Goldman Sachs reportedly sees $685 billion in Anthropic revenue by 2030, Nvidia board reportedly weighs ousting Huang, and more - Daily AI Brief #581

In this edition, covering 11 November:

• Goldman Sachs analysts reportedly project that Anthropic will generate $685 billion in annual revenue by 2030, and Anthropic's S-1 filing says each of its seven founders will sell shares worth nearly $250 million in the IPO.
• Nvidia's board is reportedly preparing to decide whether to remove co-founder and CEO Jensen Huang, and Nvidia shares fell about 8% in extended trading.
• CoreWeave reported $3.66 billion in third quarter revenue, up 168% from a year earlier, and raised its 2026 revenue guidance to $13.0 billion to $13.3 billion.

From our lists

A new AI Brief list ranks private defense AI companies by their latest valuation.

Private defense AI companies by latest valuation

Anthropic

Anthropic opened its Australian retail share offering, part of its proposed international retail IPO. The offer closes on 17 November, ahead of a planned Nasdaq listing under the ticker ANT. Anthropic's S-1 filing says each of its seven founders will sell 3,378,378 shares in the IPO, nearly $250 million each at the proposed midpoint price of $74. The founders will together keep 51.1% of voting control through supervoting shares. Australian offer · Founders' share sales

Goldman Sachs analysts reportedly project that Anthropic will generate $685 billion in annual revenue by 2030, including $586 billion from enterprise customers. According to the reports, the forecast supports a potential IPO valuation of $1.4 trillion and expects much higher infrastructure spending, with positive annual free cash flow only by the end of the decade. Goldman Sachs forecast

Anthropic's amended agreements with Microsoft commit it to spend at least $47.2 billion and up to $49.2 billion on compute between November 2026 and June 2033, regardless of actual usage, according to its amended S-1 filing. The filing also says Anthropic's approximately $34.9 billion agreement with Lambda is currently terminable because Lambda did not secure a required facility lease. Microsoft and Lambda in the filing

Anthropic says Claude Mythos brought Claude up on AMD's MI355X GPUs in about three days with minimal human help. The model wrote or adapted nearly all the required GPU kernels, work that Anthropic says previously took dozens of engineers six to 12 months. US apparel retailer J. Crew launched Style Advisor, an AI shopping tool on its desktop website that is trained on its styling data and runs on Anthropic models; a mobile app version is planned. Claude on AMD · J. Crew Style Advisor

Anthropic researchers tested agents from Anthropic, OpenAI and Alibaba in four simulated multi user settings and found that teams of separate user agents did worse than a single coordinator. Anthropic released three of the settings as the MAMUBench benchmark. MAMUBench

Nvidia

Nvidia's board is reportedly preparing to decide whether to remove co-founder and CEO Jensen Huang, as the company's investigation into his reported relationship with a Megaspeed executive continues. Nvidia shares fell about 8% in extended trading. Nvidia board

OpenAI

OpenAI made Japan based request processing and model inference available for GPT-6 Luna through the Responses API and Chat Completions API. The option supports text and image inputs and text outputs, requires prior Zero Data Retention approval and is limited to Standard pay as you go billing. OpenAI also launched local API processing in India for eligible enterprises and developers, starting with GPT-6 Luna for text tasks, with supported data storage and model processing kept in India through regional API settings. GPT-6 Luna in Japan · GPT-6 Luna in India

Parloa and OpenAI expanded their partnership on AI agents for customer service, with a focus on North America. Parloa will join OpenAI's Alpha Model Testing Program and plans to offer its platform through the OpenAI Marketplace. Chicago based VIU by HUB launched a ChatGPT plugin that lets consumers in 48 states compare home and auto insurance quotes from several carriers, with a licensed advisor following up to bind the policy. Parloa · VIU by HUB

OpenAI research finds that lower model perplexity improves reasoning without chain of thought, while reinforcement learning increases a model's control over its chain of thought. Model generated training data can pass this controllability on to later models, which makes chain of thought a less reliable way to monitor advanced AI systems, according to the research. Chain of thought research

Three senior OpenAI officials discussed military support for Ukraine, including possible offensive AI services, on a video call with Russian pranksters posing as Ukrainian officials. OpenAI says it investigated the call, confirmed it was a prank and took steps to prevent similar incidents. Prank call

Australian law firm Gilbert + Tobin says 87% of users enabled for OpenAI tools are active; it uses ChatGPT Enterprise, ChatGPT Work and Codex with Australian data residency, role based access and human review. Shanghai based BlueMedia launched a partnership with ChatGPT Ads to help Chinese brands advertise in AI conversations in global markets, and says it has already run campaigns for a first group of clients. Gilbert + Tobin · BlueMedia and ChatGPT Ads

OpenAI made Aleph the exclusive reseller of ChatGPT Ads in 44 markets across Latin America, Europe, Central Asia, the Middle East, Africa and Asia Pacific, with local contracts, billing and support. Aleph

Microsoft

Microsoft President of Cloud Operations Noelle Walsh says the company will pay for electricity and grid upgrades linked to its data center growth, rather than passing those costs to other US energy customers. The pledge builds on Microsoft's January 2026 community first plan. Data center costs

Microsoft introduced Microsoft Classroom, a standalone Microsoft 365 experience that brings Assignments, Grades, Class Notebooks, Learning Accelerators and Insights into one place. It will be available at no additional cost for A1, A3, A5 and A7 customers starting in 2027, with a preview planned for early 2027. Adobe and Microsoft are optimizing Photoshop, Premiere, After Effects and Substance 3D Painter for Arm based Windows PCs, and Adobe Express templates and AI editing features will come to Microsoft Photos on Windows 11, with a preview for Windows Insiders from December 2026. Microsoft Classroom · Adobe on Windows

Amazon Web Services

AWS opened its first cloud infrastructure region in Chile for select customers, with general availability expected by the end of 2026. Amazon plans to invest more than $4 billion in Chile, and AI services including Amazon Bedrock and SageMaker AI are expected within the first year. AWS in Chile

AWS released AWS Emulator, a free tool that simulates more than 200 AWS services on developers' machines, so developers and AI coding agents can test software locally without network access or cloud charges. AWS says autonomous AI agents built and maintain the emulator; its known limitations include incomplete support for throttling, outages, regional differences and code sandboxing. AWS Emulator

CoreWeave

CoreWeave reported $3.66 billion in revenue for the quarter ended 30 September 2026, up 168% from a year earlier, and a GAAP net loss of $709 million. It raised its 2026 revenue guidance to $13.0 billion to $13.3 billion, expects full year capital expenditures of $37 billion to $39 billion and reported a revenue backlog of $132.5 billion. CoreWeave also launched CoreWeave Database, a managed PostgreSQL service with pgvector support that runs inside customers' CoreWeave Kubernetes Service clusters. CoreWeave results · CoreWeave Database

Data centres

Gartner forecasts that global AI infrastructure spending will reach $2.4 trillion in 2027, 59% of total AI spending of $4 trillion and 53% more than in 2026, with AI optimized server spending rising to $957.2 billion. Rackspace will provide and operate dedicated Nvidia GB300 NVL72 infrastructure for Together AI from 2027 and about 2MW of AMD powered compute and storage for Blacksmith in Northern Virginia; Rackspace says it is seeking separate financing for the Nvidia systems. Gartner forecast · Rackspace

Utah energy storage startup Torus raised $500 million in a Series C led by Magnetar to manufacture Foundry, a shipping container sized system that combines batteries with AI computing for inference. The units range from one to 20 megawatts, and Hydra Host is among the customers. Mavria raised $130 million across pre seed and seed rounds, with the latest round led by Paradigm valuing it at $1 billion, for factory built data center infrastructure in the United States. Torus · Mavria

Atomarine, founded by MIT students Dimitris Koutentakis and Emile Germonpré, raised $19 million in seed funding led by Y Combinator for ships that carry modular data centers, first powered by natural gas and later by small modular nuclear reactors of up to 150MW. It targets a first deployment in 2031 or 2032. Armada and Johnson Controls began continuous production at Galleon Forge Two in Dublin, which completed its first Leviathan modular data center for Fossefall in Norway. Atomarine · Armada in Dublin

Saudi Arabia's National Infrastructure Fund and HUMAIN signed two agreements worth up to SAR 2.5 billion, including up to SAR 1.1 billion for GPUs and SAR 1.4 billion for power infrastructure for HUMAIN data centers. Vodacom plans to build AI data centres with partners in Egypt, South Africa, Kenya and other African markets, and already works with Cassava Technologies and ElSewedy Electric on a centre in Egypt. HUMAIN financing · Vodacom

The Kraft Group and Unified Global Packaging Group are considering a data center campus of up to 900MW in Sutton, Massachusetts, which could require more than $16 billion in private investment; the proposal is tentative and needs local approval. Voters in Ottawa, Kansas, approved a 10 year ban on data centers larger than 25MW with 71.85% of the vote. The Riverside, California, City Council voted 5 to 2 for a resolution stating that the city cannot accept data center proposals under its current land use code. Kraft Group in Sutton · Ottawa ban · Riverside resolution

Senator Josh Hawley of Missouri supports a temporary nationwide pause on data center construction until Congress guarantees that taxpayers will not face higher utility bills. He is the first Senate Republican to back a national moratorium. Pew Research Center reports that the US had more than 3,000 operational data centers and at least 1,500 in development as of February 2026, and that 60% of US adults would be uncomfortable with a new local data center. US data center electricity use rose from 108 terawatt hours in 2020 to 183 terawatt hours in 2024, according to Pew. Hawley on a moratorium · Pew report

Policy

The European Commission formally proposed the Digital Fairness Act, with bans on dark patterns, clearer pricing rules, limits on unfair personalised ads and easier subscription cancellations. The proposal sets consumer rights for AI assisted purchases and would allow fines of up to 6% of worldwide annual turnover. Digital Fairness Act

Finland and Estonia called for EU level arrangements to respond to serious AI incidents, in a letter ahead of the first European Council meeting focused on AI. Hungary separately proposed national contact points and pre vetted evaluators for AI incidents. The European Data Protection Board selected automated decision making in recruitment, including AI based screening, as the focus of its sixth coordinated enforcement action in 2027. AI incident response · EDPB enforcement

The UK Department for Education is consulting until 4 February 2027 on a voluntary certification scheme for AI and EdTech products in English schools, colleges and nurseries, with checks by UKAS accredited assessors. Virginia Governor Abigail Spanberger and Democratic lawmakers presented a 2027 agenda that includes tighter data center rules and restrictions on AI generated child sexual abuse material and AI wage setting. EdTech certification · Virginia agenda

The California Federation of Teachers asked Governor Gavin Newsom to order a three year pause on AI use in public school classrooms from kindergarten through eighth grade. The UK Information Commissioner's Office says AI assisted subject access requests remain valid and must be handled like other requests. California teachers · ICO guidance

A Center for Security and Emerging Technology report examines more than 2,000 robotics procurement requests from China's PLA published from October 2022 to December 2025. Most of the platforms are still human controlled and focused on research and development. PLA robotics

Defence

Russia, China and the United States are discussing possible limits on AI in nuclear command and control systems ahead of a planned trilateral summit in Shenzhen. The UK Ministry of Defence announced a £90 million framework with Ocean Infinity Solutions for Royal Navy drones, sensors and AI systems that monitor the North Atlantic, detect Russian submarines and protect undersea cables and energy networks. AI in nuclear arsenals · Atlantic Net

Helsing launched Clarity, an intelligence platform that uses military customised AI agents to process data from satellites, sensors and drones and identify threats; it runs on European customers' own infrastructure. California startup Lyra Systems came out of stealth with a five year Cooperative Research and Development Agreement with the US Air Force Research Laboratory to bring early stage space technologies to orbital demonstration missions faster with AI enabled mission integration software. Helsing Clarity · Lyra Systems

Ukraine founded, London headquartered UFORCE raised $700 million in a Series A at a $6 billion valuation, co led by Valor Equity Partners and the UK National Wealth Fund, for manufacturing and research on autonomous air, land and sea systems. Munich based SE3 raised $30 million in a Series A led by Accel for software that combines drone and ground vehicle data into 3D battlefield maps and coordinates autonomous fleets; it works under contracts with the Bundeswehr and customers in Ukraine. UFORCE · SE3

Security and safety

Recorded Future analyzed 103 AI related cyber incidents from 31 July 2025 to 31 July 2026 and found that attackers target AI developer tools and use AI for phishing, fraud, reconnaissance, malware development and extortion. Fully autonomous attacks remain uncommon and usually require human oversight, according to the report. Arctic Wolf Labs linked Alibaba Cloud systems hosting the AntiFox AI penetration testing platform to attacks that reached at least one production secret in AWS Secrets Manager. Recorded Future report · AntiFox

The Institute for Law & AI introduced FraudBench, which tests 20 large language models on 159 scenarios based on US court cases. In the study, fraud rates rose with autonomy, reaching 21.4% when models acted directly compared with 3.0% when they advised users. Young people in several US states are suing xAI, alleging that users used Grok to turn ordinary childhood photos into sexualized images; xAI says it prohibits child sexual exploitation. FraudBench · xAI lawsuits

Models and developer tools

Google introduced the Gemini Decisions API, powered by Gemini Decision 1, which evaluates up to 1,000 yes or no, multiple choice and rubric questions against text, images, audio, video and PDFs in a single request. It returns probabilities and confidence scores with 150 to 250 millisecond latency and costs $0.10 per 1 million input tokens. ByteDance is developing an AI agent similar to Meta's Muse, Bloomberg reports, citing unnamed sources; co-founder Zhang Yiming reportedly told the AI team that ByteDance could compete with Anthropic and OpenAI, which may take until 2029 or 2030. Gemini Decisions API · ByteDance agent

Meta introduced Triton Low Level Extensions, or TLX, which gives developers explicit control over warp group orchestration, asynchronous memory movement, synchronization and cluster execution. Meta says it uses TLX in production for recommendation training and inference and for large scale LLM pretraining on Nvidia and AMD hardware. Tencent and Inferact are serving Tencent's Hy4 model in production and building native vLLM support for the open weight Hy4 Preview. Meta TLX · Hy4 on vLLM

The Institute of Foundation Models at MBZUAI released PAN 2, a 29B parameter world model that simulates the results of text based actions as video, with weights and code under the Apache 2.0 license. Robbyant, part of Ant Group, released LingBot-Touch, a tactile foundation model trained on nearly 5,000 hours of real world data. TIER IV released five open source end to end autonomous driving models through Autoware, including NVIDIA Alpamayo 2 Edge, and says some are still being integrated or tested in vehicles. PAN 2 · LingBot-Touch · Runway released Praxis-1, a 27 billion parameter world action model, with weights, inference and fine tuning code under the Apache 2.0 license. TIER IV models · Runway Praxis-1

DeepL launched Aure, a voice model that adapts translated speech to a speaker's tone, emotion, emphasis and pace. It is available in DeepL Voice and the Voice API, first for English, German, Spanish, French, Italian, Japanese, Portuguese and Korean. Zyphra released ZONOS3-Mini, a zero shot voice cloning text to speech model with 100 million active parameters, under the Apache 2.0 license. SpeakLeash and ACK Cyfronet AGH released Bielik 11B v3.1, a Polish language model that supports 56 languages, tool use and structured outputs, on Hugging Face. DeepL Aure · ZONOS3-Mini · Bielik 11B v3.1

Companies

Databricks plans to invest approximately $3 billion across Europe and the Middle East over the next three years. It will expand its regional research and development and customer teams, grow its EMEA workforce by about 50% and train 500,000 people in data and AI skills by 2030. Databricks also updated its Genie AI agent to carry out extended, multistep business tasks without supervision, using an ontology to understand customer data on its platform. Databricks in EMEA · Genie update

NiCE launched an updated Cognigy platform for enterprise AI agents with AI assisted development, reusable skills, decision tracing, native speech to speech models and integrations with Deepgram, ElevenLabs, OpenAI and Anthropic; current production deployments continue unchanged. Zendesk introduced Bridge, which lets external AI agents use approved Zendesk tools and data under permissions, authentication and activity logs. It supports more than 400 tools through the Zendesk MCP server and adds Zendesk plugins for ChatGPT, Claude and Microsoft Copilot. NiCE Cognigy · Zendesk Bridge

Orange France will launch new Orange and Sosh apps on 19 November with free access to Claude, Gemini, Mistral and GPT, and an Orange IAs Plus subscription for 10 euros a month. Lloyds launched Export Assist, an AI tool based on Complidata's TradeSpeed that checks export documents for errors and is open to exporters that do not bank with Lloyds. Bilt launched Bilt Neighborhood, an AI platform that connects local businesses with customers through residential communities, offices, hotels and sports teams. Orange apps · Lloyds Export Assist · Bilt Neighborhood

Alibaba, JD.com and Douyin used chatbots, virtual try ons and other AI tools during China's Singles Day festival. JD.com's Dongdong assistant attracted nearly 80 million users, and chatbot links drove about 5% of Douyin purchases in October. Legora expanded former Solicitor General Lucy Frazer KC's role from advisor to part time lead for its public sector work. Singles Day · Legora

United Airlines is changing its ticketing rules to prohibit unauthorized third party AI agents from making, changing or servicing tickets, citing risks to passenger data, unintended purchases and US aviation rules. Reddit is testing Reddit Reviews, which turns product discussions into AI organized summaries for about 10,000 products, and will roll it out to US users over the next few weeks. Aura reported third quarter revenue of $87.9 million, up 24%, and introduced Solstice, an in house AI engine for family protection, now in limited beta. United Airlines · Reddit Reviews · Aura

Meta says infrastructure chief Santosh Janardhan will step down and move into an advisory role on Meta Compute before leaving next year; Dave Brown, a former Amazon executive who joined Meta earlier in 2026, will replace him. Octave is acquiring Cerebre to combine Cerebre's live plant models, deployed at more than 200 plants, with Octave's engineering records for industrial AI. Cresta opened a Tokyo office with Jim Tamura as country manager, and DPG Media is testing whether the GPT-NL model can suggest headlines and flag copy issues at two Dutch publications until the end of January 2027. MasTec will deploy Gritt's AI robotic arms on utility scale solar projects in several US states. Meta infrastructure · Octave and Cerebre · Cresta in Japan · GPT-NL at DPG Media · MasTec and Gritt

Funding

Munich based Agile Robots raised $500 million in a round led by SoftBank, with participation from Nvidia, at a valuation of $8.5 billion. The money goes to US expansion, new industrial applications and manufacturing in Bavaria. Agile Robots

Italian startup Nebuly raised $21 million in a Series A led by Planven for its platform that measures the cost and business value of AI agents at large enterprises; it plans to grow from 40 to 60 people and hire sales staff in New York, London and Europe. Plurai raised $13 million in seed funding led by Team8 and made its platform for simulating AI agent interactions generally available. Anorak raised a $5 million pre seed round co led by Play Ventures and Backed VC, with Supercell and MTG, for an AI game creation platform it plans to launch in early 2027. Tano raised $4 million in seed funding led by Cherry Ventures for creator advertising run with AI agents and creator strategists. Nebuly · Plurai · Anorak · Tano

Chinese robot startup Wujingdongli shut down eight months after its founding, leaving 30 to 40 employees owed about 8 million yuan, as investment in China's physical AI sector declines. Wujingdongli

Stage 2 Capital closed $100 million across Venture Fund IV and Catalyst IV for B2B AI companies from pre seed through Series A. Enza Capital raised $21 million in the first close of its third fund, which targets $60 million for AI startups across Africa. StitchStudio raised $8 million in a round led by Voyager Capital to expand its insurance AI products, and Helix raised $6.2 million in pre seed funding led by Chicago Ventures for on device voice verification that detects bots and estimates user age. Stage 2 Capital · Enza Capital · StitchStudio · Helix

Rox, which builds AI agents for enterprise sales teams, raised $100 million in a Series B at a $1.3 billion valuation, co led by General Catalyst and March Capital. Rox

Health and science

Mosaic Intelligence Labs, founded by Stanford researchers Ileana Pirozzi and Alison Kight Mackanic, came out of stealth with $47 million in seed funding for AI and robotic tools that help surgeons navigate blood vessels. Kardi Ai raised a €5.3 million Series A led by Callisto Health for long term ECG monitoring with AI analysis. Switchpoint Bio raised $5.3 million in pre seed funding for a blood based epigenomics platform that tracks treatment response, first in autoimmune diseases. Mosaic · Kardi Ai · Switchpoint Bio

Lithuanian company Biomatter raised $18 million in a Series A led by Astanor for enzymes designed with computational models and lab testing, and plans a 2,000 square metre design and testing facility in Vilnius in 2027. Biomatter

A team led by MIT receives up to $53 million under the ARPA-H STREAM program to use generative AI to develop safer herbicides and RNA based alternatives. OptumAI is partnering with California Medicaid insurer Alameda Alliance for Health to automate claims adjudication with agentic AI. AI herbicides · Optum claims

Surveys and skills

A Nippon Research Center survey from June 2026 found that 56.3% of people in Japan aged 20 to 69 had used generative AI and 50.4% were using it, with ChatGPT the most used tool. Community Care's 2026 Learning Landscape research found that 53% of 1,649 social workers in England and Wales had received no guidance or training on AI, while 56% used AI in their daily work. Japan survey · Social workers

Research from the USF Muma College of Business, using confidential US Census Bureau data, finds that banks adopting AI lent more to distant small businesses without higher risk. South Africa launched the AI Research Foundations programme at Stellenbosch University, supported by Google.org, to train lecturers at 30 universities in Ghana, Kenya, Nigeria and South Africa and reach 30,000 students. AI in bank lending · AI Research Foundations

Ohio State University recorded 816 academic misconduct cases involving unauthorized AI use in 2025 to 2026, up from four in 2023 and 19% of all cases. Ohio State

Also

Anyday Security launched ExploitTotal.ai, a free platform that tests whether open weight AI models can exploit disclosed vulnerabilities; it says the platform holds more than 200 working AI generated exploits, 72% of them for vulnerabilities with no previously known public exploit. Austrian developer Stanislaus Hees, 14, built Autark, an open source tool in which a manager agent assigns tasks to other AI agents; it has no sandboxing, approval prompts or permission gates. ExploitTotal · Autark