Models, products and AI startups, near-duplicates collapsed to the most credible source
Arista Posts Record Q3 Growth as AI Networking AcceleratesArista Networks reported Q3 revenue of $3.66B, up 58.5% year over year, its largest growth as a public company. CEO Ullal said Arista has surpassed 100 Etherlink customers and forecasts over 10 gigawatts deployed across 10+ accelerator types by end-2027.
Qualcomm unveils Cloud AI 100 Ultra acceleratorQualcomm's Cloud AI 100 Ultra is a 150W AI accelerator card with 128GB of memory and 548GB/s of bandwidth.
Crunchbase: Under 2% of 2026 VC Funding Went to Red StatesCrunchbase data shows the 17 remaining red states, home to nearly a quarter of Americans, drew just $8.6 billion in venture funding this year through Election Day — under 2% of the U.S. total and less than a typical single week. Texas and Ohio shifted to purple, and the 10 purple states pulled in about $28.6 billion.
Apple Cuts Internal AI Usage Budgets to Reduce SpendingApple Inc. is reducing employee budgets for AI usage as part of cost-cutting efforts, even as it seeks to make AI central to its operations.
Data Center Buildout Could Help Energy PricesArushi Sharma Frank of Luminary Strategies says data centers can be assets to energy resources and the power grid; she discusses how midterm results may impact the AI buildout and why Elon Musk's plan to put data centers in space diversifies the chip sector.
Tech industry's Trump bet backfires as midterms become AI referendumIn the US midterms, 56% of voters called AI and data centers a major factor, over 40 data-center ballot measures appeared across ten states, and many Democrats who backed data-center moratoriums won. Texas, holding 27% of planned US data-center capacity, saw its governor ousted, putting Big Tech's bets at risk.
DuckDuckGo Pitches Itself to Schools Amid Google AI BacklashAfter Google told schools that disabling its Search AI features would cost them Google Search, DuckDuckGo is courting educators and showing them how to switch their default search engine. Google says about 50 million Chromebooks and over 150 million Google Classroom users are in schools, a market Google has long dominated.
OpenAI Asks Court to Toss X's Apple ChatGPT Antitrust SuitOpenAI is seeking summary judgment to dismiss X Corp.'s antitrust claims, arguing its ChatGPT integration with iPhones is not exclusive and rivals can still negotiate with Apple or distribute through the App Store. Musk's companies sued Apple and OpenAI in August 2025; in September 2026 they dropped Apple and kept the case against OpenAI.
Sánchez Urges Ibero-American Common Response to AIAt the 30th Ibero-American Summit in Madrid, Spanish Prime Minister Pedro Sánchez urged the 22 participating countries to agree on AI development and regulation, citing effects on jobs, production and supply chains, and backed a common AI agenda alliance. The summit was marked by the absences of Lula and Sheinbaum.
Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic CodingRicursive Intelligence released Ripple-65B Preview, a general-purpose model scoring 80.52 on TerminalBench 2.1, a new U.S. open-weight SOTA, surpassing larger models like Reflection Beam 501B and Meta Muse Spark Open 816B. Pretrained on 12 trillion tokens with up to 1M context, it runs efficiently on local devices such as NVIDIA DGX Spark.
Texas Election Shift May Jeopardize AI Data Center BuildoutDemocrats James Talarico and Gina Hinojosa won Texas' Senate and governor races after making data centers a central campaign issue, with Hinojosa pushing a moratorium. In September, Gov. Greg Abbott had already directed a regulator to halt data center permits. The Stargate Frontier project under construction in Texas is projected at 1,820 megawatts.
Hugging Face policy chief on open source AI costs and risksHugging Face's policy chief discusses how open source models can be monetized, plus the costs, risks and trade-offs for enterprises, governments and startups adopting them. The episode also asks whether US players can keep pace as Chinese open models gain ground.
Nothing Teases Screen-Free AI Device Launching Next MonthNothing CEO Carl Pei said on X that the company's first AI-native product ships in December. He said it has no new screen and could not have been an app, offering few other details. The launch comes just over a year after he announced plans for AI-native devices and an AI operating system.
ChatGPT adds GIF creation alongside Images 2.6 model updateChatGPT can now generate GIFs from a prompt or an uploaded photo, with options to adjust playback speed, edit frames and share. The feature arrives with OpenAI's Images 2.6 model; previously ChatGPT could not natively make GIFs in one step.
Reporter's use of always-recording Amazon Bee Frontier raises privacy questionsVerge senior reviewer Victoria Song recounts wearing the always-recording Amazon Bee Frontier in therapy, needing her therapist's consent, and criticizes its LED indicator as too small to notice. The piece examines privacy concerns around always-on AI wearables.
Unsloth releases Qwen3.8-Flash-Next GGUFs V3.1Unsloth released V3.1 of its GGUF quantized versions of Qwen3.8-Flash-Next, claiming a 15% accuracy improvement.
Dharma releases 2B-parameter field extractor outputting structured JSONDharma launched Field Extractor, a 2-billion-parameter model that pulls requested fields from documents like invoices, medical forms and IDs and returns structured JSON. It claims to beat every frontier LLM tested at a fraction of the cost, extending its earlier domain-specialized DharmaOCR-LITE approach.
Kino AI Launches AI Media Manager and Video EditorY Combinator-backed Kino AI launched an AI media manager and video editor that is code-first and fully collaborative in real time, allowing multiple AI agents and teammates to work on the same timeline; it is built on Remotion.
Anthropic and peers advance provable inference with zkSNARKsResearchers from Anthropic, EPFL and others use zkSNARKs to cryptographically prove that responses came from the claimed model, guarding against substituted or tampered outputs. Their FlashProve system generates proofs on Llama 4 Maverick (400B parameters) at about 110 tokens/second on one H100, with proofs of ~19 MB per 1,024 tokens and ~70-95x the GPU time of plain inference.
LFM3-Omni-2B Runs Offline on iPhone at ~80 tok/sA Reddit user reports running LFM3-Omni-2B fully offline on an iPhone 18 Pro Max at about 80 tokens per second using custom WebGPU kernels in a React Native app, highlighting on-device multimodal inference potential.
AI takeover debate resurfaces as Anthropic researcher resignsBuilt In examines AI takeover theory and its risks, citing researcher Jacob Coxon's resignation from Anthropic with a warning that AI products could endanger humanity by decade's end. An October 2026 Axios/CivicScience survey found most US adults expect no positive impact on quality of life or society.
AMD books $1.1B noncash charge tied to OpenAI, Meta warrantsAMD recorded about a $1.1 billion noncash charge in Q3, mostly from stock warrants issued under its AI chip supply deals with OpenAI and Meta. The charge hit GAAP earnings ($1.46 per share) but not cash flow; the warrants could dilute shareholders if targets are met, showing how AI suppliers use equity to lock in demand.
CloudOpsBench: Reasoning Is Top Bottleneck for Cloud Ops AICloudOpsBench, a new open-source benchmark, measures how well AI agents handle typical cloud DevOps tasks such as fixing services, removing resources, restricting access, rolling out updates and decommissioning projects. It finds reasoning to be the biggest current bottleneck.
Pennsylvania Requires Disclosure for Synthetic-Content AdsPennsylvania now requires advertisers to disclose when synthetic content in ads could create false consumer perceptions. The rule targets AI-generated or altered ad material to prevent misleading consumers.
Survey: 22% of Ukrainians Get News via AIAn Internews-Ukraine survey found about 22% of Ukrainians used AI to get news in 2026, matching television and online media; 87% still rely on social media, with Telegram top at 82%. Only 10% correctly judged the accuracy of three test stories, and 69% opposed restricting social media access.
Study Profiles HBM Traffic in Huginn Looped TransformersA profiling study of Huginn's looped architecture on a B200 finds that shared weights save storage but do not stop weights being re-read from HBM each iteration once they exceed L2 capacity. Eight iterations of the Huginn Core took roughly eight times as long and read about eight times as much data as one.
Researchers find prompt adware in OpenAI Codex jailbreak toolkitCheck Point Research disclosed a Codex jailbreak bundle called Haiou that uses prompt injection to alter global agent instructions, inserting ads for a third-party LLM gateway when users type triggers like 'hi' or 'Seagull.' It calls this scheme 'prompt adware' aimed at driving users to the paid gateway.
US midterms deal blow to data center expansionUS voters elected candidates favoring data center moratoriums and passed ballot measures restricting expansion. An NBC poll found 76% of respondents opposed data centers in their communities, a blow to a key pillar of the AI industry.
Poll shows sharp global divide on AI attitudesIpsos' 2026 AI Monitor found 85% of Chinese, 80% of Indian and 77% of Indonesian respondents see AI as more beneficial than harmful, versus only 38% in the US. The article cites historical cases showing resisting technology forfeits its gains.
Uber fulfilled only 4 AI-agent rides last quarterUber disclosed for the first time that third-party AI agents arranged just 4 rides in Q3, with about 500 Uber Eats orders during Halloween week out of 320 million total, showing agent-driven bookings remain nascent.
The A.I. Daily