Reports
Report 2026-03-26 158 sources Input Tokens: 92,152 · Output Tokens: 5,121 · Total Tokens: 99,326

AI 戰情報告 (2026-03-26)

metaanthropicopenaigooglenvidiacloudflareoraclegithub claude-codenemotronlyria-3deepseek-v4gemini-3.1gpt-4o-mini agentagentic-commercesecuritydata-centerorchestrationquantum-cryptography

AI 戰情報告 (2026-03-26)

昨日 AI 重點新聞

祖克柏開發 AI Agent 擔任「數位共同執行長」協助管理 Meta ref-127

Mark Zuckerberg 正開發專屬 AI 代理,擔任其「數位 co-CEO」。該代理能深入分析 Meta 內部數據、整理情報並參與策略決策,反映出頂尖科技高層正率先進入 AI 協同管理 模式。

中方模型 DeepSeek 與 MiniMax 在 Token 消耗量上超越美國對手 ref-68

根據 OpenRouter 數據,低成本且效能卓越的中方模型(如 DeepSeek、MiniMax)在過去一個月的 Token 消耗量已領先美國對手。這顯示中方在模型推理效率與商業化滲透上已具備極強競爭力。

OpenAI 領投 Isara 9400 萬美元,攻克「代理軍團」協作技術 ref-92, ref-13

OpenAI 支持的新創公司 Isara 獲得 9400 萬美元融資,旨在開發能協調數千個 AI 代理同時運作的編排軟體。這解決了當前代理系統難以大規模並行與複雜任務分發的瓶頸。

NVIDIA 領軍 Nemotron 聯盟,定義「開放與私有」並存的 AI 未來 ref-120

NVIDIA 發表 Nemotron 聯盟,與 Mistral AI 合作開發開放模型。執行長 Jensen Huang 強調,未來 AI 並非單一模型,而是由多個專門化代理組成的「協奏系統」。

Google 發表 Lyria 3 Pro,音樂生成進入「長曲與結構化」時代 ref-16, ref-41, ref-14

Google 升級其音樂模型至 Lyria 3 Pro,支援生成長達 3 分鐘 的樂曲,並允許使用者精確控制前奏、副歌與間奏等結構,目前已在 Gemini API 開放付費預覽。

法律 AI 新創 Harvey 估值衝上 110 億美元,紅杉資本領投 ref-99, ref-43

專業垂直領域 AI 領頭羊 Harvey 再次獲得 Sequoia 與 a16z 加碼,估值達到 110 億美元。這證明了針對法律等高門檻行業進行深度優化的代理,其商業護城河極深。

川普政府組建 AI 顧問委員會,祖克柏與黃仁勳入列 ref-120, ref-55, ref-127

川普 政府任命 Mark Zuckerberg、Jensen Huang、Larry Ellison 與 Sergey Brin 加入總統科學技術顧問委員會。此委員會將主導美國的 AI 與加密貨幣政策,預示監管環境將更趨向科技巨頭利益。

OpenAI 正式終止 Sora 計畫,算力全面傾注於 Spud 模型 ref-92, ref-141

延續昨日報告,OpenAI 正式關閉 Sora 影片生成項目。資源將全面轉向代號為 「Spud」 的新模型,該模型定位為 Agentic Commerce (代理商務) 的核心,旨在對抗 Anthropic 並主導線上購物的自動化執行。

GitHub Copilot 宣布將使用使用者程式碼訓練 AI 模型 ref-2, ref-3

GitHub 更新政策,從 4 月 24 日起將默認使用 Copilot 的互動數據與程式碼片段進行模型訓練。企業與個人用戶若重視隱私,必須手動進行 Opt-out (退出) 設定。

Google 發表 TurboQuant 算法,將 AI 記憶體需求降低六倍 ref-108, ref-1

Google 揭曉 TurboQuant 壓縮算法,能將 LLM 的 KV 快取壓縮至 3 位元且維持精度。這讓 NVIDIA H100 的推理性能提升達 8 倍,有效緩解目前全球算力與記憶體供不應求的壓力。

Anthropic Claude 電腦使用功能獲市場極大迴響 ref-130, ref-8, ref-56

Anthropic 的 Computer Use 功能被視為其史上最成功的產品發佈。透過 Auto Mode,Claude 能在執行開發任務時自主決策,並透過專屬的安全性掃描機制預防惡意操作,大幅提升了開發者的工作效率。

會議 AI Granola 融資 1.25 億美元,轉型企業代理應用 ref-9

Granola 估值飆升至 15 億美元,其定位從簡單的會議記錄工具轉向「企業 AI 應用」,增加更多 AI Agent 支持以處理會議後的後續執行任務。

Meta Reality Labs 大重組,資源全面轉向 AI 原生 ref-65, ref-75, ref-46

Meta 裁撤 Reality Labs 約 700 名員工,並將剩餘人員重組為 AI-native pods。這標誌著 Meta 的戰略重心已從純虛擬實境轉向由 AI 驅動的增強實境與智慧穿戴設備。

美國國會提出法案,要求暫停新建數據中心 ref-40, ref-93

參議員 Bernie Sanders 與 AOC 提出法案,要求在通過全面 AI 監管法規前,暫停所有新的數據中心建設。此舉引發業界對算力基礎設施增長受阻的擔憂。

Jentic Mini 發表:解決 AI 代理洩漏憑證的資安漏洞 ref-87

針對 OpenClaw 等代理工具容易在對話中洩漏 API 密鑰的問題,Jentic 推出 Jentic Mini。這是一個「憑證防火牆」,讓代理在無需接觸明文密鑰的情況下安全調用企業服務。


市場洞察

  1. AI 代理從「單兵作戰」轉向「軍團編排」 (ref-10, ref-11, ref-48, ref-60)
    延續前幾日對 AI 代理 (Agents) 的趨勢報導,今日 OpenAI 領投 Isara 以及 Sierra 推出自造代理工具,顯示市場焦點已從「如何做一個代理」轉向「如何管理數千個代理」。這種「代理編排 (Orchestration)」技術將成為企業級 AI 的新標配。正如前日報告提到的個人超級智慧,未來的企業運作將是多個 AI 代理組成的協奏系統 (ref-39)。

  2. 中方模型憑藉推理成本優勢在全球生態系中崛起 (ref-9, ref-13, ref-139)
    OpenRouter 的數據證實了一個危險信號:DeepSeek 等中方模型正在憑藉極高的性價比侵蝕美國模型的市佔。儘管美國司法部正加強對晶片的禁運與走私打擊 (ref-13, ref-109),但軟體層面的效能突破已讓中方模型在 Token 消耗量上實現反超。這預示著 2026 年底前,全球 AI 將出現「美國算力 vs 中國演算法」的格局。

  3. 代理商務 (Agentic Commerce) 正式取代影片生成成為營收引擎 (ref-35, ref-53, ref-136)
    OpenAI 殺掉 Sora 是極具戰略意義的動作。這標誌著 AI 市場正從內容生成的「噱頭期」轉向執行交易的「變現期」。當 MIT 探討代理商務需要「事實與上下文」 (ref-35),而 OpenAI 轉向 Spud 購物助手時,2026 年的主戰場已確定為:誰能替使用者在真實世界中完成決策與採購,而非僅僅生成一段影片。

  4. 算力與能源瓶頸引發的政策逆風 (ref-47, ref-78, ref-135)
    延續過去一週對「算力主權」與「AI 工廠」的分析,今日 Bernie Sanders 提出的數據中心禁令法案以及 Micron 在新加坡遭遇的電力變壓器短缺 (ref-78),顯示 AI 的擴張正撞上能源與環境的物理牆。這將迫使軟體商從追求「大模型」轉向追求如 Google TurboQuant 般的「極致效率壓縮」 (ref-83, ref-117)。

Sources 158 items
ref-0 Hacker News Apple randomly closes bug reports unless you "verify" the bug remains unfixed ref-1 Hacker News Ensu – Ente’s Local LLM app ref-2 Hacker News Updates to GitHub Copilot interaction data usage policy ref-3 Hacker News 90% of Claude-linked output going to GitHub repos w <2 stars ref-4 Hacker News False claims in a widely-cited paper ref-5 Hacker News Quantization from the Ground Up ref-6 Hacker News I tried to prove I'm not AI. My aunt wasn't convinced ref-7 Hacker News Health NZ staff told to stop using ChatGPT to write clinical notes ref-8 Hacker News Show HN: A plain-text cognitive architecture for Claude Code ref-9 Hacker News Building a coding agent in Swift from scratch ref-10 Hacker News Show HN: Optio – Orchestrate AI coding agents in K8s to go from ticket to PR ref-11 Hacker News Show HN: Robust LLM Extractor for Websites in TypeScript ref-12 Hacker News The Military Failures of Fascism ref-13 OpenAI Blog Inside our approach to the Model Spec ref-14 Google AI Blog Build with Lyria 3, our newest music generation model ref-15 Google AI Blog Lyria 3 Pro: Create longer tracks in more Google products ref-16 Google DeepMind Lyria 3 Pro: Create longer tracks in more ref-17 Google Developers Blog Closing the knowledge gap with agent skills ref-18 Google Developers Blog Announcing the Colab MCP Server: Connect Any AI Agent to Google Colab ref-19 NVIDIA Blog The Future of AI Is Open and Proprietary ref-20 NVIDIA Blog Blowing Off Steam: How Power-Flexible AI Factories Can Stabilize the Global Energy Grid ref-21 Cursor Blog Run cloud agents in your own infrastructure ref-22 LangChain Blog Skills in LangSmith Fleet ref-23 Vercel Blog Unified reporting for all AI Gateway usage ref-24 Vercel Blog Meet the 2026 Vercel AI Accelerator Cohort ref-25 Vercel Blog New GitHub App permissions for Actions and Workflows ref-26 AWS ML Blog Unlocking video insights at scale with Amazon Bedrock multimodal models ref-27 AWS ML Blog Deploy voice agents with Pipecat and Amazon Bedrock AgentCore Runtime – Part 1 ref-28 AWS ML Blog Reinforcement fine-tuning on Amazon Bedrock with OpenAI-Compatible APIs: a technical walkthrough ref-29 The Rundown AI OpenAI's Sora gets the axe ref-30 Latent Space [AINews] The Biggest Claude Launch of All Time ref-31 Simon Willison datasette-files-s3 0.1a1 ref-32 Simon Willison Thoughts on slowing the fuck down ref-33 Simon Willison datasette-llm 0.1a1 ref-34 Simon Willison LiteLLM Hack: Were You One of the 47,000? ref-35 TechCrunch AI The least surprising chapter of the Manus story is what’s happening right now ref-36 TechCrunch AI Mercor competitor Deccan AI raises $25M, sources experts from India ref-37 TechCrunch AI The AI skills gap is here, says AI company, and power users are pulling ahead ref-38 TechCrunch AI Google unveils TurboQuant, a new AI memory compression algorithm — and yes, the internet is calling it ‘Pied Piper’ ref-39 TechCrunch AI Melania Trump wants a robot to homeschool your child ref-40 TechCrunch AI Bernie Sanders and AOC propose a ban on data center construction ref-41 TechCrunch AI Google launches Lyria 3 Pro music generation model ref-42 TechCrunch AI Reddit takes on the bots with new ‘human verification’ requirements for fishy behavior ref-43 TechCrunch AI Harvey confirms $11B valuation: Sequoia triples down ref-44 TechCrunch AI Granola raises $125M, hits $1.5B valuation as it expands from meeting notetaker to enterprise AI app ref-45 TechCrunch AI Meta launches new initiative to support entrepreneurship, drive AI adoption ref-46 TechCrunch AI Meta turns to AI to make shopping easier on Instagram and Facebook ref-47 TechCrunch AI With Sift, two ex-SpaceX engineers are bringing the software that helped launch rockets to the factory floor ref-48 TechCrunch AI Lucid Bots raises $20M to keep up with demand for its window-washing drones ref-49 The Verge AI Meta is laying off hundreds of employees as it pours money into AI ref-50 The Verge AI Disney’s big bets on the metaverse and AI slop aren’t going so well ref-51 The Verge AI Can you monitor a situation without monitors? The Polymarket sports bar tried ref-52 The Verge AI Reddit accounts with ‘fishy’ bot-like behavior will soon need to prove they’re human ref-53 The Verge AI Google Lyria 3 Pro makes longer AI songs ref-54 The Verge AI Senate Democrats are trying to ‘codify’ Anthropic&#8217;s red lines on autonomous weapons and mass surveillance ref-55 The Verge AI Mark Zuckerberg and Jensen Huang are part of Trump’s new ‘tech panel’ ref-56 The Verge AI Anthropic’s Claude Code gets ‘safer’ auto mode ref-57 Ars Technica Google bumps up Q Day deadline to 2029, far sooner than previously thought ref-58 MIT Technology Review AI This startup wants to change how mathematicians do math ref-59 MIT Technology Review AI Agentic commerce runs on truth and context ref-60 MIT Technology Review AI The AI Hype Index: AI goes to war ref-61 TechMeme How Theia, a satellite imaging startup that raised $250M+ by 2020, collapsed after the founder and four other execs were indicted on federal fraud charges (Brent Crane/Bloomberg) ref-62 TechMeme Back-to-back jury verdicts against Meta could spur a flood of litigation against social media companies and threaten to undermine Section 230 protections (Wall Street Journal) ref-63 TechMeme The US DOJ charged a Chinese national and two Americans for allegedly trying to smuggle ~$62M worth of export-controlled A100 and H100 Nvidia chips into China (Michael Kan/PCMag) ref-64 TechMeme Deccan AI, which supplies post-training data and evaluation work, raised a $25M Series A led by A91 Partners; most of its workforce of experts is based in India (Jagmeet Singh/TechCrunch) ref-65 TechMeme Leaked memo: Meta's Reality Labs is reorganizing staff into AI-native "pods" focused on specific outcomes, flattening the organization's leadership structure (Charles Rollet/Business Insider) ref-66 TechMeme Sources: Reflection AI, which is developing open foundation models, seeks to raise $2.5B at a pre-money valuation of $25B; JPMorgan is in talks to participate (Wall Street Journal) ref-67 TechMeme Filing: Apple settled its lawsuit against former Vision Pro designer Di Liu, who agreed to return confidential documents to Apple and pay monetary damages (Chance Miller/9to5Mac) ref-68 TechMeme OpenRouter data: lower-cost Chinese AI models by groups such as DeepSeek and MiniMax have surpassed US rivals in token consumption since last month (Zijing Wu/Financial Times) ref-69 TechMeme San Francisco has become a laboratory for police surveillance after early resistance; SFPD recorded 700 drone flights last month, up from 93 in February 2025 (Cyrus Farivar/The San Francisco Standard) ref-70 TechMeme Isara, which aims to build software that can coordinate the work of thousands of AI agents, raised $94M and says OpenAI backed the startup at a $650M valuation (Berber Jin/Wall Street Journal) ref-71 TechMeme GitHub says it will use Copilot interaction data, including inputs, outputs, and code snippets, to train its AI models starting April 24, unless users opt out (Corbin Davenport/How-To Geek) ref-72 TechMeme Google sets a 2029 deadline for its post-quantum cryptography migration, aiming to "secure the quantum era" as "frontiers may be closer than they appear" (Dan Goodin/Ars Technica) ref-73 TechMeme Musk's lawyers ask a Delaware judge to step back from cases involving him, after her account "liked" a LinkedIn post celebrating his defeat in a California case (Sujeet Indap/Financial Times) ref-74 TechMeme Singapore-based Startale, developer of the Strium blockchain for tokenized securities and JPYSC and USDSC stablecoins, raised a $63M Series A from SBI and Sony (Francisco Rodrigues/CoinDesk) ref-75 TechMeme Source: Meta on Wednesday laid off around 700 employees in the Reality Labs unit, as well as some in recruiting, sales, and Facebook (Eli Tan/New York Times) ref-76 InfoQ AI/ML QCon London 2026: Tools That Enable the Next 1B Developers ref-77 InfoQ AI/ML Uber Launches IngestionNext: Streaming-First Data Lake Cuts Latency and Compute by 25% ref-78 InfoQ AI/ML Podcast: [Video Podcast] Agentic Systems Without Chaos: Early Operating Models for Autonomous Agents ref-79 InfoQ AI/ML AWS Load Balancer Controller Reaches GA with Kubernetes Gateway API Support ref-80 InfoQ AI/ML Presentation: Panel: Security Against Modern Threats ref-81 InfoQ AI/ML .NET 11 Preview 2 Updates MAUI with Performance Improvements and Platform Refinements ref-82 The New Stack Edera spent years calling KVM less secure. Here’s why it changed its mind. ref-83 The New Stack Your Kubernetes isn’t ready for AI workloads, and drift is the reason ref-84 The New Stack Fivetran donates its SQLMesh data transformation framework to the Linux Foundation ref-85 The New Stack HPE’s AI agents cut root cause analysis time in half ref-86 The New Stack Why online stores keep showing the wrong products — and why tensors fix it ref-87 The New Stack OpenClaw’s biggest security flaw is why Jentic Mini exists ref-88 The New Stack Claude can now open apps, click buttons, and complete tasks on your Mac — but Anthropic says risks remain ref-89 The New Stack Why most AI projects fail after the demo actually works ref-90 Wired AI There’s Something Very Dark About a Lot of Those Viral AI Fruit Videos ref-91 Wired AI OpenClaw Agents Can Be Guilt-Tripped Into Self-Sabotage ref-92 Wired AI OpenAI Enters Its Focus Era by Killing Sora ref-93 Wired AI New Bernie Sanders AI Safety Bill Would Halt Data Center Construction ref-94 Wired AI Arm’s CEO Insists the Market Needs His New CPU. It Could Piss Everyone Off ref-95 TechCrunch Startups Delve did the security compliance on LiteLLM, an AI project hit by malware ref-96 TechCrunch Startups Riding the GLP-1 boom, VITL lands $7.5M to overhaul cash-pay clinic prescribing ref-97 TechCrunch Startups How soap opera-TikTok hybrids became a billion-dollar business ref-98 TechCrunch Startups ReelShort made $1.2 billion on werewolf romances. Watch Club wants to do it better. ref-99 TechCrunch Startups After pivoting, Y Combinator grad Glimpse raises $35M led by a16z ref-100 TechCrunch Startups Harbinger’s next product will be hybrid emergency vehicles ref-101 TechCrunch Startups Arinna raises $4M seed round to solve the space power problem ref-102 Stratechery Arm Launches Own CPU, Arm’s Motivation, Constraints and Systems ref-103 Tom's Hardware Geekbench 6 warns about inconsistent benchmarking performance from new Core Ultra 200S Plus chips — says Intel's IPC boosting Binary Optimization Tool modifies scores in 'unclear' fashion ref-104 Tom's Hardware Micron's $24 billion Singapore fab could need 500 transformers, more than double the output of any single manufacturer — heavy electrical infrastructure the latest AI buildout bottleneck ref-105 Tom's Hardware Samsung’s 870 EVO SATA SSD quietly gets 8TB variant despite storage shortage and skyrocketing pricing — new model spotted in Europe for €1,300 with higher cache and endurance ref-106 Tom's Hardware Intel Arc Pro B70 and Arc Pro B65 GPUs bring 32GB of RAM to AI and pro apps — bigger Battlemage finally arrives, but it's not for gamers ref-107 Tom's Hardware PC makers face shortages of Intel and AMD CPUs that stretch up to six months — lead time for orders jumps from just two weeks in the face of AI demand ref-108 Tom's Hardware Google's TurboQuant reduces AI LLM cache memory capacity requirements by at least six times — up to 8x performance boost on Nvidia H100 GPUs, compresses KV caches to 3 bits with no accuracy loss ref-109 Tom's Hardware Seagate FireCuda 530R 2TB SSD Review: A stealthy workstation powerhouse ref-110 Tom's Hardware Intel officially releases Xeon 600 chips, announces new vPro Panther Lake CPUs — ‘all-new’ vPro platform goes all-in on AI ref-111 Tom's Hardware Start your AM5 build off right with an incredible deal on a Ryzen 9 9900X processor — get $185 off and walk away with the centerpiece of your new or upgraded PC for only $314.99, an all-time low ref-112 Tom's Hardware Kentucky farm family rejects $26 million offer for 600 acres of land from unnamed AI data center suitor — declines 7x offer, wants to ‘Stay and hold and feed a nation’ ref-113 Tom's Hardware Gigabyte MO27Q28G 27-inch 280 Hz OLED gaming monitor review: Bright, colorful, and quick ref-114 Tom's Hardware Grab this $8.98 TP-Link gigabit Ethernet switch for lag-free 4K streaming and gaming — save nearly 50% on silent unmanaged 5-port hub to instantly expand your LAN ref-115 Tom's Hardware The Super Micro AI accelerator smuggling scandal proves how cut-throat the global AI race has become — as global trade evolves, so does export control evasion ref-116 Tom's Hardware Russian ‘Starlink Rival’ established with 16 satellites launched, aims for 900 by 2035 — commercial operation to begin next year with 250 sputniks ref-117 Tom's Hardware Flipper Zero pen-testing tool gets an AI-powered companion app — natural language interface allows for faster, easier hacking ref-118 Tom's Hardware HP's new AI workstation can expand horizontally to add 15% more internal volume — alternate side panel also includes more active cooling. ref-119 Politico Tech Meta, YouTube found liable for social media addiction in landmark trial ref-120 Politico Tech Jensen Huang and Mark Zuckerberg among tech leaders appointed to White House advisory council ref-121 The Hacker News (Security) LeakBase Admin Arrested in Russia Over Massive Stolen Credential Marketplace ref-122 The Hacker News (Security) GlassWorm Malware Uses Solana Dead Drops to Deliver RAT and Steal Browser, Crypto Data ref-123 The Hacker News (Security) The Kill Chain Is Obsolete When Your AI Agent Is the Threat ref-124 The Hacker News (Security) Russian Hacker Sentenced to 2 Years for TA551 Botnet-Driven Ransomware Attacks ref-125 The Hacker News (Security) Device Code Phishing Hits 340+ Microsoft 365 Orgs Across Five Countries via OAuth Abuse ref-126 Lobsters Your First Parser ref-127 Google News Mark Zuckerberg is building an AI agent to help run Meta as a digital 'co-CEO' ref-128 Google News Oracle expands AI Agent Studio for Fusion Applications ref-129 Google News Mystery AI model Hunter Alpha may be DeepSeek V4 in disguise ref-130 Google News Anthropic Introduces ‘Auto Mode’ for Claude to Advance Autonomous AI Coding with Built-In Safeguards ref-131 Google News Nvidia plans open-source AI agent platform 'NemoClaw' for enterprises: Wired ref-132 Google News AI model reads cardiac MRI scans with near expert accuracy ref-133 Google News Reco Launches AI Agent Security to Tackle AI Sprawl and Agent Sprawl Across the SaaS Stack ref-134 Google News Cloudflare launches Dynamic Workers for AI agent execution ref-135 Google News Domo Launches AI Agent Builder and MCP Server to Connect Enterprise Data to the AI Ecosystem ref-136 Google News Introducing Oracle AI Agent Memory: A Unified Memory Core for Enterprise AI Systems ref-137 Google News Mistral Forge Takes Aim at RAG. But Who Actually Needs Custom Models? ref-138 Google News Apple Readies Introduction of AI Agent-Like Siri ref-139 Google News OpenAI Launches Safety Bug Bounty Program Targeting AI Agent Vulnerabilities ref-140 Google News Mercedes puts Zhipu AI model in new Maybach to woo Chinese buyers ref-141 Google News Generative AI: OpenAI to Shut Down Sora, Disney Investment Canceled ref-142 Google News Exclusive | OpenAI Backs New AI Startup Seeking Bot Army Breakthroughs ref-143 Google News AI Startup Sierra Unveils Self-Service Agent-Building Product ref-144 Google News Legal AI startup Harvey valued at $11 billion in funding round, as VCs spread bets beyond model companies ref-145 Google News Mitsubishi Electric Invests in AI Startup Sakana AI ref-146 Google News Microsoft and Nvidia announce AI partnership to fast-track nuclear power plants ref-147 Google News China bans founders of AI startup Manus from travelling after deal with Meta ref-148 Google News ARM gains on plan to sell own chips for first time in push for AI revenue ref-149 Google News New Bernie Sanders AI Safety Bill Would Halt Data Center Construction ref-150 Google News Decoding the 2026 White House AI Blueprint: U.S. AI Policy Starts to Take Shape ref-151 Google News Double win: Washington Gov. Ferguson signs two major AI safety bills into law ref-152 Google News Anthropic reopens talks with Pentagon after AI safety feud ref-153 Google News Trump names Nvidia, Meta and Oracle CEOs to council that will influence AI policy ref-154 Google News From Scale to Optimization: GTC 2026 Signals the Next Phase of AI Infrastructure ref-155 Google News Feds Arrest Trio for Nvidia GPU Smuggling Scheme Involving Supermicro Servers ref-156 Google News Arm Enters AI Chip Market with Launch of Arm AGI CPU in Strategic Shift ref-157 Google News SK Hynix files for US listing to fund AI chip-driven expansion