中文 ↗RSS ↗

THE NEWSROOM / AI 新闻

AI News · Page 11

Source-linked briefings on models, agents, products and practical AI.

YYSI publication dates appear on the left; original announcement dates appear under each brief.

Search and categories filter this page. Use the page links below for earlier news.

REPORTED

Product updates AI industry

Marvell to acquire Lumentum in $115B all-stock deal, challenging Broadcom in AI optical interconnect

According to IT Home, Marvell announced an all-stock acquisition of optical communications firm Lumentum valued at roughly $115 billion, pending shareholder and regulatory approval. TrendForce estimates the combined company would hold nearly 7% of the wired communications semiconductor market, ranking behind Nvidia and Broadcom. Marvell brings DSP and switching chips while Lumentum supplies InP lasers and optical modules, potentially offering cloud providers an alternative to Broadcom's integrated optical-electrical portfolio.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Practical guides AI industry

China Pushes World AI Cooperation Organization Toward Full Operation With 38 Founding Members

According to IT Home, China's National Development and Reform Commission said at a press conference that it is advancing the World AI Cooperation Organization toward full operational status, with 38 countries joining as founding members. An NDRC official said China will implement the Global AI Governance Initiative, promote open-source AI access, and coordinate with partners on development strategies, governance rules and technical standards.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Panasonic Industry to Invest 92 Billion Yen in AI-Related Materials and Components

According to IT Home, Panasonic Industry announced a 92 billion yen investment in AI-related business, per a media report. The plan adds MEGTRON multilayer substrate material plants in Suzhou, China and Ayutthaya, Thailand, expands SP-Cap conductive polymer capacitors in Kyoto and Batam for AI servers, and builds an R&D facility in Yokkaichi for next-generation semiconductor packaging materials.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Practical guides AI industry

New York Times and Other Publishers Sue Brave Over Scraping Content for AI Clients

According to IT Home, The New York Times sued Brave in Manhattan federal court, alleging the privacy-focused browser maker used an undisclosed crawler to build an index of over 30 billion web pages and supplied copyrighted publisher content to AI clients including Anthropic, Perplexity, Mistral, xAI and Cohere. The Daily News, Chicago Tribune and Orlando Sentinel filed a similar suit the same day. The Times seeks an injunction and damages.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates Google

Google commits over $6M in Singapore university research grants

Google announced more than $6 million in new research grants and capacity-building funding for the National University of Singapore and Nanyang Technological University, alongside a new Singapore research hub. The grants cover AI for Alzheimer's research, sustainable compute and regional language processing, AI safety and privacy-preserving watermarking, and a regional AI for Science capacity fund. Google.org and Google DeepMind add $3 million for INSEAD's Thrive Lab.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates Google

Google expands Dreambeans AI story app to India with 13 local languages

According to TechCrunch, Google launched its experimental Dreambeans app in India, its first market outside the U.S., with support for 13 local languages including Hindi, Tamil and Bengali. The app draws on connected Google services such as Gmail, Photos and Gemini to generate personalized illustrated stories without prompts, and can animate images using Veo. Google cited strong Indian demand and high overlap across its services as reasons for the expansion.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates Anthropic

Anthropic IPO Valuation Expectations Cut to $1.5–1.6 Trillion Amid Tech Selloff

According to IT Home, The Wall Street Journal reports that institutional investors planning to join Anthropic's IPO now tell the company and its advisers to expect a valuation of $1.5–1.6 trillion, down from roughly $2 trillion discussed earlier, after a tech selloff. Anthropic has not finalized its valuation target or share sale size; its roadshow could start as soon as next week. The filing showed quarterly revenue of $17 billion through September 30.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates NVIDIA

NVIDIA open-sources Attribution for diagnosing distributed AI workload failures

NVIDIA published a developer blog announcing that NVIDIA Attribution, its workload-centric failure-attribution system, is open source. The tool correlates workload-scoped logs, metrics, telemetry and scheduler state into a structured diagnosis covering outcome, root cause, affected nodes, symptoms and an evidence trail. It ships as a REST-based Attribution Service and a Python Blueprint with connectors for Slurm, Kueue, Loki, Kubernetes pod logs and PromQL. NVIDIA says it runs in production across its AI infrastructure.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Tencent TenPay Global and Mastercard sign MoU covering inbound QR payments, remittances and agentic payments

According to IT Home, At Hong Kong FinTech Week on November 2, Tencent's cross-border payment platform TenPay Global and Mastercard signed a memorandum of understanding. The partners plan to let inbound visitors pay at WeChat Pay-enabled merchants in China using existing digital wallets or bank apps, and to extend cooperation to cross-border remittances via Mastercard Move and to agentic commerce around Mastercard's Agent Pay, Agent Connect and Virtual C-Suite products.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Trajectory raises $45M Series A and opens its continual-learning post-training platform

According to TechCrunch, Trajectory, a startup founded by three Stanford classmates, raised a $45 million Series A at a $300 million valuation led by Sequoia, with Nvidia, Bessemer and Bain Capital Ventures participating. The company says its continual-learning platform lets nontechnical specialists retrain open source models from feedback, and it is now making the post-training platform publicly available. Customers cited include Harvey, Rogo, Decagon, Mercor and Nvidia.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Qualcomm validates first-gen HBC memory architecture, AI250 sampling in H1 2027

According to IT Home, Qualcomm said it completed hardware validation of its first-generation High Bandwidth Compute (HBC) technology at its San Diego lab, confirming full read/write and near-memory compute capability. The design stacks LPDDR memory above a compute die. Qualcomm claims the AI250 accelerator, planned to sample in the first half of 2027, will deliver 133 TB/s effective memory bandwidth, 18 times the LPDDR5X-based AI200.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates NVIDIA

NVIDIA open-sources Recovery for automated distributed training recovery

NVIDIA released NVIDIA Recovery, an open-source workload-level service within its DSX OS portfolio that automates recovery across a cluster's training jobs. It detects interruptions via application logs, GPU metrics and scheduler signals, attributes failures, isolates implicated nodes and restarts workloads from checkpoints. A Slurm integration ships as a Helm-deployable Advisor Controller, and new deployments default to monitor mode. NVIDIA reports roughly 8,550 anomaly detections over a three-month production monitoring period.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates Anthropic

Anthropic Says High-Spend Customers Top 11,900 as Net Revenue Retention Hits 900%

According to IT Home, Anthropic disclosed that as of Sept. 30, 2026, customers spending at least $100,000 a year exceeded 11,900, up from about 950 a year earlier, while net revenue retention reached roughly 900%. It said over half of these customers now use Claude plus a first-party app such as Claude Code or Cowork, crediting its three-tier API-to-Cowork strategy for rising per-customer value.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates SGLang

SGLang v0.5.23 adds TPU7x support, low-latency kernels and Paged Experts

SGLang released v0.5.23, adding Google TPU7x support via a PyTorch-native plugin, opt-in NVIDIA megakernel and AMD monokernel low-latency modes, and Paged Experts for running MoE models larger than GPU memory on a single consumer GPU. The release also reports faster DeepSeek-V4.1 prefill and lower first-token latency, plus LoRA serving on quantized MoE models.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates Anthropic

Anthropic Updates Claude Code Prompt to Block Game Clones

According to The Verge, Anthropic said it added two sentences to Claude Code's system prompt on Opus 5.5 to steer the tool away from recreating specific existing video games, citing its usage policy against IP infringement. According to The Verge, original games, genre games and games built from users' own designs and assets are meant to keep working, and Anthropic said it will continue iterating on the balance.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Agents & tools DeepSeek

DeepSeek details Super-agent architecture with Router and Worker layers

According to IT Home, DeepSeek Harness published a design article on a Super-agent architecture that splits work between a Router handling human interaction and global context and isolated Worker sessions running tasks in parallel. It describes event triggers, speculative execution and an asynchronous ReAct loop, and says a prototype plugin is for validating the approach, not an official product.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Models & research Google

Google's Project Suncatcher satellite runs Gemma 4 in orbit

Google says its Project Suncatcher prototype satellite returned the first AI response generated by a data-center-class TPU in orbit. During a ground-station pass, the team uplinked a prompt to an onboard Gemma 4 open model, which produced a haiku about being in space. Google plans further tests of TPU performance under thermal stress and radiation, ahead of twin interconnected satellites targeted for 2027.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Meta

Meta and CNIB Partner to Expand AI Glasses Access for Blind and Low-Vision Canadians

Meta announced a partnership with CNIB, a Canadian non-profit, to bring the lived experience of people who are blind, Deafblind or have low vision into the development of its AI glasses. Meta is funding community listening, peer-led demonstrations and training, and feedback to its accessibility and product teams. Meta is also donating AI glasses through CNIB's Come to Work employment program.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Gemini app adds in-chat file generation, preview and editing

Google says the Gemini app can now generate, preview and edit files directly inside a chat, including documents, PDFs, presentations, spreadsheets, Markdown, TXT and code. Users can expand to full screen to type edits, highlight text to ask questions, or ask Gemini to refine files. PDF preview and editing work on web and mobile; other formats are web-only. Google says the capabilities are available to all users in all languages where the Gemini app is offered.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Product updates Google

Gemini Notebook adds free textbook access for students

Google says Gemini Notebook now lets students add hundreds of textbooks at no charge from providers in the U.S. and India, including OpenStax, Open Textbook Library and S Chand Publishing. The company claims answers are grounded in the added textbook with inline citations and links to the original source, and that content can be turned into study guides, practice quizzes and flashcards.

AI summary · Official sourceOriginal source ↗Read brief →