128 N4 cores per die on TSMC N3P with LPDDR6, 256MB of L3 and 128 lanes of PCIe 7 — doubling socket performance over CSS N3 just as Arm announced Oracle and ByteDance's Volcano Engine are joining the AGI CPU ecosystem for agentic AI.
-
Tools 中N3P 製程塞進 128 核心、每瓦效能再省 20%：Arm Neoverse CSS N4「Ranger」改寫半客製化伺服器晶片規則
Arm 的 Neoverse CSS N4「Ranger」運算子系統可在台積電 N3P 上實現單晶片 8 至 128 個 N4 核心，支援 LPDDR6、256MB L3 快取與 128 條 PCIe 7 通道，效能較 CSS N3 翻倍；同場 Arm 更宣布 Oracle 與位元組跳動火山引擎加入 AGI CPU 生態系，為代理式 AI 鋪路。
-
Industry ENThe Broker of Last Resort: Huawei Now Bankrolls China's Lithography Startups and Matchmakes Their SMIC Deals
A Financial Times investigation reveals Huawei is funding Chinese lithography equipment makers through its Hubble arm and personally brokering their sales to SMIC — the clearest sign yet that the sanctioned giant has decided if it cannot buy the machines, it will own the industry that makes them.
-
Industry 中最後掮客現身：華為出資中國微影設備新創，還親自撮合它們與中芯的交易
英國《金融時報》調查揭露，華為正透過哈勃投資注資中國微影曝光設備廠商，並親自居中撮合這些廠商與中芯國際的採購協議——這是這家被制裁巨頭至今最明確的訊號：買不到的機器，就乾脆買下整個製造這些機器的產業。
-
Industry ENOne Flight to Beijing, Four Hundred Lost Jobs: Belgium Charges a Former BelGaN Executive With GaN Chip Espionage
Belgian federal prosecutors say a 52-year-old dual national spent three years as a senior BelGaN researcher while directing a Chinese look-alike startup — the clearest public blueprint yet of how gallium-nitride know-how left Europe's last GaN fab.
-
Industry 中一趟北京航班與四百個失去的工作：比利時以間諜罪起訴 BelGaN 前高層，氮化鎵技術外洩案曝光
比利時聯邦檢方指出，一名 52 歲雙重國籍男子一邊在 BelGaN 擔任研究高層，一邊遙控中國的「鏡像公司」——這是歐洲最後一座氮化鎵晶圓廠技術外流至今最清晰的路徑圖。
-
Industry ENSix Years in Exile, One Fold Forward: Huawei's Kirin 9050 Pro Reboots the Flagship Chip Race Without US Silicon
Huawei's Mate XT2 launches with the Kirin 9050 Pro, the first flagship SoC built on its homegrown LogicFolding 3D-stacking architecture — a 42% uplift claimed with zero US supply chain.
-
Industry 中六年晶片流放後的一步：華為 Kirin 9050 Pro 用 LogicFolding 重啟旗艦晶片競賽，完全繞開美國供應鏈
Mate XT2 三折機搭載 Kirin 9050 Pro 登場 — 這是首款採用華為自研 LogicFolding 3D 堆疊架構的旗艦 SoC，官方宣稱效能提升 42%，且整機不含任何受美國管制零組件。
-
Policy ENThe $5.6 Billion Backdoor: How Blacklisted Inspur Kept Buying Nvidia's Best Chips Through Its American Arm
An NYT investigation reveals Aivres, the California-based subsidiary of blacklisted Inspur, shipped over $3B in Blackwell servers to Southeast Asia — exposing a structural hole in US chip export controls.
-
Policy 中56 億美元的後門：被列黑的浪潮集團如何透過美國子公司持續取得 Nvidia 最強晶片
《紐約時報》調查揭露，列入實體清單的中國浪潮集團，其美國子公司 Aivres 兩年內向東南亞出口逾 30 億美元 Blackwell 伺服器，暴露美國晶片出口管制的結構性漏洞。
-
Industry ENNT$921.77 Billion in One Month: Foxconn's August Sales Confirm the AI Server Supercycle Is Still Accelerating
Foxconn's August consolidated sales jumped 51.98% YoY to NT$921.77 billion (US$29.14B), the second-highest month in its history, as cloud and networking revenue surged on AI server demand.
-
Industry 中單月營收 9,217.7 億新台幣：鴻海八月營收證實 AI 伺服器超級週期仍在加速
鴻海八月合併營收年增 51.98% 至 NT$9,217.7 億（約 291.4 億美元），創史上第二高紀錄，雲端網路產品在 AI 伺服器需求帶動下強勁成長。
-
Industry ENA VC-Led Fab Fund: Coatue and MatX Near a Multibillion-Dollar Joint Venture to Finance AI Chips
Coatue Management is in early talks to form a first-of-its-kind multibillion-dollar joint venture with AI chip startup MatX, using project-finance muscle to pre-book memory and manufacturing capacity ahead of the startup's 2027 ship date.
-
Industry 中創投當起晶圓廠金主：Coatue 與 MatX 洽談數十億美元合資公司，為 AI 晶片預訂產能與記憶體
Coatue Management 正與前 Google TPU 工程師創辦的 AI 晶片新創 MatX 洽談成立首見的數十億美元合資公司，以專案融資手法在 2027 年出貨前預訂記憶體與製造產能。
-
Industry ENThe $200 Billion Policy: Swiss Re Sizes the Insurance Bill of the AI Capex Boom
At Monte Carlo, Swiss Re Institute said AI data centres and renewables could generate ~$200B in commercial insurance premiums by 2030 — while warning that $50B campuses clustered in hail, tornado and earthquake zones concentrate risk like nothing insurers have underwritten before.
-
Industry 中2000 億美元的保單：Swiss Re 估算 AI 資本支出熱潮的保險帳單
在蒙地卡羅，Swiss Re Institute 表示 AI 資料中心與再生能源可望在 2030 年前創造約 2,000 億美元的商業保險保費——同時警告造價 500 億美元的園區群聚在冰雹、龍捲風與地震帶，風險集中度是保險業前所未見。
-
Industry ENThe Quiet Winner of IFA 2026: D-Robotics Wants to Be the Android of Home Robots
IFA's flashiest home robots — TCL's hey AiMe, Vbot's SuperDog, the xLean TR1 — all run on Sunrise AI chips from D-Robotics, the Horizon Robotics spinoff that just raised $270M to own the compute layer under embodied AI.
-
Industry 中IFA 2026 的隐形赢家：D-Robotics 想当家用机器人的「安卓」
IFA 最受瞩目的三款家用机器人——TCL hey AiMe、Vbot SuperDog、xLean TR1——全都跑在 D-Robotics 的 Sunrise AI 晶片上。这家从地平线机器人拆分出来的公司刚募得 2.7 亿美元，要拿下具身智慧的运算层。
-
Policy ENSilicon as Statecraft: WSJ Reveals the US Brokered the Armenia-Azerbaijan Peace Deal With Nvidia Chips
A Wall Street Journal investigation reveals US negotiators dangled access to Nvidia Blackwell GPUs for an Armenian AI data center to help secure the August 2025 White House peace deal — turning export-controlled silicon into instruments of diplomacy.
-
Policy 中矽晶片外交：WSJ 揭露美國以 Nvidia 晶片換取亞美尼亞—亞塞拜然和平協議
《華爾街日報》調查報導揭露，美國談判代表以輝達 Blackwell GPU 進口許可作為外交籌碼，促成 2025 年 8 月白宮簽署的亞美尼亞—亞塞拜然初步和平協議——出口管制的尖端矽晶片，正式成為國家外交工具。
-
Industry ENSovereignty for Silicon: G42 Weighs Selling Majority Control to US Firms to Keep the Chips Coming
Bloomberg reports Abu Dhabi's G42 has held exploratory talks to sell a majority stake to American companies — betting that US ownership is the only durable way to guarantee advanced AI chip access beyond its April 2027 license-free window.
-
Industry 中以主權換晶片：G42 考慮出售多數股權給美企，只為讓輝達晶片不斷貨
彭博報導，阿布達比的 G42 已展開初步磋商，考慮將多數股權出售給美國公司——押注唯有「美資化」才能在 2027 年 4 月的免許可窗口到期後，持續取得先進 AI 晶片。
-
Industry EN160,000 Huawei Chips for DeepSeek: Inside the Order That Could Reshape China's Compute Endgame
Bloomberg reports DeepSeek will deploy at least 160,000 Huawei Ascend 950DT accelerators at its 1GW Inner Mongolia data center — potentially the largest known Ascend cluster and a concrete step toward replacing Nvidia in China.
-
Industry 中16 萬顆華為晶片訂單：DeepSeek 的算力豪賭如何改寫中國 AI 佈局
Bloomberg 報導 DeepSeek 計畫在內蒙古 1GW 資料中心部署至少 16 萬顆華為 Ascend 950DT 加速器，可望成為已知最大規模的昇騰叢集，也是中國取代 Nvidia 算力最具體的一步。
-
Industry ENOne Workload, Six Kinds of Silicon: Gimlet Labs Raises $300M at a $3B Valuation to Orchestrate AI's Heterogeneous Future
Six months after an $80M Series A, Gimlet Labs has raised $300M led by a16z at a $3B valuation, with Arm and Microsoft's M12 joining, to run AI workloads across NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix chips.
-
Industry 中一種工作負載、六種晶片：Gimlet Labs 以 30 億美元估值融資 3 億美元，打造 AI 異構運算的調度中樞
在 8,000 萬美元 A 輪融資僅六個月後，Gimlet Labs 完成 a16z 領投的 3 億美元新一輪融資，估值達 30 億美元，Arm 與微軟 M12 均參投，要把 AI 工作負載分配到 NVIDIA、AMD、Intel、ARM、Cerebras 與 d-Matrix 等不同晶片上執行。
-
Industry ENAI Chips Tripled to $16.7B and the Stock Still Fell: Inside Broadcom's Massive Q3 and Its $230 Billion Bet
Broadcom's Q3 FY26 AI semiconductor revenue grew 221% to $16.7B and management now sees $115B in FY27 and $230B in FY28 — yet shares fell on a soft Q4 guide. Here's what the numbers actually say.
-
Industry 中AI 晶片營收暴增三倍至 167 億美元，股價卻下跌：解析 Broadcom 超狂 Q3 財報與 2,300 億美元的豪賭
Broadcom FY26 Q3 的 AI 半導體營收年增 221% 達 167 億美元，管理層更上調 FY27 展望至 1,150 億美元、勾勒 FY28 達 2,300 億美元——但 Q4 財測不及預期，股價應聲下挫。數字背後的真相是什麼？
-
Industry ENOne Category, Half the Exports: South Korea's Chip Sales Triple to a Record $46.65 Billion
AI infrastructure demand from Google and Amazon pushed South Korea's August semiconductor exports up 209% to a record $46.65 billion — 47.5% of everything the country sold abroad.
-
Industry 中單一品項撐起半壁出口：南韓晶片出口暴增 209% 至 466.5 億美元歷史新高
Google 與 Amazon 的 AI 基礎設施需求，推動南韓 8 月半導體出口年增 209% 至 466.5 億美元歷史新高——佔該國當月總出口的 47.5%。
-
Industry ENRecord Revenue, Falling Stock: HPE's Q3 FY26 Shows AI Demand Is Real — and Chained to Memory
HPE posted record Q3 FY26 results — $12.2B revenue up 34%, record $2.4B AI orders, raised guidance — yet shares slid as CFO Marie Myers admitted memory remains the bottleneck for AI server delivery.
-
Industry 中營收創紀錄、股價卻下跌：HPE 第三季財報證明 AI 需求是真的——但被記憶體綁住了
HPE 交出 2026 會計年度第三季最佳成績——營收 122 億美元、年增 34%，AI 訂單創紀錄達 24 億美元、上調全年財測——股價卻下跌，因為 CFO Marie Myers坦言記憶體仍是 AI 伺服器出貨的瓶頸。
-
Industry ENBroadcom's AI Chip Revenue Triples to $16.7 Billion — and Hock Tan Is Now Doubling Down Through 2028
Broadcom's Q3 FY26 print: AI semiconductor revenue up 221% to $16.7B, Q4 guided to $21.7B, and a staggering new roadmap targeting $115B in FY27 and $230B in FY28.
-
Industry 中Broadcom AI 晶片營收三倍暴增至 167 億美元——陳福陽豪賭 2028 年 2,300 億美元
Broadcom 2026 會計年度第三季財報：AI 半導體營收年增 221% 至 167 億美元，第四季指引 217 億美元，更喊出 FY27 達 1,150 億、FY28 直逼 2,300 億美元的路線圖。
-
Industry ENEurope's €387.8M LUMI-AI Bet: AMD MI430X GPUs and the Supercomputer Built to Win the AI Sovereignty Race
EuroHPC has signed a €387.8 million contract with Bull to build LUMI-AI, a next-generation AMD-powered AI supercomputer in Kajaani, Finland — 10x the AI capacity of today's LUMI, deployment in 2027, and the clearest signal yet of Europe's sovereign-compute ambitions.
-
Industry 中歐洲 3.878 億歐元的 LUMI-AI 豪賭：AMD MI430X GPU 與為 AI 主權競賽而生的超級電腦
EuroHPC 與 Bull 簽署 3.878 億歐元合約，將在芬蘭 Kajaani 打造次世代 AMD 架構 AI 超級電腦 LUMI-AI —— AI 算力達現有 LUMI 的 10 倍，2027 年部署，是歐洲主權算力企圖心最明確的訊號。
- Industry EN
Ex-Apple Face ID Engineers Triple Lyte's Valuation to $1.6B With $165M Series C
Physical-AI perception startup Lyte raised a $165M Series C led by Maverick Silicon at a $1.6B post-money valuation, just eight months after emerging from stealth — to put its LyteVision 'visual brain' into more robots, faster.
- Industry EN
前 Apple Face ID 工程師團隊以 1.65 億美元 C 輪融資，將 Lyte 估值推升至 16 億美元
實體 AI 感知新創 Lyte 完成 1.65 億美元 C 輪融資，由 Maverick Silicon 領投，投後估值達 16 億美元——距離走出隱身模式僅八個月，目標是讓 LyteVision「視覺大腦」更快進駐更多機器人。
-
Industry ENSouth Korea's $919 Billion Sovereign AI Bet: 18.4GW of Datacenters and a Tournament That Eliminated Its Best Model
A SemiAnalysis deep dive reveals Korea's $919B, 18.4GW-by-2035 sovereign AI buildout — and how the national foundation-model tournament just eliminated the best non-Chinese open-source model in the world.
-
Industry 中韓國 9,190 億美元主權 AI 豪賭：18.4GW 資料中心與一場淘汰掉自己最強模型的國家級賽局
SemiAnalysis 深度報告揭露韓國 9,190 億美元、2035 年前 18.4GW 的主權 AI 建設計畫——而國家基礎模型錦標賽剛淘汰了全世界最強的非中國開源模型。
-
Industry ENGoPro's $285M Nasdaq Escape: Inside the Starman Optical Merger That Turns a Camera Maker Into an AI Infrastructure Play
Cash-burning and delisting-threatened GoPro agreed to a $285 million merger with optical-photonics firm Starman Optical, pivoting 2,500 optics patents toward AI data centers and defense — and the stock doubled.
-
Industry 中GoPro 以 2.85 億美元的併購逃離下市危機：相機老牌變身 AI 基礎設施概念股的來龍去脈
現金枯竭、面臨下市威脅的 GoPro 與光電公司 Starman Optical 簽定 2.85 億美元合併協議，將 24 年累積的 2,500 多項光學專利轉向 AI 資料中心與國防市場——股價兩天翻倍。
-
Tools ENDLSS 5 Goes Live September 3: NVIDIA's Neural Rendering Era Begins With NBA 2K27 and 370 FPS
NVIDIA has locked September 3 for DLSS 5's official debut in NBA 2K27 — the first shipping build of its 3D-guided neural rendering stack, running at up to 370 FPS in 4K on an RTX 5090 and reaching down to RTX 5060 laptops.
-
Industry ENCerebras Bets €1.7 Billion on Finland: Inside the 165 MW Mikkeli AI Data Centre
Cerebras and Compute Nordic will build a 165 MW AI data centre in Mikkeli, Finland — a seven-year contracted, phased build-out backed by an estimated €1.0-1.7 billion in regional investment.
-
Industry 中Cerebras 押注芬蘭 17 億歐元：深入解析 Mikkeli 165 MW AI 資料中心
Cerebras 與 Compute Nordic 將在芬蘭 Mikkeli 興建 165 MW AI 資料中心——七年期合約承諾、分階段擴建，背後是估計 10 至 17 億歐元的區域投資。
-
Meta ENIris Goes Into Production: Meta's Homegrown AI Chip Aims Straight at Nvidia's Bill
Meta's custom 'Iris' accelerator enters manufacturing this month after a clean six-week validation, the opening move in a four-generation silicon plan built to bend the cost curve of a $145 billion AI buildout.
-
Meta 中Iris 進入量產：Meta 自研 AI 晶片正面瞄準 Nvidia 的帳單
Meta 自研加速器「Iris」在僅六週的乾淨驗證後於本月進入製造階段，這是四代自研晶片計畫的第一步，目標是扭轉 1,450 億美元 AI 基礎建設的成本曲線。
-
Industry ENAMD, Saudi MCIT, and DCO Launch Open Developer Ecosystem to Advance AI Innovation
At LEAP 2026, AMD, Saudi Arabia's MCIT, and the Digital Cooperation Organization launched a joint initiative to give developers and startups open AI tools, ROCm training, and governance resources across 16 member states.
-
Industry 中AMD 攜手沙烏地 MCIT 與 DCO 啟動開放開發者生態系，加速 AI 創新
在 LEAP 2026 上，AMD、沙烏地阿拉伯通訊與資訊科技部（MCIT）與數位合作組織（DCO）共同宣布啟動開發者生態系計畫，將 ROCm 開源軟體、AI 開發者培訓與治理資源帶給 16 個成員國的開發者與新創。
-
Industry ENChina's CXMT Begins First HBM3E Production, Cracking Open the AI Memory Bottleneck
ChangXin Memory Technologies has started small-batch production of HBM3E, China's first domestically made advanced AI memory — a generation behind the leaders, but a strategic breakthrough against export controls.
-
Industry 中中國 CXMT 首度投產 HBM3E，撬開 AI 記憶體瓶頸
長鑫存储（CXMT）已開始小批量生產 HBM3E，這是中國首度自產先進 AI 記憶體——雖落後龍頭一個世代，卻是對抗出口管制的戰略性突破。
-
Industry ENNVIDIA Puts $3.5 Billion Into MediaTek to Wire Custom AI Chips Into Its Empire
NVIDIA announced a $3.5 billion convertible bond investment in MediaTek alongside an expanded partnership that puts NVLink Fusion at the center of custom XPU development — extending its rack-scale AI factory strategy from its own GPUs to everyone else's silicon.
-
Industry 中輝達豪擲 35 億美元投資聯發科：把所有人的自研 AI 晶片都接上自己的帝國
NVIDIA 宣布以 35 億美元認購聯發科可轉換公司債，並擴大雙方合作——聯發科將採用 NVLink Fusion 平台，讓超大規模雲端業者自研的 XPU 能直接接入輝達的機櫃級 AI 工廠。輝達的算力帝國，正從自家 GPU 延伸到所有人的晶片。
-
Industry ENTaiwan Raids Nvidia Supplier Unimicron Over China-Parts Relabeling
Prosecutors raided Unimicron's Taoyuan headquarters over allegations it relabeled China-made PCBs as Taiwanese, hitting shares 10% and exposing origin-fraud risk at the heart of the AI supply chain.
-
Industry 中台灣檢方搜索輝達供應商欣興電子 揭中國製零件「洗產地」疑雲
桃園檢方搜索欣興電子總部，質疑其將中國製 PCB 回台重新標示為台灣製，股價重挫約 10%，也讓 AI 供應鏈的產地詐欺風險一次浮上檯面。
-
Industry ENOne-Third of US GDP Growth Is Now AI, ING Estimates — But It's Capex and the Wealth Effect, Not Consumers
ING's chief international economist James Knightley decomposes the AI investment boom: after subtracting imported chips, tech capex still drives roughly a third of US GDP growth — while only 2-3% of households pay for AI and layoffs citing AI hit a fifth straight month.
-
Industry ENTernus Takes Over Apple Tomorrow — and Bloomberg Says AI Is Job No. 1
John Ternus becomes Apple CEO on September 1 with an AI catch-up mandate: a Gemini-powered Siri rewrite shipped at WWDC after an 18-month delay, a $250M false-advertising settlement behind it, memory costs quadrupling, and a Sept 9 iPhone Fold launch days away.
-
Industry 中Ternus 明天正式接掌 Apple——Bloomberg：AI 是他的第一要務
John Ternus 於 9 月 1 日接任 Apple 執行長，首要任務是追趕 AI：延遲 18 個月的 Gemini 版 Siri 已在 WWDC 上架、2.5 億美元廣告不實和解案待收尾、記憶體成本翻漲四倍，9 月 9 日首款摺疊 iPhone 發表會更近在眼前。
-
Policy ENFed Chair Warsh Calls AI a Potential 'Fourth Factor of Production' in Landmark Jackson Hole Keynote
In his first Jackson Hole keynote, Fed Chairman Kevin Warsh framed AI as a potential fourth factor of production, cited $100B+ annualized token sales, and launched a task force on productivity and jobs.
-
Policy 中Fed 主席 Warsh 在 Jackson Hole 演說中稱 AI 為潛在的「第四種生產要素」
Fed 主席 Kevin Warsh 首次 Jackson Hole 演說將 AI 定調為潛在第四種生產要素，引用逾 1,000 億美元年化 token 銷售額，並成立生產力與就業任務小組。
-
Policy ENUS Drafts Rule to Close the Remote GPU Rental Loophole — but Lawyers Say BIS Lacks the Authority
Commerce is preparing a September rule to block Chinese AI firms from renting Nvidia GPU time through Thai and Singapore data centers, but export-control lawyers say BIS has no statutory authority over remote compute access.
-
Policy 中美國擬新規堵住遠端 GPU 租用漏洞——但律師說商務部根本沒有執法權
美國商務部正起草九月新規，要阻止中國 AI 公司透過泰國、新加坡資料中心租用 Nvidia GPU 算力，但出口管制律師普遍認為 BIS 對遠端算力存取缺乏法定授權。
-
Industry ENNvidia's $96 Billion Quarter and a 70% Growth Forecast That Stunned Wall Street
Nvidia reported Q2 FY2027 revenue of $96.2 billion, guided Q3 to $108 billion, and told investors fiscal 2028 revenue should grow about 70% — far above the ~44% analysts expected — alongside a plan to deploy 2 million more GPUs with AWS.
-
Industry 中Nvidia 單季營收 962 億美元，70% 成長預測震撼華爾街
Nvidia 公布 2027 財年第二季營收 962 億美元、年增 106%，並預估下一財年營收將成長約 70%，遠超市場預期的 44%；同時宣布與 AWS 合作於 2027–2028 年再部署 200 萬顆 GPU。
-
Policy ENCXMT Sues the Pentagon to Escape the 'Chinese Military Company' Blacklist
China's top DRAM maker ChangXin Memory Technologies has sued the U.S. Defense Department over its Section 1260H 'Chinese military company' designation — armed with an unusual piece of evidence: the Pentagon once told CXMT it was off the list, then changed its mind within hours.
-
Policy 中CXMT 控告美國國防部：中國記憶體龍頭全力反擊「中國軍事企業」黑名單
中國最大 DRAM 廠長鑫存儲（CXMT）向美國哥倫比亞特區聯邦地方法院提起訴訟，挑戰五角大廈依《國防授權法》第 1260H 條將其列入「中國軍事企業」清單的決定——而且握有一項不尋常的證據：五角大廈曾通知 CXMT 已將其移出名單，卻在數小時內反悔。
-
Policy ENPhase 2 Chip Tariffs Would Hit Laptops, Consoles — and AI Data Center Servers
The Trump administration is weighing a second round of semiconductor tariffs extending duties to finished goods — laptops, gaming consoles, and data center servers — with duty-free quotas pegged to pledged US fab capacity.
-
Policy 中晶片關稅第二波瞄準筆電、遊戲主機——以及 AI 資料中心伺服器
川普政府正研議第二輪半導體關稅，將課稅範圍從晶片本體擴大至終端產品——筆電、遊戲主機與資料中心伺服器，免稅額度則與企業承諾的美國本土產能掛鉤。
-
Tools ENAMD Ships ROCm 10: Agentic AI Tooling and a 3.3x Inference Claim in Its Biggest Software Release Yet
AMD's ROCm 10 marks a decade of its open compute stack with ROCm.AI — an agentic developer layer pairing ROCm CLI, AMD Skills for Claude/Cursor/Codex, and the Hyperloom auto-optimizer — plus a claimed 3.3x inference and 2.4x training uplift over ROCm 7.
-
Tools 中AMD 發布 ROCm 10：代理式 AI 工具鏈與 3.3 倍推論提升，十年來最大軟體更新
AMD 以 ROCm 10 歡慶開源運算堆疊十週年，主打 ROCm.AI 代理式開發體驗——整合 ROCm CLI、支援 Claude/Cursor/Codex 的 AMD Skills 與自動最佳化工具 Hyperloom——並宣稱推論效能較 ROCm 7 提升 3.3 倍、訓練提升 2.4 倍。
-
Industry ENCXMT Beats Samsung and SK Hynix to LPDDR6: Chinese Memory Debuts on Xiaomi's 18 Fold
China's CXMT has begun mass-producing LPDDR6 mobile memory and will debut it inside Xiaomi's 18 Fold flagship in September — the first time a Chinese DRAM maker ships next-generation memory in a premium phone ahead of Samsung, SK Hynix, and Micron.
-
Industry 中CXMT 搶先三星、SK 海力士量產 LPDDR6：中國記憶體首度登上小米 18 Fold 旗艦
中國最大記憶體廠長鑫存儲（CXMT）宣布 LPDDR6 進入量產，並將於 9 月在小米 18 Fold 旗艦摺疊手機首發——這是中國 DRAM 廠商首次與三星、SK 海力士、美光同步、甚至搶先在新一代行動記憶體上出貨。
-
Industry ENa16z's $1.1B Machine Age Fund: The Venture Giant Goes All-In on AI's Physical Layer
Andreessen Horowitz has raised $1.1 billion for its first dedicated hardware fund — backing chips, memory, networking, data centers, robotics, and even home AI appliances as the AI supply chain hits the limits of physics.
-
Industry 中a16z 11 億美元「機器時代基金」：創投巨頭全面押注 AI 的物理層
Andreessen Horowitz 成立首檔純硬體基金「機器時代基金」，規模 11 億美元，投資範圍涵蓋 AI 晶片、記憶體、網路設備、資料中心、機器人乃至家用 AI 裝置——因為 AI 供應鏈已經撞上物理極限。
-
Industry ENOpenAI's Jalapeño Chip Beats Nvidia Rubin on Performance-per-Watt in First Benchmarks
OpenAI's first custom inference chip, built with Broadcom in 16 months, delivers 1.5–1.9x more AI work per watt than the best shipping accelerators — with big implications for the CUDA moat.
-
Industry 中OpenAI 首顆自研晶片 Jalapeño 首波測試出爐：每瓦效能擊敗 Nvidia Rubin
OpenAI 與 Broadcom 花費 16 個月打造的首顆自研推理晶片，在每瓦效能上以 1.5–1.9 倍領先目前最強的量產加速器，也讓 Nvidia 的 CUDA 護城河首度受到實質挑戰。
-
Industry ENNvidia Hits Pause on Its $36B AI Cloud Financing Program Amid Antitrust Fears
Less than two months after launching the AI Compute Partnership, Nvidia has reportedly paused some deals after employees warned the 50%-of-revenue terms and customer-approval clauses could invite antitrust scrutiny.
-
Industry 中輝達喊停 360 億美元 AI 雲端融資計畫：反壟斷陰影下的急轉彎
AI Compute Partnership 推出不到兩個月，輝達據報暫停部分交易——內部員工警告，50% 超額收入分成與「客戶須經輝達核准」條款可能引來反壟斷審查。
-
Industry ENSK hynix Breaks Ground on $4B Indiana Fab: First 'Made in USA' HBM Arrives in 2029 — and the Memory Shortage Won't End Until 2030
SK hynix's West Lafayette groundbreaking starts a two-continent HBM pipeline — Korean wafers, Indiana packaging — with first US-made HBM in H2 2029, while CEO Kwak Noh-jung forecasts a global memory shortage through the end of 2030.
-
Industry 中SK hynix 印第安納 40 億美元廠動工：第一批「美國製造」HBM 2029 年問世，記憶體短缺卻要到 2030 年
SK hynix 在西拉法葉動工興建全美首座 HBM 先進封裝廠——韓國生產晶圓、印第安納負責封裝的雙洲生產線將於 2029 下半年量產，而執行長郭魯正預言全球記憶體短缺將持續到 2030 年底。
-
Industry ENIntel's Crescent Island: 480GB of LPDDR5X and a Bet That Agentic AI Inference Doesn't Need HBM
At Hot Chips 2026, Intel detailed Crescent Island — a 350W air-cooled Xe3P GPU with up to 480GB of LPDDR5X, 256 third-gen XMX engines, full-rate FP64, and a tokens-per-watt design philosophy aimed squarely at agentic AI inference.
-
Industry 中Intel Crescent Island：480GB LPDDR5X 與一場「代理式 AI 推論不需要 HBM」的豪賭
在 Hot Chips 2026 上，Intel 詳細揭露 Crescent Island——一張 350W 氣冷 PCIe 卡，搭載最高 480GB LPDDR5X、256 個第三代 XMX 引擎與全速 FP64，以「每瓦 token 數」為核心設計哲學，瞄準代理式 AI 推論市場。
-
Industry ENNvidia Compresses AI Model Release Cycle to 4–6 Weeks — Software Sprints While Hardware Walks
Nvidia's VP of Applied Deep Learning Bryan Catanzaro says the company now ships open-weight model updates every 4–6 weeks, down from 6–8 months — a 5–8x acceleration that applies to software only, while Blackwell and Vera Rubin keep their annual silicon cadence.
-
Industry 中Nvidia 將 AI 模型發布週期壓縮至 4–6 週 — 軟體衝刺，硬體慢走
Nvidia 應用深度學習研究副總裁 Bryan Catanzaro 證實，開放權重模型的更新週期已從每 6–8 個月縮短至每 4–6 週——約 5–8 倍的加速，但僅限軟體；Blackwell 與 Vera Rubin 矽晶片仍維持年度節奏。
-
Industry ENNvidia Doubles Revenue Again: Q2 FY2027 Hits $96.2B as AI Demand Stays 'Parabolic'
Nvidia's Q2 FY2027 results blew past estimates with $96.2B revenue (+106% YoY), $89B data center sales and $108B Q3 guidance — but Wall Street's reaction was muted, and supply, not demand, is now the binding constraint.
-
Industry 中輝達營收再度翻倍：FY2027 Q2 達 962 億美元，AI 需求依然「拋物線式」成長
輝達 FY2027 第二季財報再度大勝預期：營收 962 億美元（年增 106%）、資料中心營收 890 億美元、Q3 指引 1,080 億美元——但華爾街反應平淡，如今真正的瓶頸是供給，不是需求。
-
Tools ENSiFive's BigSky Ships: The First Rackable RISC-V Server Runs Ubuntu, RHEL 10 — and CUDA
SiFive's BigSky SF-2U870, the first enterprise-grade rackable 2U RISC-V server, is shipping to lead customers with out-of-the-box Ubuntu 26.04 LTS and RHEL 10 — and CUDA running on its P870-D cores as a head node for NVIDIA GPUs.
-
Tools 中SiFive BigSky 出貨：首款可上機架的 RISC-V 伺服器，原生支援 Ubuntu、RHEL 10——還能跑 CUDA
SiFive 的 BigSky SF-2U870 是首款企業級 2U 可上機架 RISC-V 伺服器，已開始出貨給先導客戶，開箱即支援 Ubuntu 26.04 LTS 與 RHEL 10，而且 CUDA 已能在其 P870-D 核心上執行、充當 NVIDIA GPU 叢集的前端主節點。
-
Industry ENMarvell Beats and Raises, but Shares Slide 8% as Investors Demand a Timeline for Google's $120B Chip Deal
Marvell posted record Q2 FY27 revenue of $2.739B (+37% YoY) and lifted FY27/FY28 guidance to $12B/$18B — yet the stock fell more than 8% as Wall Street pushed for clarity on when the Google custom-silicon agreement actually converts to revenue.
-
Industry 中Marvell 財報超預期並上調指引，股價卻暴跌 8%：華爾街要的是 Google 千億美元晶片交易的「時間表」
Marvell 2027 財年第二季營收創新高達 27.39 億美元（年增 37%），並將 FY27/FY28 營收指引上調至 120 億 / 180 億美元——但投資人追問 Google 客製晶片協議何時兌現為營收，週五早盤股價仍重挫逾 8%。
-
Industry ENAnthropic Planned, Then Walked Away From a $7 Billion Acquisition of Chip Startup MatX
A Reuters exclusive reveals Anthropic discussed buying ex-Google TPU engineers' chip startup MatX for roughly $7 billion before abandoning the deal — the clearest sign yet that frontier labs see custom silicon as existential.
-
Industry 中路透獨家：Anthropic 曾計劃以 70 億美元收購晶片新創 MatX，最終喊卡
路透社獨家揭露，Anthropic 曾洽談以約 70 億美元收購由前 Google TPU 工程師創立的晶片新創 MatX，最終放棄交易——這是前沿 AI 實驗室將客製化晶片視為生死存亡關鍵的最明確訊號。
-
Models ENApple's M6 Is the First 2nm Chip — and M5 Ultra Makes 512GB of AI Memory a Desktop Affair
Apple's M6 is the world's first 2nm silicon, and the quad-die M5 Ultra brings 512GB of unified memory at 1.2TB/s to the Mac Studio — a local-LLM workstation that quietly changes the economics of running frontier-scale models on a desk.
-
Models 中Apple M6 是全球首款 2nm 晶片——M5 Ultra 則把 512GB AI 記憶體帶上桌面
Apple M6 是全球第一款 2 奈米處理器，四晶片封裝的 M5 Ultra 更為 Mac Studio 帶來 512GB 統一記憶體與 1.2TB/s 頻寬——一部改變本機大型模型運算經濟學的桌面工作站。
-
Industry ENa16z Raises $1.1B 'Machine Age' Fund to Bankroll the Physical Buildout of AI
Andreessen Horowitz bets $1.1 billion that AI's next bottleneck is hardware — chips, memory, data centers, robots — not software.
-
Industry 中a16z 募集 11 億美元「機器時代基金」，全力押注 AI 的實體基礎建設
Andreessen Horowitz 押注 11 億美元，賭 AI 的下一個瓶頸在硬體——晶片、記憶體、資料中心與機器人，而非軟體。
-
Industry ENXPeng's IRON Humanoid Unit Raises $900 Million in China's Largest Physical AI Funding Round
XPeng's robotics unit Dogotix has raised over $900 million at a $6.3 billion valuation to mass-produce its IRON humanoid robot — the largest single-round private financing in China's physical AI industry, led by IDG Capital with Tencent and Alibaba as strategic investors.
-
Industry 中小鵬 IRON 人形機器人部門融資 9 億美元,創中國實體 AI 最大單輪紀錄
小鵬汽車旗下機器人業務 Dogotix 以超過 63 億美元估值融資逾 9 億美元,用於 IRON 人形機器人量產——這是中國實體 AI 產業迄今最大單輪私募融資,由 IDG 資本領投,騰訊與阿里巴巴戰略投資。
-
Industry ENNvidia Hits Pause on AI Cloud Revenue-Sharing Deals After Internal Antitrust Warnings
Less than two months after launching its revenue-sharing financing program for AI clouds, Nvidia has paused some deals after employees warned the structure — including a 50% revenue cut above a threshold — could draw antitrust scrutiny.
-
Industry 中內部反壟斷警告後，Nvidia 暫停 AI 雲端收益分享交易
推出 AI 雲端收益分享融資計畫不到兩個月，Nvidia 已暫停部分交易——員工警告這種結構（包括超過門檻後抽走 50% 收益）可能引來反壟斷審查。
-
Industry ENIntel's 256-Core Xeon 7 'Diamond Rapids': A Rome Moment Three Years Late, Built for the AI Era
At Hot Chips 2026, Intel detailed Xeon 7 'Diamond Rapids': up to 256 P-cores across 22 chiplets, FP8 matrix acceleration, 16 DDR5-8000 channels and 1.28GB of L3 — its most sophisticated Xeon ever, but it won't ship until 2027.
-
Industry 中Intel 256 核心 Xeon 7「Diamond Rapids」：遲到三年的 Rome 時刻，為 AI 時代而生
在 Hot Chips 2026 上，Intel 詳細揭露 Xeon 7「Diamond Rapids」：最多 256 個 P 核心、22 顆晶片堆疊、FP8 矩陣加速、16 通道 DDR5-8000 與 1.28GB L3 快取——史上最先進的 Xeon，但要等到 2027 年才會出貨。
-
Industry ENNvidia's $96.2 Billion Quarter: Vera Rubin Hits Full Production as AI Compute Becomes Revenue
Nvidia's Q2 FY2027 results doubled year over year to $96.2B with $89B from data centers, $108B guided for next quarter, and Vera Rubin racks already running at CoreWeave, Azure, Google Cloud and Oracle.
-
Industry 中Nvidia 962 億美元季度財報：Vera Rubin 全面量產，AI 算力正式成為營收
Nvidia FY2027 Q2 營收年增 106% 達 962 億美元，資料中心貢獻 890 億美元，下季指引 1,080 億美元，Vera Rubin 機架已進駐 CoreWeave、Azure、Google Cloud 與 Oracle。
-
Industry ENOpenAI's First Custom Chip 'Jalapeño' Beats Nvidia on Efficiency in First Public Benchmarks
At Hot Chips 2026, OpenAI showed its Broadcom-built inference ASIC beating Nvidia's GB200/GB300 rack systems with 1.5-1.9x more throughput per kilowatt — from a 700W part.
-
Industry 中OpenAI 首款自研晶片「Jalapeño」首次公開測試即擊敗 Nvidia 能效表現
在 Hot Chips 2026 上，OpenAI 展示與 Broadcom 共同開發的推論專用 ASIC，以 700W 之力在每千瓦吞吐量上超越 Nvidia GB200/GB300 機架系統 1.5 至 1.9 倍。
-
Industry ENNvidia Puts a Number on It: AI Labs It Finances Will Be ~25% of Next Year's Business
Fresh off a $96.2B quarter, Nvidia disclosed that the AI labs it helps finance will account for roughly a quarter of fiscal 2028 revenue — with $366B in future commitments and $108.5B in guarantee exposure on the books — while Jensen Huang calls the circular-financing critique overblown and insists 'the risk is low.'
-
Industry 中Nvidia 首次給出具體數字：它出資扶植的 AI 實驗室，將占下一個財年約四分之一業績
在 962 億美元營收的季度之後，Nvidia 披露其出資扶植的 AI 實驗室將占 2028 財年約四分之一營收——帳上未來承諾總額達 3,660 億美元、最大擔保曝險 1,085 億美元——黃仁勳則回應「循環融資」質疑，強調「風險很低」。
-
Industry ENHuawei Widens Its AI Pharma Push: From Compound Screening to the Clinic
Huawei's healthcare president William Zhang says the company will expand AI cooperation with Chinese drugmakers into drug development and clinical practice — riding its Ascend and Kunpeng chips into a market where Nvidia has already partnered with Eli Lilly and Novo Nordisk.
-
Industry 中華為擴大 AI製藥布局：從分子篩選一路推進臨床
華為醫療事業總裁張強表示，將把與中國藥企的 AI 合作從早期研究擴展到藥物開發與臨床實踐——憑藉昇騰與鯤鵬晶片，切入輝達已與禮來、諾和諾德結盟的 AI 製藥市場。
-
Industry ENNvidia's $96 Billion Quarter: Inside the Numbers Behind the AI Boom's Most-Watched Report
Nvidia's Q2 FY2027 revenue doubled to $96.2B with data center sales up 117%, but markets shrugged — here's what the numbers really say about the AI buildout.
-
Industry 中Nvidia 的 960 億美元季度：AI 熱潮中最受矚目財報背後的數字解讀
Nvidia 2027 財年第二季營收翻倍至 962 億美元、資料中心暴增 117%，市場反應卻出奇平淡——這些數字究竟透露了 AI 建設熱潮的什麼訊息？
-
Industry ENAnthropic Signs $45 Billion Compute Deal with Nscale — 460MW of Nvidia Vera Rubin Power for the Claude Era
Anthropic has agreed to pay UK neocloud Nscale roughly $45 billion over six years to lease about 460 megawatts of Nvidia Vera Rubin capacity at its Monarch campus in West Virginia — one of the largest AI infrastructure deals ever, signed weeks before its IPO roadshow.
-
Industry 中Anthropic 與 Nscale 簽下 450 億美元算力合約——460MW 的 Nvidia Vera Rubin 運算，買斷 Claude 時代的未來
Anthropic 同意在六年內支付英國新創 Neocloud 業者 Nscale 約 450 億美元，租用其西維吉尼亞 Monarch 園區約 460 百萬瓦的 Nvidia Vera Rubin 算力——這是史上最大規模的 AI 基礎設施交易之一，而距離 Anthropic 的 IPO 路演只剩幾週。
-
Industry ENCisco Puts Supermicro Inside: Secure AI Factory Grows Rack-Scale Teeth for the Rubin Era
Cisco is folding Supermicro's liquid-cooled rack-scale systems into its Secure AI Factory with NVIDIA, targeting Vera Rubin NVL72-class builds for enterprises, neoclouds and sovereign clouds — a bet that the next wave of AI infrastructure gets bought as one validated system, not assembled part by part.
-
Industry 中Cisco 把 Supermicro 收進體系：Secure AI Factory 長出機架級利牙，迎戰 Rubin 時代
Cisco 宣布將 Supermicro 的液冷機架級伺服器納入 Secure AI Factory with NVIDIA 架構，瞄準 Vera Rubin NVL72 等級的企業、Neocloud 與主權雲部署——押注下一波 AI 基礎設施將以「整套驗證系統」而非零組件拼裝的方式被採購。
-
Industry ENAWS Triples Down on Nvidia: 2 Million More GPUs, Trainium Gets NVLink Fusion, and Federal AI Factories at IL6
Announced on Nvidia's earnings call, the expanded AWS-Nvidia pact adds 2M more GPUs for 2027-28, brings Vera CPUs to EC2, wires NVLink Fusion and NVHBM into Amazon's own Trainium chips, and targets 100K GPUs of classified federal AI infrastructure.
-
Industry 中AWS 加碼押注 Nvidia：再採購 200 萬顆 GPU、Trainium 導入 NVLink Fusion、IL6 聯邦 AI 工廠
在 Nvidia 財報電話會議上宣布的擴大合作，將於 2027-28 年為 AWS 再增添 200 萬顆 GPU、把 Vera CPU 帶進 EC2、讓 Amazon 自家 Trainium 晶片整合 NVLink Fusion 與 NVHBM，並瞄準 10 萬顆 GPU 規模的機密聯邦 AI 基礎設施。
-
Industry ENNvidia Doubles Down: Q2 FY27 Revenue Hits $96.2B, Guides $108B as AI Buildout Goes 'Full Steam'
Nvidia's fiscal Q2 2027 revenue doubled to $96.2B with $89B from data centers, a $108B Q3 guide, and a $2 trillion backlog — compute has officially become revenue.
-
Industry 中輝達財報炸裂：FY27 Q2 營收 962 億美元翻倍成長，Q3 指引 1,080 億美元，AI 建設「全速前進」
輝達 FY27 第二季營收翻倍至 962 億美元，資料中心貢獻 890 億美元，Q3 指引 1,080 億美元，積壓訂單突破 2 兆美元——運算正式成為營收。
-
Industry ENYMTC's $4.9 Billion IPO Filing Takes Aim at Samsung and SK Hynix
China's NAND champion YMTC has filed for a $4.9 billion Shanghai STAR Market listing through parent CCSH — and told IPO investors it intends to become the world's largest flash supplier by the end of 2027, overtaking Samsung and SK Hynix.
-
Industry 中YMTC 49 億美元 IPO 申請，劍指三星與 SK 海力士
中國 NAND 快閃記憶體龍頭長江存儲（YMTC）透過母公司長江存儲科技控股（CCSH）申請在上海科創板上市，募資 49 億美元，並向 IPO 投資人表明：目標在 2027 年底前超越三星與 SK 海力士，成為全球最大快閃記憶體供應商。
-
Industry ENOpenAI's Jalapeño Chip Posts First Benchmarks: Up to 1.9x Nvidia Blackwell Efficiency, 3.6x Lower Latency
OpenAI's first custom inference chip, built with Broadcom, delivered 1.5-1.9x more AI work per watt and up to 3.6x lower latency than Nvidia's GB200/GB300 systems in its first public benchmarks.
-
Industry 中OpenAI 自研晶片 Jalapeño 首批實測出爐：能效最高達 Nvidia Blackwell 的 1.9 倍、延遲低 3.6 倍
OpenAI 與 Broadcom 共同打造的首款自研推理晶片 Jalapeño 公布首批實測數據：每瓦效能最高達 Nvidia GB200/GB300 系統的 1.9 倍，端對端延遲最低僅其 3.6 分之一。
-
Industry ENIntel Details Xeon 7 'Diamond Rapids': 256 Cores, 1.28 GB of Cache, and a 2027 Date
At Hot Chips 2026, Intel revealed its most modular Xeon ever — up to 256 P-cores across 22 chiplets with FP8 matrix acceleration, 16 DDR5 channels, and a fan-out fabric that finally answers AMD's chiplet playbook.
-
Industry 中Intel 揭曉 Xeon 7「Diamond Rapids」：256 核心、1.28 GB 快取，2027 年登場
在 Hot Chips 2026 上，Intel 公開了史上最模組化的 Xeon 伺服器處理器——最多 256 個 P 核心、22 顆晶粒組合、支援 FP8 矩陣運算與 16 通道 DDR5，要用「扇出型架構」正面迎戰 AMD 的小晶片布局。
-
Tools ENNVIDIA Announces Jetson Orin Nano 2: 78 TOPS Robotics Computer for Entry-Level Physical AI
NVIDIA's new Jetson Orin Nano 2 doubles edge inference performance in the same form factor while cutting power 40%, putting frontier-class generative AI into robots and drones.
-
Tools 中NVIDIA 發布 Jetson Orin Nano 2：78 TOPS 入門級物理 AI 運算模組
NVIDIA 新推出的 Jetson Orin Nano 2 在相同體積下將邊緣推論效能翻倍、功耗降低 40%，把前沿級生成式 AI 帶進機器人與無人機。
-
Models ENNVIDIA's Groq 3 LPX Enters Full Production: 3,400 Tokens/sec for the Agentic AI Era
NVIDIA's first product from its $20B Groq acqui-hire — the Groq 3 LPX inference rack — is now in full production, pairing 256 LPU accelerators with Vera Rubin NVL72 and posting a record 3,400 output tokens/sec on Gemma 4 31B.
-
Models 中NVIDIA Groq 3 LPX 進入量產：Agentic AI 時代的 3,400 Tokens/秒推論引擎
NVIDIA 以 200 億美元收購 Groq 團隊後的首款產品 Groq 3 LPX 推論機架正式進入量產，每座機架整合 256 顆 LPU 加速器並與 Vera Rubin NVL72 協同部署，在 Gemma 4 31B 上創下每秒 3,400 輸出 tokens 的紀錄，Nebius 成為首家採用的 AI 雲。
-
Industry ENEnflame, the Last of China's Four AI Chip Dragons, Opens Its $892 Million STAR Market IPO
Tencent-backed Enflame opens subscriptions September 2 for a 6 billion yuan Shanghai STAR Market listing — the final IPO among China's 'four little dragons' of AI silicon, and a referendum on Beijing's chip-independence push.
-
Industry 中燧原科技掛牌倒數：中國「AI 晶片四小龍」最後一員啟動 892 億台幣級 STAR 科創板 IPO
騰訊投資的燧原科技宣布 9 月 2 日開放上交所科創板 IPO 申購，募資約 60 億人民幣（8.92 億美元）——這是「四小龍」中最後掛牌的一家，也是對中國晶片自主之路的一場公開投票。
-
Industry ENIndia's $8 Billion AI Bet: AM Intelligence Orders 9,000 NVIDIA Vera Rubin Systems
Hyderabad-based AM Intelligence, backed by Greenko's founders, has placed a binding order for ~9,000 NVIDIA Rubin GPUs as Vera Rubin NVL72 systems — the first tranche of an $8B+, 1GW compute plan that could make India a major AI factory hub.
-
Industry 中印度 80 億美元的豪賭：AM Intelligence 下單 9,000 套 NVIDIA Vera Rubin 系統
由 Greenko 創辦人支持的海德拉巴公司 AM Intelligence，已簽下約 9,000 顆 NVIDIA Rubin GPU 的約束性訂單（Vera Rubin NVL72 機架級系統），作為 80 億美元、1GW 運算計畫的第一步——印度可望躍居全球 AI 工廠重鎮。
-
Models ENOpenAI's Jalapeño Chip Posts First Benchmarks: Up to 1.9x the Efficiency of Blackwell
At Hot Chips, OpenAI's first custom inference ASIC beat Nvidia Blackwell on SemiAnalysis' InferenceX: 1.5-1.9x more work per watt and 1.7-3.6x lower latency.
-
Models 中OpenAI 自研晶片 Jalapeño 首批基準測試出爐:每瓦效率最高達 Blackwell 的 1.9 倍
在 Hot Chips 會議上,OpenAI 首顆自研推理 ASIC 在 SemiAnalysis 的 InferenceX 基準測試中擊敗 Nvidia Blackwell:每瓦工作量提升 1.5-1.9 倍,端對端延遲降低 1.7-3.6 倍。
-
Models ENCaltech Duo Launches Accelerated Understanding, a Physics AI That Ditches Transformers
Anima Anandkumar and Benedikt Jenik unveiled Accelerated Understanding Inc, an enterprise physics AI built on neural operators that handled 5 trillion data points in a single prompt — after walking away from a Bezos-backed Prometheus offer.
-
Models 中Caltech 團隊創辦 Accelerated Understanding：捨棄 Transformer 的物理 AI
Anima Anandkumar 與 Benedikt Jenik 推出 Accelerated Understanding Inc，以神經算子打造、單一提示可處理 5 兆個資料點的企業級物理 AI——兩人先前拒絕了 Bezos 支持的 Prometheus 招募。
-
Models ENApple's M6 and M5 Ultra: First 2nm Chip and Quad-Die Monster Aim squarely at On-Device AI
Apple's M6 is its first 2nm chip with a Dual 16-core Neural Engine, while M5 Ultra fuses four dies into the most powerful M-series SoC ever — 512GB unified memory, 1.2TB/s bandwidth, built to run frontier AI models locally.
-
Models 中Apple M6 與 M5 Ultra：首款 2nm 晶片與四晶粒巨獸，正面搶攻裝置端 AI
Apple M6 是首款 2nm 製程晶片、配備雙 16 核心神經網路引擎；M5 Ultra 則以四晶粒架構成為史上最強 M 系列晶片——512GB 統一記憶體、1.2TB/s 頻寬，目標是讓前沿 AI 模型直接在本機跑。
-
Industry ENNvidia's $92 Billion Question: What to Watch When Q2 Earnings Land on August 26
Nvidia reports fiscal Q2 2027 results on August 26 with consensus at ~$92B revenue and $2.08 EPS. Between Vera Rubin's production ramp, a $104B Q3 guide, and $725B of hyperscaler capex, here's the full checklist for the most-watched print in tech.
-
Industry 中輝達的 920 億美元大考：8 月 26 日 Q2 財報五大觀察重點
輝達將於 8 月 26 日盤後公布 2027 財年第二季財報，市場共識營收約 920 億美元、EPS 2.08 美元。從 Vera Rubin 量產爬坡、1,040 億美元的 Q3 財測，到雲端巨頭 7,250 億美元資本支出，本文整理這場科技界最重要財報的完整檢查清單。
-
Industry ENIBM's Dual-ISA Mainframe Chip Runs Arm and z/Architecture in the Same Cores
At Hot Chips 2026, IBM unveiled the first dual-architecture mainframe processor: a 2nm, 11-core chip where every core natively executes both IBM Z and Arm AArch64 instructions at 5.7+ GHz, plus a second-gen AI inference accelerator with 96GB of HBM3e.
-
Industry 中IBM 雙架構主機晶片問世：同一核心原生執行 Arm 與 z/Architecture
在 Hot Chips 2026，IBM 發表首款雙架構主機處理器：2nm 製程、11 核心、時脈超過 5.7 GHz，每個核心都能原生執行 IBM Z 與 Arm AArch64 指令，並搭配內建 96GB HBM3e 的第二代 AI 推論加速器。
-
Industry ENIntel's Three-Layer Bet on Agentic AI: 256-Core Diamond Rapids, 480GB Crescent Island, and Wildcat Lake at the Edge
At Hot Chips 2026, Intel detailed a full-stack silicon strategy for agentic AI: the 256-core Diamond Rapids Xeon on 18A-P, the 480GB LPDDR5X Crescent Island inference GPU that skips HBM entirely, and Wildcat Lake — the first Intel processor to use the UCIe chiplet standard — for the edge.
-
Industry 中英特爾的代理式 AI 三層布局：256 核心 Diamond Rapids、480GB Crescent Island 與邊緣端 Wildcat Lake
在 Hot Chips 2026，英特爾公開了完整的代理式 AI 矽晶片策略：採用 18A-P 製程、256 核心的 Diamond Rapids Xeon，完全捨棄 HBM、改用 480GB LPDDR5X 的 Crescent Island 推論 GPU，以及首款導入 UCIe 小晶片標準的英特爾處理器 Wildcat Lake。
-
Industry ENSEC Subpoenas Four Wall Street Banks Over the Situational Awareness Collapse
US regulators have subpoenaed Bank of America, Citi, Goldman Sachs and JPMorgan as an investigation opens into the near-collapse of Leopold Aschenbrenner's leveraged AI hedge fund.
-
Industry 中SEC 傳喚四大華爾街銀行，調查 Situational Awareness 基金崩塌案
美國監管機構已傳喚美國銀行、花旗、高盛與摩根大通，針對 Leopold Aschenbrenner 的高槓桿 AI 避險基金近乎崩盤一事展開調查。
-
Industry ENSpaceXAI Puts NVIDIA's Vera CPU to Work — and a Vera Rubin NVL72 Is Heading to Orbit
NVIDIA says SpaceXAI will deploy Vera — the first CPU built for AI agents — across its agentic AI stack, scale Grok on Vera Rubin toward gigawatts, and fly an optimized Vera Rubin NVL72 in its first-generation Starmind AI satellite.
-
Industry 中SpaceXAI 讓 NVIDIA Vera CPU 上陣——還有一套 Vera Rubin NVL72 即將升空
NVIDIA 宣布 SpaceXAI 將部署首款為 AI 代理人打造的 Vera CPU，以 Vera Rubin 平台擴展 Grok 基礎設施朝 GW 級邁進，並讓最佳化的 Vera Rubin NVL72 搭乘第一代 Starmind AI 衛星進入軌道。
-
Industry ENUK Signs Landmark Deal for Access to Ukraine's Avengers AI Labs Battlefield Data
Britain becomes the first international partner to tap Ukraine's Avengers AI Labs, a vast battlefield dataset that will train AI models to protect UK bases, railways and energy infrastructure.
-
Industry 中英烏簽署里程碑式 AI 協議：英國取得 Avengers AI Labs 戰場資料存取權
英國成為首個獲准存取烏克蘭 Avengers AI Labs 的國際夥伴，這座戰場資料寶庫將用於訓練 AI 模型，保護英國基地、鐵路與能源基礎設施。
-
Industry ENGartner: Global Semiconductor Revenue to Hit $1.6 Trillion in 2026 as AI Reshapes the Chip Industry
Gartner's latest forecast puts 2026 semiconductor revenue at $1.6 trillion, up 92% from $809 billion in 2025, with memory climbing to 54% of the market on AI data center demand.
-
Industry 中Gartner 預測:2026 年全球半導體營收將達 1.6 兆美元,AI 正重塑晶片產業
Gartner 最新預測將 2026 年半導體營收上調至 1.6 兆美元,較 2025 年的 8,090 億美元成長 92%,記憶體在 AI 資料中心需求推動下占整體市場 54%。
-
Industry ENNVIDIA's Groq 3 LPX Enters Full Production: 3,400 Tokens/Sec Inference Racks Built for the Agentic Era
NVIDIA's Groq 3 LPX — the low-latency inference rack born from its $20B Groq acqui-hire — is now in full production, pairing 256 LPU accelerators with 128GB of on-chip SRAM per rack and hitting a record 3,400 output tokens/sec on Gemma 4 31B. Nebius is the first AI cloud to deploy it inside its Token Factory.
-
Industry 中NVIDIA Groq 3 LPX 全面量產：為代理式 AI 時代打造的每秒 3,400 Token 推理機櫃
NVIDIA 源自 200 億美元 Groq 收購的低延遲推理機櫃 Groq 3 LPX 正式進入全面量產,每個機櫃整合 256 顆 LPU 加速器與 128GB 晶片上 SRAM,在 Gemma 4 31B 寫下每秒 3,400 token 的紀錄。Nebius 成為首家採用的 AI 雲,將部署於其 Token Factory 平台。
-
Industry ENNVIDIA Sends the AI Factory to Orbit: SpaceXAI Adopts Vera CPUs and a Vera Rubin NVL72 Satellite
NVIDIA announced that SpaceXAI will deploy Vera CPUs for agentic AI workloads and base its first-generation Starmind AI satellite on an optimized Vera Rubin NVL72 — the first time a rack-scale AI factory architecture is being adapted for space.
-
Industry 中NVIDIA 把 AI 工廠送上軌道：SpaceXAI 採用 Vera CPU，首款 Starmind AI 衛星將以 Vera Rubin NVL72 為基礎
NVIDIA 宣布 SpaceXAI 將部署 Vera CPU 加速代理式 AI 工作負載，並以最佳化的 Vera Rubin NVL72 系統打造第一代 Starmind AI 衛星——這是機架級 AI 工廠架構首次被搬上太空。
-
Industry ENAnthropic Hires Google TPU Founder Amir Salek to Build Its Own AI Chips
Anthropic has hired Amir Salek, the engineer who co-founded Google's custom silicon program and shipped seven TPU generations, as it lays the groundwork for in-house AI chips — the strongest sign yet that the Claude maker wants to own its compute stack.
-
Industry 中Anthropic 聘請 Google TPU 之父 Amir Salek,要親自下場造 AI 晶片
Anthropic 聘請了 Google 自研晶片計畫共同創辦人、七代 TPU 推手 Amir Salek,為自研 AI 晶片鋪路——這是 Claude 開發商至今最明確的訊號:它想掌控自己的運算堆疊。
-
Industry ENNVIDIA's Groq 3 LPX Enters Full Production: 3,400 Tokens/sec and the Coming Disaggregation of AI Inference
At Hot Chips 2026, NVIDIA announced its Groq 3 LPX inference rack has entered full production, delivering a record 3,400 output tokens/sec on Gemma 4 31B at 100K context — with Nebius as the first cloud customer for its Token Factory and SpaceXAI standardizing on Vera CPUs.
-
Industry 中NVIDIA Groq 3 LPX 進入全面量產：每秒 3,400 Token，AI 推論的解構時代來臨
NVIDIA 在 Hot Chips 2026 宣布 Groq 3 LPX 推論機架進入全面量產，在 Gemma 4 31B、100K 上下文下創下每秒 3,400 個輸出 token 的紀錄——Nebius 成為首個導入的雲端客戶，SpaceXAI 則宣布採用 Vera CPU 打造次世代 AI 架構。
-
Industry ENXiaomi's Xring O3 Breaks 5 Million on AnTuTu — and Unveils a Three-Chip Silicon Strategy
Xiaomi launched three in-house chips at its Beijing Xring conference: the 3nm Xring O3 flagship SoC (first mobile chip past 5 million on AnTuTu, first with LPDDR6), the 6nm O100 NPU with 1.22 Tbps near-memory bandwidth for on-device MiMo LLM, and the 3nm D100 driving chip that runs 200B-parameter models locally — a coordinated push to cut dependence on Qualcomm and MediaTek.
-
Industry 中小米 Xring O3 突破安兔兔 500 萬分——三款自研晶片揭示的矽智財戰略
小米在北京 Xring 晶片技術發表會上一次推出三款自研晶片：3 奈米旗艦 SoC Xring O3（首款安兔兔破 500 萬分、首支援 LPDDR6 的手機晶片）、6 奈米 O100 NPU 以 1.22 Tbps 近記憶體頻寬實現 MiMo 大模型端側運行，以及可在車內本地運行 2000 億參數模型的 3 奈米智駕晶片 D100——目標直指降低對高通與聯發科的依賴。
-
Industry ENCerebras CS-4: 30x Faster AI Inference From Three Overclocked Wafers and a Reinvented Rack
Cerebras unveiled the CS-4, a rack-scale AI accelerator built from three overclocked WSE-3 Turbo wafer-scale processors on its new modular Nexus platform — claiming up to 30x faster inference than GPU systems, 1,000+ tokens/sec on 10T-parameter models, and native support for disaggregated inference with AMD Helios and AWS Trainium handling prefill.
-
Industry 中Cerebras CS-4:三片超頻晶圓與重新發明的機架,帶來 30 倍速 AI 推論
Cerebras 發表 CS-4 機架級 AI 加速器,以三顆超頻版 WSE-3 Turbo 晶圓級處理器搭配全新模組化 Nexus 平台,號稱推論速度最快達 GPU 系統的 30 倍、10 兆參數模型可跑出每秒 1,000+ token,並原生支援與 AMD Helios、AWS Trainium 協作的分解式推論架構。
-
Industry ENWaymo's Secret Weapon: A Custom 5nm Chip Pushing 1,000+ TOPS Inside Its Next-Gen Robotaxi
Waymo revealed it designed its own 5nm ASIC for the sixth-generation Driver in the Ojai robotaxi — over 1,000 TOPS of pre-processing power rivaling Nvidia's DRIVE AGX Thor, with partners spanning TSMC, Samsung, and AMD. Vertical integration from silicon up is now the robotaxi industry's core bet on profitability.
-
Industry 中Waymo 的祕密武器：自研 5 奈米晶片，在次世代自駕計程車內狂飆 1,000+ TOPS
Waymo 證實為第六代 Driver 自行設計 5 奈米 ASIC 晶片，部署於 Ojai 自駕計程車，提供超過 1,000 TOPS 的感測資料前處理效能，直逼 Nvidia DRIVE AGX Thor，合作夥伴橫跨台積電、三星與 AMD。從晶片層級開始的垂直整合，正成為自駕計程車產業押注獲利的核心策略。
-
Industry ENNvidia's $92 Billion Question: What to Watch at Q2 FY2027 Earnings on August 26
Wall Street expects roughly $92 billion in revenue when Nvidia reports fiscal Q2 2027 results on Wednesday, with Blackwell demand, the Rubin roadmap, and AI capex endurance all in focus.
-
Industry 中輝達的 920 億美元考題：8 月 26 日財報的五個觀察重點
輝達將於週三美股收盤後公布 2027 財年第二季財報，市場共識營收約 920 億美元，Blackwell 需求、Rubin 路線圖與 AI 資本支出的持續性是三大焦點。
-
Industry ENNvidia to Raise AI Server Prices Up to 17% as Memory Costs Explode
Server makers have told customers that Grace Blackwell 300 and Vera Rubin 200 systems will cost up to 17% more from early 2027, as the AI-driven memory shortage turns DRAM and HBM into the industry's most expensive components.
-
Industry 中記憶體成本暴漲，Nvidia AI 伺服器最多將漲價 17%
伺服器廠商已通知客戶，Grace Blackwell 300 與 Vera Rubin 200 系統自 2027 年初起最高漲價 17%——AI 掀起的記憶體超級週期，正讓 DRAM 與 HBM 成為整個產業最昂貴的零組件。
-
Industry ENNvidia Tells Its Biggest Customers: AI Servers Will Cost 15%+ More Next Year
Bloomberg reports Nvidia's largest customers have been warned that AI servers — including Vera Rubin and Grace Blackwell systems — will cost more than 15% extra when they ship in early 2027, as soaring memory prices rewrite the economics of the AI buildout.
-
Industry 中Nvidia 告訴最大客戶：明年起 AI 伺服器將漲價超過 15%
彭博報導，Nvidia 已通知最大客戶：搭載 Vera Rubin 與 Grace Blackwell 平台的 AI 伺服器，明年初出貨時價格將上漲超過 15%，記憶體成本飆漲正在改寫 AI 建置的經濟學。
-
Industry ENHot Chips 2026 Kicks Off at Stanford: OpenAI's Jalapeño Finale and the Inference Silicon Wars
Hot Chips 2026 opens today at Stanford with NVIDIA's Vera and Rubin, Google's TPU 8, Microsoft's MAIA 200, and OpenAI's Jalapeño inference chip closing the show — the clearest sign yet that the AI battle has moved to custom silicon.
-
Industry 中Hot Chips 2026 史丹佛登場：OpenAI Jalapeño 壓軸與推論晶片大戰
Hot Chips 2026 今天在史丹佛開幕，NVIDIA Vera 與 Rubin、Google TPU 8、Microsoft MAIA 200 同場較勁，OpenAI 的 Jalapeño 推論晶片擔綱壓軸——AI 戰場正式轉移到客製化矽晶片的最明確訊號。
-
Industry ENEtched Doubles to $21B as Jane Street Bets on Transformer-Only Silicon
The AI chip startup's valuation more than doubled in under a month as quant giant Jane Street led a $700M round — and became Sohu's first customer.
-
Industry 中Etched 估值一個月翻倍至 210 億美元，Jane Street 領投並成為首位客戶
這家只做 Transformer 推論晶片的新創，以不到一個月的時間讓估值翻倍；量化巨頭 Jane Street 領投 7 億美元，同時採用其 Sohu 晶片。
-
Industry ENBroadcom's $80 Billion Gamble: The Biggest AI Debt Deal Ever Is Being Built for Anthropic's Chips
Broadcom is in talks to raise $70-80 billion in debt — potentially $100 billion — through a special-purpose vehicle to bankroll custom AI chips for Anthropic and other customers, in what would be the largest single AI financing ever attempted.
-
Industry 中Broadcom 的 800 億美元豪賭：史上最大 AI 債務交易，為 Anthropic 的晶片而生
Broadcom 正洽談透過特殊目的載具募集 700 至 800 億美元債務（總規模可能逼近 1,000 億美元），為 Anthropic 等客戶的訂製 AI 晶片提供資金——這將是 AI 產業史上最大規模的單一融資案。
-
Industry ENNvidia's China Comeback Play: A Groq-Based LPU That Complies With US Export Rules — and the Denial That Followed
The Information reported Nvidia is preparing a China-specific LPU built on Groq-licensed inference technology that could clear US export rules without compute caps — Nvidia denied it within a day. Either way, the report reveals how the China AI chip endgame is now being fought.
-
Industry 中Nvidia 的中國復活計畫：取得 Groq 授權、符合美國出口管制的新型 LPU——以及隨之而來的官方否認
《The Information》報導 Nvidia 正準備一款中國專用 LPU，採用 Groq 授權的推論技術，無需降規即可符合美國出口管制——Nvidia 在 24 小時內否認。無論真相為何，這份報導揭露了中國 AI 晶片終局之戰的新打法。
-
Industry ENFrom $10B to $21B in a Month: Jane Street Leads $700M Bet on Etched's Transformer-Only Silicon
AI chip startup Etched more than doubled its valuation to $21 billion in under a month after quant giant Jane Street installed its first inference cluster and was impressed enough to lead a $700 million round — the fastest re-rating of any chip startup this cycle.
-
Industry 中一個月從 100 億變 210 億美元：Jane Street 領投 7 億美元，豪賭 Etched 的純 Transformer 晶片
AI 晶片新創 Etched 在不到一個月內將估值從 103 億美元翻倍至 210 億美元——因為量化巨頭 Jane Street 先實際安裝了它的首套推理叢集、驗證效果後，才領投這輪 7 億美元融資。
-
Industry ENAmazon Quietly Hikes Echo, Kindle, and Fire TV Prices — the AI Memory Squeeze Reaches Your Living Room
Amazon overnight raised prices across Echo, Kindle, Fire TV, and eero hardware, citing 'significant increases in memory and storage component costs' — the clearest sign yet that the AI data-center buildout is now taxing ordinary consumer electronics.
-
Policy ENChina Is Quietly Throttling Germanium and Quartz Exports to Taiwan
Nikkei reveals Chinese customs is slow-walking germanium, quartz and magnet shipments to Taiwan — a materials squeeze one layer above the chip fabs that feed the world's AI accelerators.
-
-
Industry ENAnthropic Hires the Founder of Google's TPU as It Lays the Groundwork for Its Own Chips
Bloomberg reports Anthropic has hired Amir Salek, who founded Google's TPU program in 2013 and ran it until 2022, to join its compute team — the clearest signal yet that the Claude maker intends to design its own AI silicon.
-
Industry 中Anthropic 聘請 Google TPU 之父，為自研晶片鋪路
根據 Bloomberg 報導，Anthropic 已聘請 2013 年為 Google 創立 TPU 計畫、並領導該業務至 2022 年的 Amir Salek 加入其運算團隊——這是這家 Claude 開發商有意自研 AI 晶片迄今最明確的訊號。
-
Industry ENNvidia in Early Talks With Korean AI Chip Startup Rebellions
Bloomberg reports Jensen Huang met Rebellions CEO Sunghyun Park to discuss a partnership, investment, or acquisition — the latest move in Nvidia's AI-silicon consolidation campaign.
-
Industry 中輝達傳與韓國 AI 晶片新創 Rebellions 洽談合作
Bloomberg 報導黃仁勳本週在輝達總部會見 Rebellions 執行長朴成賢，討論技術合作、投資甚至收購——這是輝達 AI 矽晶片整併行動的最新一步。
-
Industry ENGoogle Hands Marvell a $12.2 Billion Warrant — and Reshapes the Custom Silicon Chessboard
Google struck a sweeping custom-chip agreement with Marvell covering AI inference accelerators and TPU-adjacent silicon — and granted the search giant warrants to buy up to $12.2 billion in Marvell stock, the boldest supplier-financing move yet in the AI chip wars.
-
Industry 中Google 交付 Marvell 122 億美元認股權——重新改寫客製晶片棋局
Google 與 Marvell 簽署涵蓋 AI 推論加速器與 TPU 周邊矽晶片的客製晶片大單，並授予 Google 最高 122 億美元的 Marvell 認股權——這是 AI 晶片戰爭至今最大膽的供應商綁定手法。
-
Tools ENWaymo Opens the Trunk: Custom 5nm ASIC Pushes Robotaxi Brain Past 1,000 TOPS
Waymo has revealed its in-vehicle compute architecture for the first time — a purpose-built 5nm ASIC delivering 1,000+ TOPS for front-end sensor processing, flanked by a seven-supplier list and a dual-redundant design.
-
Industry ENBroadcom Lines Up a $100 Billion AI Debt Machine — and Anthropic Is the First Customer
Broadcom is in talks to raise more than $60 billion in debt for an AI chip financing deal benefiting Anthropic — a senior tranche of $60–70 billion plus $30 billion in junior debt that could push the total toward $100 billion.
-
Industry 中博通集結千億美元 AI 債務機器——Anthropic 是第一位客戶
博通正洽談募集超過 600 億美元債務，用於一項讓 Anthropic 受惠的 AI 晶片融資計畫——600 至 700 億美元優先順位債券加上 300 億美元次順位債券，總規模可能直逼 1,000 億美元。
-
Industry ENNvidia's Reported China Comeback Plan: A Groq-licensed LPU Inference Chip — and a Swift Denial
The Information reports Nvidia is readying a China-specific inference chip built on Groq LPU technology for small-batch shipments by year-end. Nvidia calls the report 'incorrect' — but its denial leaves room to maneuver.
-
Industry 中Nvidia 傳以 Groq 授權 LPU 推理晶片重返中國——隨即遭官方否認
The Information 報導 Nvidia 正開發中國專用推理晶片，採用 Groq LPU 技術、年底前小批量出貨；Nvidia 回應「報導有誤」，但聲明措辭留下了日後翻轉的空間。
-
Policy ENWashington's Ultimatum: Pick Pax Silica or Beijing's AI Bloc
The US is warning 35 signatories of its AI Opportunity Statement that joining China's World AI Cooperation Organization will get them excluded from the US-led coalition — and Beijing is pushing back hard.
-
Policy 中華府最後通牒：在 Pax Silica 與北京 AI 聯盟之間選邊站
美國警告「AI 機會宣言」35 個簽署國：加入中國主導的世界人工智慧合作組織，就會被排除在美國領導的聯盟之外——北京則強力反擊。
-
Industry ENEtched Doubles to a $21 Billion Valuation in a Month as Jane Street Puts Its Sohu Racks to Work
Transformer-only chip startup Etched raised $700M at a $21B valuation led by Jane Street, its first customer — after the quant firm ran live workloads on its first Sohu rack and customer contracts topped $1B.
-
Industry 中Etched 一個月內估值翻倍至 210 億美元，Jane Street 讓 Sohu 機櫃正式上線
專做 transformer 推論晶片的 Etched 完成 Jane Street 領投的 7 億美元融資，估值達 210 億美元——這家量化交易公司在自家機房以 Sohu 晶片跑實際工作負載後，決定同時成為客戶與投資人。
-
Industry ENSK hynix Fires a $29 Billion Signal: Korea's Largest-Ever Buyback Bets the AI Memory Boom Is Here to Stay
SK hynix will buy back and cancel ₩40 trillion of shares in three months — the largest treasury cancellation in Korean history — a blunt reply to investors questioning how long the AI memory supercycle can run.
-
Industry 中SK hynix 投下 290 億美元的訊號：韓國史上最大庫藏股，賭 AI 記憶體榮景不會落幕
SK hynix 將在三個月內買回並註銷 40 兆韓元股份——韓國史上最大規模的庫藏股註銷——直接回應市場對 AI 記憶體超級週期还能持續多久的質疑。
-
Industry ENSamsung Hikes Foundry Prices Up to 15% as AI Demand Shifts Orders to Korea
AI chip demand has filled Samsung's advanced lines, giving its loss-making foundry rare pricing power — with Chinese customers accepting the steepest increases.
-
-
Industry ENFractile Jumps to $6.5B Valuation After Anthropic Signs $250M Chip Deal
The Oxford-born inference chip startup is raising ~$600M at a $6.5B pre-money valuation — a six-fold jump since May — after Anthropic signed an initial $250M supply agreement for its SRAM-based 'memory-compute fusion' chips.
-
Industry 中Anthropic 簽下 2.5 億美元晶片訂單，Fractile 估值三個月暴漲六倍至 65 億美元
牛津出身的英國推論晶片新創 Fractile 正以 65 億美元估值洽談募資約 6 億美元——較五月翻漲逾六倍——主因是 Anthropic 簽下了其 SRAM「記憶體運算融合」晶片的初始供應協議。
-
Industry ENMarvell Grants Google a $12.2 Billion Warrant in Landmark Custom AI Chip Deal
Marvell issued Google a warrant for up to 58.97 million shares at $206.58 — about $12.18B — tied to a custom silicon deal spanning AI inference accelerators, controllers, and near-memory compute for the TPU ecosystem.
-
Industry 中Marvell 授予 Google 122 億美元認股權證，打造客製 AI 晶片里程碑交易
Marvell 向 Google 發行最多 5,897 萬股、每股 206.58 美元的認股權證（約 121.8 億美元），綁定涵蓋 AI 推論加速器、各類控制器與近記憶體運算的 TPU 生態系客製晶片協議。
-
Industry ENChina Quietly Lets Nvidia's H200 Back In — ByteDance and Tencent Each Get 10,000 Chips
An FT exclusive reveals Beijing has allowed small batches of Nvidia's H200 into mainland China, with ByteDance and Tencent each receiving about 10,000 units — a cautious thaw in the AI chip war after months of blocked shipments.
-
Industry 中中國低調放行 Nvidia H200：字節跳動、騰訊各拿到一萬顆
FT 獨家揭露，北京已允許小批 Nvidia H200 進口中國大陸，字節跳動與騰訊近期各自接收約一萬顆——歷經數月封鎖後，AI 晶片戰出現謹慎解凍的跡象。
-
Industry ENChip Rout Goes Global: Samsung and SK Hynix Slump 7% as Bond-Market Angst Spreads
The Philadelphia Semiconductor Index just suffered its worst day since late July, and the pain went global overnight as Samsung and SK Hynix plunged more than 7% in Seoul.
-
Industry 中晶片拋售蔓延全球：債市恐慌下三星與 SK 海力士重挫 7%
費城半導體指數寫下七月底以來最大單日跌幅後，恐慌一夜之間蔓延至亞洲——三星與 SK 海力士在首爾股市雙雙暴跌逾 7%。
-
Industry ENGuardian Investigation: Microsoft Has 2.2M AI Chips Installed — Far Below What Its Own Capacity Claims Imply
Internal documents seen by The Guardian show Microsoft has 2.2 million AI chips installed after a $280B build-out, less than half the ~6.4M GPUs its claimed 10GW of capacity should imply — exposing an industry-wide accountability gap in how AI capacity is measured.
-
Industry 中衛報調查：微軟實際安裝的 AI 晶片僅 220 萬顆——遠低於其自身產能宣稱所隱含的規模
衛報取得的內部文件顯示，微軟在砸下約 2,800 億美元擴建後，實際安裝的 AI 晶片僅約 220 萬顆，不到其宣稱的 10GW 產能所應隱含的 640 萬顆 GPU 的一半——揭露了整個產業在 AI 產能衡量上的問責黑洞。
-
Policy ENTen Thousand H200s Each: ByteDance and Tencent Get Their Nvidia Chips — but Beijing Keeps Them in Hong Kong
FT reveals ByteDance and Tencent have each received ~10,000 Nvidia H200 processors — the first sizeable shipments since US approval — but Chinese regulators are directing the hardware to Hong Kong to shield domestic chipmakers.
-
Policy 中各一萬顆 H200：字節跳動與騰訊拿到輝達晶片——但北京把它們留在香港
FT 獨家揭露：字節跳動與騰訊已各自收到約一萬顆 Nvidia H200 處理器——這是美國放行後首批具規模的出貨——但中國監管機構指示這些硬體必須部署在香港，以保護本土晶片產業。
-
Models ENCerebras Unveils CS-4: A 750-PFLOP Wafer-Scale Rack That Claims Up to 30x GPU Inference Speed
Cerebras's new rack-scale CS-4, built from three WSE-3 Turbo wafers, delivers 750 PFLOPs, 129.6 PB/s of memory bandwidth, and over 4,400 tokens/sec per user on GPT-OSS-120B — up to 30x faster than GPU systems, with first shipments this quarter.
-
Models 中Cerebras 發表 CS-4：750 PFLOPs 晶圓級機架，宣稱推論速度最快可達 GPU 系統的 30 倍
Cerebras 新推出的機架級 CS-4 由三片 WSE-3 Turbo 晶圓構成，提供 750 PFLOPs 運算力、129.6 PB/s 記憶體頻寬，在 GPT-OSS-120B 上每秒每用戶可產出超過 4,400 個 token——最快達 GPU 系統的 30 倍，本季開始出貨。
-
Industry ENVelaura AI Hits $1 Billion Valuation With $110M Series A to Slash AI's Power Bill
Chip IP startup Velaura AI raised a $110M Series A led by Seligman Ventures at a $1B+ valuation, licensing Titan Core technology that cuts AI accelerator power consumption by up to 2x.
-
Industry 中Velaura AI 以 1.1 億美元 A 輪融資躋身獨角獸，要幫 AI 省下電費
晶片 IP 新創 Velaura AI 完成 1.1 億美元 A 輪融資、估值突破 10 億美元，由 Seligman Ventures 領投；其 Titan Core 技術宣稱可將 AI 加速器功耗降低最多 2 倍。
-
Industry ENEtched Doubles to $21 Billion: Jane Street Leads $700M Round and Buys the Racks
Transformer-only chip startup Etched raised $700M led by Jane Street at a $21B valuation — double its July mark — with the trading firm doubling as both lead investor and server-rack customer.
-
Industry 中Etched 估值三週翻倍至 210 億美元：Jane Street 領投 7 億美元融資，同時下單買機櫃
專做 transformer 推論晶片的新創 Etched 以 210 億美元估值完成 7 億美元融資，由 Jane Street 領投——這家量化交易巨頭同時也是其伺服器機櫃客戶，估值較七月翻倍。
-
Industry ENMusk Says Memory, Not GPUs, Is AI's Real Bottleneck — and Memory Stocks Explode
Elon Musk's repeated warnings that memory and storage — not compute — now constrain AI have ignited a global memory-chip rally, with Micron crossing $1,000 and 2027 supply already sold out.
-
Industry 中馬斯克:AI 真正的瓶頸不是 GPU,而是記憶體——記憶體股應聲暴漲
馬斯克在財報電話會議與 X 平台多次示警:記憶體與儲存——而非算力——才是 AI 擴張的真正約束,帶動美光站上 1,000 美元、2027 年產能已被預訂一空。
-
Policy ENWashington's Ultimatum: 35 Countries Told to Pick Pax Silica or Beijing's AI Bloc
A draft State Department letter warns 35 countries they cannot belong to both the US-led Pax Silica coalition and China's Shanghai-based WAICO — with dual-member Kazakhstan as the flashpoint.
-
Industry ENGroq Raises $350M at Half Its Peak Valuation, Completing Its Reinvention as an Inference Neocloud
Bloomberg reports Groq raised $350M at a $3.5B valuation — roughly half its September 2025 peak — with Nvidia joining a round led by Disruptive, capping the LPU pioneer's pivot from chip rival to AI inference data-center operator.
-
Industry 中Groq 以 35 億美元估值募資 3.5 億美元，正式完成從晶片挑戰者到推論雲營運商的轉型
Bloomberg 報導，Groq 以 35 億美元估值完成 3.5 億美元融資——約為 2025 年 9 月高峰的一半——由 Disruptive 領投、Nvidia 參投，為這家 LPU 先驅從 Nvidia 挑戰者轉型為 AI 推論資料中心營運商的旅程寫下新章。
-
Industry ENThe CPU Comeback: Agentic AI's Unexpected Bottleneck Is the Humble Processor
IEEE Spectrum reports that agentic AI has flipped CPUs into the new performance bottleneck — Intel server chips are sold out, AMD doubled forecasts, and AWS is rationing CPU cycles.
-
Industry 中CPU 的逆襲：Agentic AI 意外讓古老處理器成為新瓶頸
IEEE Spectrum 報導指出，Agentic AI 已讓 CPU 成為新的效能瓶頸——Intel 伺服器晶片售罄、AMD 倍增出貨預測、AWS 開始 ration CPU 運算資源。
-
Models ENOpenAI's Ultrafast Mode Runs GPT-5.6 Sol at 750 Tokens per Second on Cerebras
OpenAI's new Ultrafast API tier, powered by Cerebras wafer-scale chips, serves its flagship GPT-5.6 Sol model at up to 750 output tokens per second — up to 14× standard speed.
-
Models 中OpenAI Ultrafast 模式：Cerebras 晶圓級晶片讓 GPT-5.6 Sol 每秒輸出 750 個 Token
OpenAI 推出 Ultrafast API 等級，由 Cerebras 晶圓級晶片驅動，旗下最強模型 GPT-5.6 Sol 每秒可輸出 750 個 token，速度最高達標準版的 14 倍。
-
Policy ENIndia Bets Big: 10 Million Youth to Be Trained in AI Within One Year
From the Red Fort on India's 80th Independence Day, PM Modi announced a nationwide mission to train one crore (10 million) youth in AI skills within a year — the centerpiece of a ₹1 lakh crore high-tech package spanning semiconductors, nuclear power, and jobs.
-
Industry ENBroadcom Sinks 6% as BofA Puts a $370 Billion Price Tag on Its AI Debt Machine
A Bank of America note pegged Broadcom's off-balance-sheet AI chip-financing vehicle at $370B in senior debt by mid-2029 — and the stock's 6% plunge shows markets are starting to price the plumbing of the AI buildout, not just the chips.
-
Industry 中美銀點出博通 AI 債務機器規模達 3,700 億美元，股價應聲重挫 6%
美國銀行分析師估算，博通帳外的 AI 晶片融資平台在 2029 年中前可能累積約 3,700 億美元高級債務——股價單日大跌 6%，顯示市場開始為 AI 建設的「資金管線」定價，而不只是晶片本身。
-
Tools ENOpenAI's Ultrafast Mode Puts GPT-5.6 Sol on Cerebras Silicon at 750 Tokens Per Second
OpenAI's new Ultrafast API tier runs GPT-5.6 Sol on Cerebras accelerators at up to 750 output tokens per second — 14x standard throughput — collapsing the speed-versus-intelligence tradeoff for time-critical AI workloads.
-
Tools 中OpenAI Ultrafast 模式攜手 Cerebras：GPT-5.6 Sol 每秒 750 Token 的推論革命
OpenAI 推出 Ultrafast API 層級，以 Cerebras 晶片驅動 GPT-5.6 Sol，輸出速度可達每秒 750 token——為標準版的 14 倍——徹底打破速度與智慧的取捨困境。
-
Industry ENAI Boom Drives Taiwan's Fastest Economic Growth in Nearly Four Decades
Taiwan lifts its 2026 GDP forecast to 11.05% — the strongest expansion since 1987 — as AI chip demand pushes export growth to a 50-year high.
-
-
Research ENAI Chip Performance Per Dollar Is Growing 49% Per Year, Doubling Every 1.7 Years
Epoch AI's latest data insight reveals AI chip price-performance is accelerating: 49% annual growth since 2023, doubling every 1.7 years — driven by Blackwell's dominance and a shift from logic to memory bottlenecks.
-
Research 中AI 晶片性價比每年成長 49%，每 1.7 年翻倍——Epoch AI 最新數據洞見
Epoch AI 最新數據分析揭示，自 2023 年以來 AI 晶片的性價比每年成長 49%、每 1.7 年翻倍——由 Blackwell 架構主導，但記憶體成本攀升帶來隱憂。
-
Industry ENAnthropic Eyes $6 Billion Decart Acquisition in Bid to Control Inference Costs
Anthropic is in advanced talks to acquire Israeli AI startup Decart for roughly $6 billion, a deal that would bring critical inference optimization technology in-house as compute costs become the dominant battleground for frontier AI labs.
-
Industry 中Anthropic 洽談以 60 億美元收購 Decart，全力掌控推論成本
Anthropic 正與以色列 AI 新創公司 Decart 進行收購談判，金額約 60 億美元。此交易將把關鍵的推論優化技術納入內部，凸顯運算成本已成為前沿 AI 實驗室的主要戰場。
-
Industry ENCerebras Beats Expectations as Cloud Revenue Explodes 281%: The Wafer-Scale Threat to NVIDIA
Cerebras Systems posted Q2 2026 revenue of $209.9M with cloud revenue up 281% YoY, beating estimates. The wafer-scale chipmaker is turning its OpenAI mega-deal into real numbers — even as margin pressure persists.
-
Industry 中Cerebras 財報超預期、雲端營收暴增 281%：晶圓級晶片對 NVIDIA 的最大威脅
Cerebras Systems 公布 2026 年第二季營收 2.099 億美元，雲端營收年增 281%，超出市場預期。這家晶圓級晶片公司正在將與 OpenAI 的超級合約轉化為實質數字——即使利潤率壓力依然存在。
-
Industry ENSuper Micro's Q4 FY2026 Shock: Margins Double, Revenue Nearly Hits $40 Billion, FY2027 Guidance Blows Past Wall Street
Super Micro Computer posted record fiscal 2026 revenue of $39.1B with Q4 gross margins nearly doubling to 17.5%, guided FY2027 to $65–72B — far above consensus — and disclosed a $60B order book amid surging AI infrastructure demand.
-
Industry 中Super Micro FY2026 第四季財報震撼：毛利率翻倍、營收近 400 億美元、FY2027 指引遠超華爾街預期
Super Micro Computer 公布 FY2026 營收 391 億美元的歷史新高，Q4 毛利率近乎翻倍至 17.5%，FY2027 營收指引設定在 650 至 720 億美元，遠超分析師共識，並揭露 600 億美元訂單存量。
-
Industry ENFoxconn's AI Servers Overtake the iPhone: Q2 Earnings Mark a Historic Inflection Point
Foxconn's cloud and networking segment surpassed 50% of revenue for the first time, as Q2 profit jumped 35% to NT$59.97 billion on surging AI server demand.
-
Industry 中鴻海 AI 伺服器超越 iPhone：Q2 財報標誌歷史性轉折點
鴻海雲端網路產品營收佔比首次突破 50%，第二季淨利大增 35% 至 599.7 億新台幣，AI 伺服器需求全面爆發。
-
Industry ENIntel Raises $20 Billion in Historic Stock Offering to Fuel AI Chip Ambitions
Intel upsized its stock offering from $15B to $20B at $95 per share, the largest equity raise in semiconductor history, to fund its AI-driven foundry turnaround under CEO Lip-Bu Tan.
-
Industry 中Intel 募資 200 億美元 寫下半導體史上最大規模股票發行
Intel 將股票發行規模從 150 億美元上調至 200 億美元，每股定價 95 美元，為半導體產業史上最大股權融資，全力推動 CEO Lip-Bu Tan 主導的 AI 晶圓代工復興計畫。
-
Models ENNVIDIA Bets on a Trillion: Nemotron 4 Aims to Be the World's Best Open-Source AI Model
NVIDIA is developing Nemotron 4, an open-source AI model family with a flagship exceeding 1 trillion parameters — roughly double the current Nemotron 3 Ultra — targeting dominance in the open-weight race against DeepSeek, Qwen, and Llama.
-
Models 中NVIDIA 押注一兆參數：Nemotron 4 劍指全球最強開源 AI 模型
NVIDIA 正在開發 Nemotron 4 開源 AI 模型家族，旗艦模型將突破 1 兆參數，約為現有 Nemotron 3 Ultra 的兩倍，目標在開源權重競賽中挑戰 DeepSeek、Qwen 與 Llama 的領先地位。
-
Industry ENSemiconductor Sales Shatter Records: Q2 2026 Hits $403 Billion as AI Devours Every Chip
Global semiconductor sales reached $403.3 billion in Q2 2026 — up 35.1% from Q1 and 123.6% year-over-year in June — as AI-driven demand pushes the industry toward a projected $1.5 trillion year.
-
Industry 中半導體銷售創歷史新高：2026 年第二季突破 4030 億美元，AI 吞噬每一顆晶片
全球半導體銷售額在 2026 年第二季達到 4033 億美元——季增 35.1%，六月單月年增 123.6%——AI 驅動的需求正將整個產業推向 1.5 兆美元的歷史大關。
-
Tools ENGoogle's Pixel 11 Launches with Gemini AI at Its Core
At its Made by Google event, Google unveiled the Pixel 11 series with deep Gemini AI integration, the 2nm Tensor G6 chip, and a new Create A Widget feature.
-
Tools 中Google Pixel 11 發表：Gemini AI 成為核心靈魂
Google 在 Made by Google 發表會上推出 Pixel 11 系列，搭載深度整合的 Gemini AI、2 奈米 Tensor G6 晶片與全新 Create A Widget 功能。
-
Industry ENIntel Raises $20 Billion in Massive Stock Offering to Fuel AI Chip Ambitions
Intel upsized its stock offering from $15B to $20B at $95/share, betting big on surging AI compute demand after a year-long 175% stock surge.
-
Industry 中英特爾募資 200 億美元 大規模釋股全力衝刺 AI 晶片佈局
英特爾將股票發行規模從 150 億美元擴大至 200 億美元，每股定價 95 美元，在股價一年大漲 175% 後押注 AI 運算需求持續爆發。
-
Industry ENAnthropic Joins the Custom Silicon Race: Building In-House Chips for Claude
Anthropic confirms an in-house chip design team for Claude AI, partnering with Samsung and poaching OpenAI's silicon talent to cut inference costs.
-
Industry 中Anthropic 加入自研晶片競賽：為 Claude 打造專屬客製化矽晶片
Anthropic 確認組建內部晶片設計團隊，與三星合作並挖角 OpenAI 矽晶片人才，目標降低 Claude 推論成本。
-
Industry ENSingapore Raises 2026 Growth Forecast to 5.5% as AI Investment Boom Defies Global Headwinds
Singapore upgraded its 2026 GDP growth forecast from 2-4% to 4.5-5.5% as AI-driven semiconductor and electronics demand powered Q2 growth to 5.9% — even as its central bank warns the boom may be unsustainable.
-
Industry 中新加坡上調 2026 年經濟成長預測至 5.5%——AI 投資熱潮無視全球逆風
新加坡將 2026 年 GDP 成長預測從 2-4% 大幅上調至 4.5-5.5%，AI 驅動的半導體與電子需求帶動第二季成長率達 5.9%——但央行同時警告這股熱潮可能無法持續。
-
Industry ENAMD Acquires Taalas: Baking AI Models Directly Into Silicon
AMD's acquisition of Toronto startup Taalas bets that the future of AI inference is not software running on GPUs, but model weights etched into silicon itself.
-
Industry 中AMD 收購 Taalas：將 AI 模型直接刻進矽晶片
AMD 收購多倫多新創 Taalas，押注 AI 推論的未來不是在 GPU 上跑軟體，而是把模型權重直接蝕刻進矽晶片本身。
-
Industry ENAMD Acquires Taalas: Etching AI Models Directly Into Silicon
AMD's acquisition of Toronto chip startup Taalas bets that the future of AI inference lies in hardwiring model weights into dedicated silicon — one chip per model.
-
-
Industry ENAnthropic Goes Silicon: Claude's Creator Builds Its Own Chip Design Team
Anthropic confirmed it is assembling an in-house custom silicon team to design chips for Claude, offering up to $485K salaries and aiming to reduce Nvidia dependence.
-
Industry 中Anthropic 進軍晶片：Claude 創造者組建自有晶片設計團隊
Anthropic 確認正在組建內部自訂矽晶圓團隊，為 Claude 設計專屬晶片，薪資最高達 48.5 萬美元，目標是降低對 Nvidia 的依賴。
-
Industry ENMicrosoft Gears Up to Unveil Maia 300 AI Chip in Bid to Cut Nvidia Dependency
Microsoft is preparing to publicly unveil its next-generation Maia 300 AI inference chip this fall, with talks underway for over 300,000 units from TSMC for 2027 delivery.
-
Industry 中微軟蓄勢待發：Maia 300 AI 晶片今秋亮相，劍指降低對 Nvidia 的依賴
微軟正準備於今秋公開發表次世代 Maia 300 AI 推論加速晶片，並與台積電洽談 2027 年交付超過 30 萬顆的產能。
-
Industry ENTSMC's July Revenue Hits Record NT$467.58 Billion as AI Chip Demand Shows No Signs of Slowing
TSMC posted all-time record July revenue of NT$467.58 billion, up 44.7% year-over-year, as insatiable AI chip demand keeps the world's largest foundry running flat out — with CoWoS capacity sold out through 2027.
-
Industry 中台積電七月營收創歷史新高 NT$4675.8 億——AI 晶片需求毫無放緩跡象
台積電七月營收創下 NT$4675.8 億的歷史新高，年增 44.7%，凸顯 AI 晶片需求持續爆發。CoWoS 先進封裝產能已滿載至 2027 年，全球最大晶圓代工廠正全力運轉。
-
Models ENHuawei's Ascend 950DT Debuts This August: China's Answer to Nvidia Takes Shape
Huawei's Ascend 950DT AI chip — with 144GB HiZQ memory, 4TB/s bandwidth, and native FP8 support — debuts in August 2026, with DeepSeek V4.2 tipped as an early adopter in China's push for AI hardware self-sufficiency.
-
Models 中華為昇騰 950DT 八月登場：中國挑戰輝達的關鍵一步
華為昇騰 950DT AI 晶片配備 144GB HiZQ 記憶體、4TB/s 頻寬與原生 FP8 支援，於 2026 年 8 月亮相，DeepSeek V4.2 可望成為首批採用者，標誌中國 AI 硬體自主化的重大里程碑。
-
Tools ENDiscovered Materials Raises $9M to Let AI Agents Hunt for Cooler Chips
Y Combinator-backed Discovered Materials closed a $9M seed to deploy swarms of AI agents that discover novel semiconductor materials capable of taming the heat crisis in AI data centers.
-
Tools 中Discovered Materials 籌得 900 萬美元，用 AI 智能體尋找更涼的晶片材料
Y Combinator 培育的新創 Discovered Materials 完成 900 萬美元種子輪融資，部署 AI 智能體群來發現新型半導體材料，解決 AI 資料中心日益嚴峻的散熱危機。
-
Industry ENSony and TSMC Bet $6.3 Billion on Next-Gen Image Sensors for AI
Sony and TSMC are pouring $6.3 billion into a joint venture to build next-generation image sensors in Kumamoto, Japan, aiming to supply Apple's iPhones and power the coming wave of physical AI — autonomous vehicles, robots, and smart factories.
-
Industry 中Sony 與 TSMC 豪擲 63 億美元，打造次世代 AI 影像感測器
Sony 與台積電計畫在日本熊本合資 63 億美元，興建次世代影像感測器廠，目標供應 Apple iPhone，並為自駕車、機器人與智慧工廠等物理 AI 應用提供關鍵的感知硬體。
-
Industry ENIntel Raises $15 Billion in Massive Stock Offering to Fuel AI Foundry Ambitions
Intel announced a $15 billion common stock offering — its largest ever — to fund soaring capital expenditures driven by AI chip demand and its 18A foundry ramp.
-
Industry 中Intel 公開募資 150 億美元 全面衝刺 AI 晶圓代工野心
Intel 宣布啟動 150 億美元公開增資——公司史上最大規模——以因應 AI 晶片需求暴增與 18A 奈米製程擴產所需的龐大資本支出。
-
Industry ENSK Hynix Bets $38 Billion on Two New AI Memory Fabs as 2027 Shortage Looms
SK Hynix will invest 54 trillion KRW ($38.1B) in new DRAM and NAND fabs in Yongin and Cheongju, racing to close an AI-driven memory gap its CEO calls the worst shortage in the industry's history.
-
Industry 中SK 海力士豪擲 380 億美元興建兩座 AI 記憶體晶圓廠 迎戰 2027 大缺貨
SK 海力士宣佈投資 54 兆韓元（約 381 億美元）在龍仁與清州興建全新 DRAM 與 NAND 晶圓廠，全力追趕其執行長所稱產業史上最嚴重的 AI 記憶體缺口。