Reports
Report 2026-06-27 147 sources Input Tokens: 96,176 · Output Tokens: 6,254 · Total Tokens: 105,252

AI 戰情報告 (2026-06-27)

openaianthropicvercelawsshopifydeepseekappleintel gpt-5.6mythos-5fable-5jalapeñoclaude-4.8gemma-4 agent-infrastructuresession-awaremcpchip-warverifiable-executionsupply-chain

AI 戰情報告 (2026-06-27)

昨日 AI 重點新聞

OpenAI 發表 GPT-5.6 系列模型 Sol/Terra/Luna,受美國政府嚴格管控存取權 ref-19, ref-61, ref-0

OpenAI 正式揭曉其下一代模型 GPT-5.6 預覽版,包含旗艦模型 Sol、平衡型 Terra 與輕量級 Luna。受川普政府要求,此模型目前僅限於少數經政府審核通過的「受信任夥伴」使用,旨在控制其在網路安全與生物領域的高階能力。據悉,Sol 模型在編碼任務上超越了 Mythos 5,但在網路攻擊測試中未達到「網路危機」臨界值,這被視為一種安全權衡。

Anthropic Mythos 5 獲准重啟,供美國關鍵基礎設施機構使用 ref-25, ref-2, ref-60

在經歷兩週的艱難談判後,美國商務部解除對 Anthropic 最強模型 Mythos 5 的封鎖,授權超過 100 家 操作關鍵基礎設施的美國機構使用。然而,面向公眾的 Fable 5 仍處於審查階段,恢復時間尚未確定,這顯示政府對「公眾存取高階 AI」仍持極度謹慎態度。

OpenAI 與 Broadcom 合作推出首款自研推理晶片「Jalapeño」 ref-141, ref-28, ref-30

OpenAI 揭露與 Broadcom 共同研發的客製化 ASIC 晶片 Jalapeño,主要用於加速 ChatGPT 及其代理產品的推理任務。此舉顯示 OpenAI 正在加速垂直整合硬體供應鏈,以降低對 Nvidia 的長期依賴並優化推理成本。

Vercel Ship 2026:發布開源代理框架 eve 與 Agent Stack 基礎設施 ref-10, ref-50

Vercel 在倫敦大會上發表專為 AI 代理設計的完整基礎設施,包含 Agent Stack 與開源框架 eve。該系統提供 Durable Workflow、沙箱運算與安全連接器 (Vercel Connect),讓代理能以單一目錄結構在隔離環境中執行代碼與自動部署軟體。

AWS、Microsoft、Google 推動「Session-aware」代理運行架構變革 ref-144

雲端三巨頭與 Anthropic 共同轉向「會話覺知 (Session-aware)」的代理執行模式。這意味著雲端負載平衡已從單次 API 請求,演進為針對長期運行的代理會話進行隔離與狀態管理,以支撐需要保存歷史脈絡與工具狀態的複雜代理工作流。

Lindy 因成本壓力將底層模型從 Claude 全面切換至 DeepSeek ref-132, ref-25

AI 助手新創 Lindy 宣布捨棄 Anthropic 模型,全面改用 DeepSeek。創辦人表示此舉能為公司節省數百萬美元的代幣支出,反映出市場在「性能領先」與「推理經濟學」之間的權衡正向高性價比模型傾斜。

Stripe 與 Cara 展示金融與保險業之生產級代理實踐 ref-17, ref-49

Stripe 透過 ReAct 框架 建立的合規代理將審核處理時間縮短了 26%;保險科技 Cara 則利用 AWS Bedrock 自動化保險經紀人的繁雜工作流。這些案例證實垂直領域代理已具備處理高敏感度、多步驟決策的能力。

代理身分標準化:中國發布首個代理連通標準與「數位身分證」 ref-52

中國 推出首個 AI 代理連通性標準,為每個代理配發「數位身分證」以供識別與授權管理。同時,Dapr 1.18 亦導入 Verifiable Execution,利用加密技術確保代理的執行軌跡不可篡改且具備來源證明。

Shopify Catalog 利用 AI 解決購物代理的商品重複辨識難題 ref-14

Shopify 推出 Catalog 系統,利用 LLM 將不同商家的產品數據進行歸一化與分群。這讓 AI 購物代理能精準識別不同頁面上的相同商品,大幅優化了「代理電商 (Agentic Commerce)」的搜尋與比價體驗。

SpaceX 簽署 63 億美元算力合約支援開源 AI 實驗室 Reflection ref-134

馬斯克的 SpaceX 簽署了一筆金額龐大的運算能力協議,將為開源 AI 啟動計畫 Reflection 提供價值高達 63 億美元 的硬體資源。這延續了馬斯克近期對 AI 工程與算力基礎設施的大規模佈局。

AWS 宣佈調漲 Nvidia GPU 實例租賃價格 20% ref-145, ref-49

Amazon Web Services 通知客戶,將於 7 月 1 日起調漲 EC2 GPU 算力價格約 20%。值得注意的是,AWS 自研的 Trainium 晶片價格維持不變,反映出雲端大廠正引導客戶轉向自研硬體以應對 Nvidia 供應鏈的溢價壓力。

MIT 研發「Masked IRL」提升機器人對模糊指令的理解力 ref-139

MIT 研究人員利用 LLM 開發出一種新訓練方法,讓機器人能自動補全模糊指令(例如將「靠近一點」補全為「靠近桌面面層」)。透過屏蔽無關環境細節,機器人只需極少量的示範數據,即可在複雜環境中安全地執行家務任務。

業界聯合成立 Akrites:以 AI 技術防禦 AI 驅動的開源漏洞威脅 ref-29

Linux 基金會 攜手 AWS、OpenAI、Anthropic 等巨點成立 Akrites 組織。該組織旨在應對 AI 模型快速掃描並發現開源漏洞的風險,從傳統的「分散發現」轉向「集體修復與披露」的防禦模式。

Baseten 完成 15 億美元融資,專注 AI 推理架構優化 ref-117

AI 推理軟體供應商 Baseten 完成 15 億美元 融資,估值達 130 億美元。這顯示資本市場對「模型執行層」與「基礎設施優化」的關注度已超越單純的模型研發。

亞馬遜披露 MCP 服務器實例:實現 S3 PDF 數據交互式提取 ref-101

AWS 發布技術指南,展示如何利用 Model Context Protocol (MCP) 在 S3 存儲中建立交互式 PDF 文本提取伺服器。此技術讓 AI 代理能直接從雲端文件庫中精準定位並提取法律或合規條款,無需繁重的批次處理管線。

市場洞察

1. 代理運行架構從「請求」轉向「會話 (Session)」

延續前幾日對於 AgentCore 與代理化管理的討論,今日 AWS、Microsoft、Google 同步轉向 Session-aware 架構 (ref-88),這是一個關鍵的技術分水嶺。傳統雲端架構是為了處理短暫、無狀態的 API 請求設計的,但 AI 代理需要長期運作、管理工具狀態並保存對話脈絡。Vercel eve (ref-39, ref-75) 的發布更進一步將此概念框架化,預示著「代理託管 (Agent Hosting)」將成為獨立於 Web 託管的新興利基市場。

2. 模型出口管制引發的「主權審核」常態化

從 Anthropic Mythos 5 獲准重啟但受限 (ref-1, ref-109),到 OpenAI GPT-5.6 實施「客戶對客戶 (Customer-by-customer)」的逐一審核制 (ref-14, ref-81),顯示高階 AI 模型已完全等同於戰略軍用物資。這種趨勢正推動非美系模型(如 DeepSeek)的快速成長,因為企業開發者不願承擔模型隨時可能因政策「斷供」的風險,昨日提到的「模型主權」競爭已進入深水區。

3. 推理經濟學與供應鏈的垂直整合

延續前日對於記憶體漲價與晶片短缺的分析,今日 Apple 遊說政府採購中資 CXMT 記憶體 (ref-0) 以及 OpenAI 發表 Jalapeño 自研晶片 (ref-4, ref-32),均顯示 AI 巨頭正竭力擺脫單一供應商限制。同時,Lindy 轉投 DeepSeek (ref-94) 顯示在 B2C/B2B 應用層,廠商對「單位推理成本」的敏感度已超越對「模型品牌」的忠誠度,推理效能優化與成本控制將是未來一年的競爭主軸。

4. AI 代理身分與可驗證執行的標準化

隨著 AI 代理在企業內部權限的擴張,身分認證與執行軌跡的不可篡改性變得至關重要。中國推出的「代理身分證」標準 (ref-120) 與 Dapr 1.18 的可驗證執行 (ref-27),均是為了解決代理行為的「歸責性」問題。這與昨日提到的「代理安全防禦」趨勢吻合:唯有解決了「誰在動數據」與「這動作是否合法」的驗證,AI 代理才能真正進入核心商務決策流程。

Sources 147 items
ref-0 Hacker News U.S. government will decide who gets to use GPT-5.6 ref-1 Hacker News Previewing GPT‑5.6 Sol: a next-generation model ref-2 Hacker News U.S. allows Anthropic to release Mythos AI to ‘trusted’ US organizations ref-3 Hacker News Ultrasound imaging of the brain ref-4 Hacker News The gap between open weights LLMs and closed source LLMs ref-5 Hacker News Show HN: Smart model routing directly in Claude, Codex and Cursor ref-6 Hacker News AI in mathematics is forcing big questions ref-7 Hacker News Gossamer: a Rust-flavoured language with real goroutines and pause-free memory ref-8 Hacker News Show HN: Hacker News on a train station-style flip board ref-9 Hacker News Anatomy of a Failed (Nation-State?) Attack ref-10 Hacker News AgentKits – 60 production-ready AI agent blueprints with guardrails ref-11 Hacker News Show HN: Autofit2 – End-to-end pipeline for multilingual text classification ref-12 Google Developers Blog How A2A is Building a World of Collaborative Agents ref-13 Google Developers Blog DiffusionGemma: The Developer Guide ref-14 Vercel Blog Vercel Ship 2026 recap ref-15 AWS ML Blog Build interactive PDF text extraction from Amazon S3 ref-16 AWS ML Blog How Cara pioneers domain-specific AI for enterprise insurance brokerages with AWS ref-17 AWS ML Blog Production-grade AI agents for financial compliance: Lessons from Stripe ref-18 The Rundown AI White House reins in OpenAI's GPT-5.6 ref-19 Latent Space [AINews] OpenAI GPT-5.6 Sol / Terra / Luna — restricted to trusted partners ref-20 Simon Willison Quoting Dean W. Ball ref-21 Simon Willison Quoting Timothy B. Lee ref-22 Simon Willison What happened after 2,000 people tried to hack my AI assistant ref-23 Simon Willison Incident Report: CVE-2026-LGTM ref-24 Simon Willison Quoting OpenAI ref-25 TechCrunch AI Trump Admin releases Anthropic Mythos to be used by more than 100 US companies, agencies ref-26 TechCrunch AI OpenAI limits GPT-5.6 rollout after government request, says restrictions shouldn’t be the norm ref-27 TechCrunch AI OpenAI poaches Uber India chief to lead its biggest market outside the US ref-28 TechCrunch AI Why everyone from OpenAI to SpaceX is building their own chips (and turning up the heat on Nvidia) ref-29 TechCrunch AI It’s not about Anthropic vs. OpenAI anymore ref-30 TechCrunch AI OpenAI’s Jalapeño chip is Big Tech’s spiciest move away from Nvidia ref-31 TechCrunch AI Early Bird pricing ends tonight for TechCrunch Founder Summit ref-32 The Verge AI Anthropic’s Mythos 5 is back ref-33 The Verge AI OpenAI unveils GPT-5.6 amid US AI regulatory drama ref-34 The Verge AI Anthropic’s Mythos mess is only getting worse ref-35 TechMeme A look at advanced chip packaging, now more reliant on TSMC and its partners in Taiwan than ever, and the efforts to address this bottleneck in the US (Don Clark/New York Times) ref-36 TechMeme Sources: Meituan, Baidu, Xiaomi, and other Chinese tech giants have been trimming their workforces, fueling Chinese workers' concerns of being replaced by AI (Wency Chen/South China Morning Post) ref-37 TechMeme Japanese financial giant SBI agrees to acquire Bitbank, a top 10 Japanese crypto exchange by trading activity, for ~$289M, with the deal set to close in October (Jamie Crawley/CoinDesk) ref-38 TechMeme Source: Intel has promised to deliver SpaceX and Apple a toolkit this fall to test its 14A node before they make final commitments to produce chips with Intel (Tripp Mickle/New York Times) ref-39 TechMeme Insurance tech startup Corgi denies accusations that it used Papermark's open source software code to develop its software and present it as its own (Julie Bort/TechCrunch) ref-40 TechMeme Sources: the CFTC began an extensive investigation of Polymarket earlier this year; the agency's former acting head killed a separate investigation in July 2025 (New York Times) ref-41 TechMeme Sources: Apple is lobbying the Trump admin for clearance to buy memory chips from US-blacklisted Chinese company CXMT to ease pressure from rising chip prices (Financial Times) ref-42 TechMeme Sources: SpaceX and Charter have held executive-level talks on a consumer mobile phone offering, which would help SpaceX become a DTC mobile phone provider (Kelcee Griffis/Bloomberg) ref-43 TechMeme Anthropic says the US government is allowing Mythos 5 to be redeployed to critical infrastructure operators, and is working to restore general access to Fable 5 (@anthropicai) ref-44 TechMeme Sources: Meta lobbyists are urging California lawmakers to exempt social media platforms from legislation that would increase penalties in child-harm cases (Tyler Katzenberger/Politico) ref-45 TechMeme Letter: the US lifts its block on Mythos 5, allowing Anthropic to release it to more than 100 US institutions; sources: talks about Fable 5 are ongoing (Semafor) ref-46 TechMeme Sources: Anthropic and the White House are moving closer to an agreement to lift US restrictions on Fable 5 and Mythos 5 (Bloomberg) ref-47 TechMeme How AI-native law firms use "management services organization" structures to access capital historically barred from US law firms, including PE and VC funds (Stephen Foley/Financial Times) ref-48 TechMeme Sources: Zuckerberg urged execs to explore Polymarket and Kalshi partnerships, as the Arena prediction app targets 100M monthly active "predictors" aged 18-34 (Mike Isaac/New York Times) ref-49 TechMeme AWS hikes prices for Nvidia GPUs in its EC2 Capacity Blocks service, which let businesses rent AI compute in advance, by 20%; Trainium chip pricing is unchanged (Catherine Perloff/The Information) ref-50 InfoQ AI/ML Vercel Introduces Eve, an Open-Source Framework for Building AI Agents ref-51 InfoQ AI/ML Presentation: AI Works, Pull Requests Don’t: How AI Is Breaking the SDLC and What To Do About It ref-52 InfoQ AI/ML Dapr 1.18 Introduces Verifiable Execution, Bringing Cryptographic Trust to AI Agents and Workflows ref-53 InfoQ AI/ML Argo CD 3.5 Tightens Supply Chain Security with Internal mTLS and Source Integrity ref-54 The New Stack After Fable 5 ban, Anthropic and 19 organizations launch open source security body ref-55 The New Stack The US government just told OpenAI who’s allowed to use the next GPT 5.6 model ref-56 The New Stack AWS, Microsoft, and Google agree the session is the new unit of compute. They disagree on how to isolate it. ref-57 The New Stack Your engineering org needs an AI slop registry ref-58 The New Stack Shopify taught AI to spot duplicate products. Here’s why retailers are scrambling. ref-59 The New Stack The AI agent identity problem nobody’s talking about ref-60 Wired AI Trump Administration Allows Anthropic to Release Mythos to Select US Organizations ref-61 Wired AI OpenAI Has New AI Models. Here’s Why You Can’t Use Them ref-62 Wired AI Europe Is Fed Up and Wants Its Own AI ref-63 Wired AI How Qatar Became FIFA’s Technology Test Lab ref-64 TechCrunch Startups Corgi, the buzzy Y Combinator-backed insurance tech startup, says it didn’t steal an open source product ref-65 TechCrunch Startups Robotaxis drive miles just to get cleaned and charged; this new startup wants to fix that ref-66 Crunchbase News The Week’s 10 Biggest Funding Rounds: AI Drives Another Spree Of Megadeals ref-67 Crunchbase News 6 Startup Investors On What It Will Take To Fund More Black Founders ref-68 Stratechery 2026.26: Summer Vibes ref-69 Tom's Hardware Incredible Ryzen 7 9800X3D prebuilt deal comes with an RX 9070 XT and 32GB of DDR5 for $750 off — get a prime iBuyPower 4K gaming rig for just $1,749 ref-70 Tom's Hardware TP-Link's 8-port 2.5G unmanaged Ethernet switch is a smokin' bargain at $50 — upgrade your home network for half price ref-71 Tom's Hardware Gigabyte Aero X16 gaming laptop delivers RTX 5060, 32GB RAM, and a 16-inch 165Hz 1600p display for 21% off ref-72 Tom's Hardware This is what Tom's Hardware readers clicked the most during Prime Day — Portable displays, DVD burners, the Nintendo Switch 2, and two decent SSD deals, among others ref-73 Tom's Hardware Save $250 on this AMD AM5 bundle, get Samsung 990 Pro SSD for basically free — grab the AMD Ryzen 5 9600X, Gigabyte B850 Aorus Elite motherboard, and 990 Pro 1TB SSD for 36% off ref-74 Tom's Hardware Sony WH-1000XM5 active noise-canceling headphones for an all-time low $198 at Amazon — audiophile-grade audio and ANC for an affordable price ref-75 Tom's Hardware Meta Quest 3S drops to pre-RAMpocalypse prices for a Prime Day — wireless VR headset goes on sale for a limited time, get 15% off and save more than $50 if you get it right now ref-76 Tom's Hardware The Logitech G29 racing wheel and pedal set with real force feedback is on a deep 40% discount for Prime Day — get a racing sim setup for just $180 ref-77 Tom's Hardware Asus' ROG Zephyrus G16 with an RTX 5070 & 240Hz OLED display is on sale for just $1,575 — Save $575 on an excellent gaming laptop with a 16-core flagship CPU & top-tier build quality ref-78 Tom's Hardware Onsemi buying cash-strapped Synaptics in $7 billion all-stock deal — smart power meets edge AI hardware ref-79 Tom's Hardware The Ryzen 7 5800X3D is tough to find, but these Prime Day CPUs with DDR5 are cheaper — offset the cost of a DDR5 upgrade with a CPU discount ref-80 Tom's Hardware OpenAI's ChatGPT-5.6 gets the same banhammer treatment as Anthropic’s Mythos from the federal government — source says that Washington cautioned OpenAI against releasing the model without receiving approval ref-81 Tom's Hardware Amazon Eero Max 7 Wi-Fi 7 mesh router 4-pack price slashed by 50 percent — flagship Wi-Fi 7 at a major discount for Prime Day ref-82 Tom's Hardware Framework's Laptop 13 Pro DIY Edition now costs less than before, but a CPU price hike might be coming — Cheaper PCIe 5.0 drives from Adata upgrade customers from 500GB to 1TB for free ref-83 Tom's Hardware Last chance to save up to 55% on these brilliant Hoto tools for PC builders and hobbyists, starting from $14 — super low prices set to end soon on cordless electric screwdrivers, drills, flashlights, vacuum cleaners, and more ref-84 Tom's Hardware Apple will skip its high-end M6 Mac chips and fast-track an AI-focused M7 generation for 2027, report claims — may release a base M6 chip for entry-level Macs this year ref-85 Tom's Hardware Solidigm VP talks PCIe 6.0 SSDs, next-gen floating gate NAND, liquid cooled storage and more — Avi Shetty, VP of AI, Solutions & Market Enablement discusses the future of enterprise storage tech ref-86 Tom's Hardware AMD Ryzen 7 5800X3D vs Intel Core i7-14700K faceoff — A new battle for DDR4 supremacy in 2026 ref-87 Tom's Hardware AMD's new 10th Anniversary Ryzen 7 5800X3D CPU back in stock — Here's where to buy the revamped processor that's the fastest gaming chip available for AM4 platforms [Updated] ref-88 Tom's Hardware Microsoft extends free Windows 10 security updates for a second year — program now ends on October 12, 2027 ref-89 Tom's Hardware Digitize your library with these Blu-Ray or DVD drives from as little as $27 in these Prime Day deals — revive your old DVD/Blu-Ray collection with these drive deals ref-90 Tom's Hardware IBM goes sub-1nm, develops 0.7nm-class technology — offering up to 50% higher performance and 70% higher energy efficiency compared to IBM's 2nm-class node ref-91 Tom's Hardware Corsair's "Bundle and Save Big" sale slashes prices on high-end gaming PCs — save up to $300 on premium Corsair gear ref-92 Tom's Hardware Keep your gaming keyboard and PC fans clean with these electric air blower deals, starting from $17 — high-RPM cordless air dusters are great replacements for canned air to blast away the dust and dirt ref-93 Tom's Hardware Rosewill’s M.2 SSD Cloner and Eraser drops to its lowest-ever price of $47 — become an IT hero, or just save yourself some time and frustration ref-94 Politico Tech Trump administration partially lifts Anthropic’s AI export ban ref-95 Politico Tech Key Democrats urge House to reject kids’ safety proposal ref-96 Politico Tech Online safety coalition urges House to reject KIDS Act compromise ref-97 The Hacker News (Security) FBI Warns Russian Intelligence Hackers Target Signal Backup Recovery Keys ref-98 The Hacker News (Security) New SharkLoader Malware Deploys Cobalt Strike in StrikeShark Cyberattacks ref-99 The Hacker News (Security) Chinese-Speaking APT Deploys New TinyRCT Backdoor in Southeast Asia Campaign ref-100 The Hacker News (Security) New Linux pedit COW Exploit Enables Root Access by Poisoning Cached Binaries ref-101 The Hacker News (Security) Amazon Q Developer Flaw Could Let Malicious Repos Run Code via MCP Configs ref-102 The Hacker News (Security) CISA Adds Exploited PTC Windchill RCE Flaw to KEV as Web Shell Attacks Continue ref-103 The Hacker News (Security) New DirtyClone Linux Kernel Flaw Lets Local Users Gain Root via Cloned Packets ref-104 The Hacker News (Security) Guardian Agents: The Next Layer of Identity Governance ref-105 The Hacker News (Security) Miasma Malware Targets npm Packages and GitHub Actions in Supply Chain Attack ref-106 The Hacker News (Security) Microsoft Warns of Photo ZIP Phishing Campaign Targeting Hotels with Node.js Implant ref-107 The Hacker News (Security) Russia Used Cellebrite on Jailed Activist's iPhone Months After Sales Cutoff ref-108 EU Startups Weekly funding round-up! All of the European startup funding rounds we tracked this week (June 22 – June 26) ref-109 EU Startups Lab-free testing moves closer to home as Germany’s Intu Diagnostics raises €1.1 million ref-110 EU Startups EIFO commits €200 million to Scaleup Europe Fund to back European tech scaleups ref-111 EU Startups UK BioTech Thalia Therapeutics raises €3.1 million and acquires Sanmirna Therapeutics ref-112 EU Startups What’s actually ‘hot’ in the new era of post-hype TechBio ref-113 MIT Research How data centers can better manage energy use ref-114 MIT Research LLMs help robots understand vague instructions and focus on key details ref-115 Lobsters Chatbots vs Ozone ref-116 Lobsters AI Agents Enable Adaptive Computer Worms ref-117 Google News Trump admin allows Anthropic to release Mythos AI model to some companies, government agencies ref-118 Google News Why Generative AI Isn’t Enough: The Case for Causal Reasoning in Medicine ref-119 Google News The AI agent identity problem nobody’s talking about ref-120 Google News Swiss Watchdog FINMA Builds Generative AI to Police Crypto Markets ref-121 Google News Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial ref-122 Google News A McMaster University AI model called SyntheMol-RL designed a new antibiotic candidate by searching a chemical space of 46 billion possible compounds — a scale far beyond physical lab screens, which usually test hundreds of thousands to a few million m ref-123 Google News OpenAI staggers AI model release after Trump administration request ref-124 Google News This AI Agent Survived 6,000 Hack Attempts—Here’s How ref-125 Google News San Antonio musicians found in massive AI training datasets, sparking concerns ref-126 Google News Supabase Raises $500M At $10.5B Valuation Driven By AI Coding Agents ref-127 Google News AI agents get ‘digital ID cards’ under China’s first connectivity standard ref-128 Google News FDA gives generative AI in radiology two breakthrough designation nods ref-129 Google News The AI coding apocalypse isn't happening, at least not for experienced engineers ref-130 Google News AI Coding Experiment Leaves Fintech Startup With $80K Bill ref-131 Google News Deloitte launches AI agent network across Omnia audit ref-132 Google News AI startup Lindy ditched Claude entirely for Deepseek, saving millions as cost pressure mounts on Anthropic ref-133 Google News AI Revenue Starts to Exceed a Key Infrastructure Cost But Not By Much ref-134 Google News SpaceX signs computing power deal with open-source AI startup Reflection worth up to $6.3 billion ref-135 Google News The engineer who built Swift and LLVM sold his AI startup to Qualcomm for $4B ref-136 Google News Chinese AI startup DeepSeek to double headcount in all departments | The round valued the company at over $50 bn | Inshorts ref-137 Google News A24 defends AI partnership with Google DeepMind: 'We'd rather have a seat at the table than on the sidelines' ref-138 Google News Read our white paper on a pragmatic approach to AI governance in America. ref-139 Google News OPAQUE and TII Launch Open Standard for Verifiable AI Governance ref-140 Google News Meta hires Oasis founder Dawn Song for AI safety push ref-141 Google News OpenAI Unveils Jalapeño, Its First Custom AI Chip With Broadcom ref-142 Google News Qualcomm Just Promised $15 Billion in AI Chip Sales, but There’s a Catch: The Chips Don’t Exist Yet ref-143 Google News At 3rd Circuit, Judges Press ROSS and Thomson Reuters on Fair Use, AI Training and Market Harm ref-144 Google News Apple and Microsoft hike device prices as AI chip costs soar ref-145 Google News Amazon Web Services raises GPU cloud pricing to $14.04 per hour starting July 1 ref-146 Google News Google Tells News Publishers to Share Content for AI Training or Lose Fees