Google's most powerful model, Gemini 4 Carbon, officially launched; Anthropic's Nasdaq IPO roadshow is underway and Claude is expanding into a workspace; OpenAI rolled out a business hub inside ChatGPT. We round up the announcements the three big AI labs poured out in a single day, the shock from Nvidia, and Korean startup FuriosaAI's inference chip entering mass production.
🤖 Google Officially Launches Its Most Powerful Model, 'Gemini 4 Carbon'
Google officially launched Gemini 4 Carbon, billed as its most powerful model, for developer APIs, enterprise customers, and paid consumers. Aimed at demanding agent work like long coding, research, and security tasks, the model claimed top scores on multiple evaluations in Google's own benchmarks. API pricing is set at $2 per million input tokens and $10 per million output tokens through November 30, doubling after that. A "/swarm" feature for orchestrating multiple coding agents was also previewed.
📈 Anthropic IPO Roadshow in Full Swing… Targeting up to $140 Billion Valuation
Claude developer Anthropic has begun its roadshow for a Nasdaq listing. It aims to offer 825 million shares at $70–$78 each, targeting a valuation of up to about $140 billion — below the $200 billion figure initially discussed. Seven big tech companies, including Nvidia, Google, and Microsoft, have also reportedly committed $26 billion in private investment separate from the offering. However, the SEC registration has not yet taken effect, so shares cannot be sold yet.
🗂️ Claude Expands into a Workspace with Spreadsheets, Whiteboards, and Meetings
Anthropic unveiled new features expanding Claude from a simple chatbot into a collaborative workspace. "Claude Sheets," with co-editing, formulas, and Excel export; a "Whiteboard" for organizing ideas together; and "Meetings" for handling meeting content launched in beta, while the existing Dashboard was renamed "Claude Data." Interactive connectors that embed partner apps like Adobe and Atlassian directly inside Claude were also added.
OpenAI began the beta of "Businesses in ChatGPT," letting companies open profiles directly inside ChatGPT. Companies can register profiles in a business hub, manage product information, and view aggregated insights into how users interact with them. U.S. e-commerce and online service companies can also handle customer service through a separate "business agent" conversation, making ChatGPT's expansion as a gateway for search and shopping increasingly clear.
🔐 DeepMind Mathematically Verifies Signal's Encryption with AI… Generated 6 Million Lines of Proof
Google DeepMind used its AI-agent-based formal verification framework "AlphaProof Code" to mathematically prove that Signal messenger's encryption protocol's actual production code satisfies its core security properties. Gemini-based agents generated more than 6 million lines of Lean proofs over four weeks to complete the verification, and also confirmed that an older version with a known flaw is rejected. The caveat: verification covers the specified properties and code scope, not every possible attack.
📉 Nvidia Loses $430 Billion in Market Value in Two Days… Korea's AI Supply Chain on Edge
As allegations linking CEO Jensen Huang to a Chinese intelligence agency and controversy over diverted AI chip supplies spread, Nvidia shares fell about 9.5% over two days, erasing roughly $430 billion in market value. The board opened an independent investigation into the reports about Huang. Korea is on edge too: Samsung Electronics and SK Hynix participate in the HBM supply chain for Nvidia's AI chips, while Naver, Hyundai Motor, and LG are tied in through GPU adoption and robot and autonomous-driving partnerships. However, no disruption to domestic supply or partnerships has been confirmed yet.
🇰🇷 FuriosaAI Starts Mass Production of Inference Chip 'RNGD'… "Twice the Power Efficiency"
FuriosaAI, led by CEO Baek Jun, a former Samsung Electronics executive, began mass production of its AI inference chip "RNGD (Renegade)" this month, challenging Nvidia's dominance. RNGD is rated at more than twice the power efficiency of Nvidia's top-tier chip and passed technical validation from LG AI Research and OpenAI. FuriosaAI was valued at about $700 million in its recent fundraising and reportedly turned down an acquisition offer from Meta.