🪖 Plus: Anthropic stands firm against the Pentagon, Google's Gemini 3.1 Pro crushes benchmarks, and Chinese AI labs caught red-handed distilling Claude
Sponsored by
🎉 Celebrating $100M ARR! 🎉 To mark the milestone, they are offering a flat 95% discount on their Standard monthly plan, bringing it down to just $1 per month to build full stack apps with AI. No catch. No fine print. This offer will not last, so do not miss it.
Try Emergent for $1/monthWelcome back to the world's #1 AI newsletter.
Today, we've got Meta inking a jaw-dropping $100B+ chip deal with AMD, Google's Gemini 3.1 Pro doubling its reasoning prowess overnight, Anthropic refusing to cave to the Pentagon on AI guardrails, and Chinese AI labs getting busted for secretly mining Claude.
Buckle up — it's been one of the wildest weeks in AI yet.
On this day in 2025…
Samsung held its Galaxy Unpacked event, revealing the Galaxy S25 series with advanced on-device AI features — the first smartphones to deeply integrate Google's Gemini Nano. The move signaled a new era where AI wouldn't just live in the cloud, but right in your pocket.
💰 Meta Strikes $100B+ AI Chip Deal With AMD
Meta just signed a massive multiyear agreement to deploy up to 6 gigawatts of AMD GPUs for its AI data centers — valued at over $100 billion. The deal even includes AMD issuing Meta a warrant to acquire ~10% of its stock. This comes just days after Meta committed billions to millions of Nvidia chips. The message: AI infrastructure dominance requires multiple suppliers.
🧠 Google Launches Gemini 3.1 Pro, Doubles Reasoning Scores
Google just retook the AI crown. Gemini 3.1 Pro scored 77.1% on ARC-AGI-2, more than doubling the previous version's 31.1%. It also set records on Humanity's Last Exam (44.4%) and topped the APEX-Agents leaderboard. The model targets science, research, and engineering workflows that need deep planning and synthesis.
🛡️ Anthropic Won't Budge as Pentagon Escalates AI Dispute
The Pentagon has given Anthropic until Friday to loosen its AI guardrails — or face potential penalties under the Defense Production Act. Anthropic has drawn two red lines: no autonomous weapons without human oversight, and no mass surveillance of Americans. With Claude being the only frontier AI model on classified Pentagon networks, this is a high-stakes game of chicken neither side can easily walk away from.
🕵️ Anthropic Accuses Chinese AI Labs of Mining Claude
Anthropic dropped a bombshell, accusing DeepSeek, MiniMax, and Moonshot AI of creating over 24,000 fake accounts to extract Claude's capabilities through "industrial-scale distillation." The 16 million+ exchanges targeted Claude's agentic reasoning, tool use, and coding — its most differentiated features. The timing is no coincidence: it comes as the U.S. debates tightening AI chip export controls to China.
💸 OpenAI Nears $100B Deal at $850B+ Valuation
OpenAI is reportedly finalizing a monster funding round with Nvidia investing up to $30B, Amazon putting in $50B, and SoftBank committing $30B. If closed, this would make it one of the largest private funding rounds in history and put OpenAI's valuation north of $850 billion.
⚡ OpenClaw is the open-source personal AI agent that went from zero to 196K GitHub stars in weeks. It automates tasks across email, messaging, calendars, and web browsers — all running locally on your own device. OpenClaw's creator Peter Steinberger has joined OpenAI, and the project will live on as an independent foundation.
📊 TimesFM 2.5 from Google Research is a pretrained time series foundation model that just got integrated into Google Sheets via Connected Sheets. No SQL, no Python, no model training required for business users to run forecasts directly in their spreadsheets. Trained on 100B+ real-world time points, it ranks #1 among open source models on GIFT-Eval and was the first foundation model to beat AutoTheta on second-level frequency. Apache 2.0 licensed.
🔍 OpenPlanter is a recursive LLM investigation agent with a terminal UI that dropped on Feb 23 and already hit 900+ stars. It ingests corporate registries, campaign finance records, lobbying disclosures, and government contracts, then surfaces non-obvious connections through evidence-backed analysis. Default: 4 levels of recursive depth with parallel subagent execution. The tagline: "so you can keep tabs on your government since they're almost certainly keeping tabs on you." Apache 2.0 licensed.
Coming in Hot
AI portraits and videos from your own photos
AI resume tailoring to land more interviews
Create viral-ready videos & stunning images in seconds. Start free
AI upscaler for natural detail recovery
Turn blog posts into AI videos in minutes
AI customer service agent that resolves tickets fast
AI video generator from text and images
Access ChatGPT, Claude, and Gemini in one place
Build full-stack web apps by chatting with AI
Access 50+ AI models in one workspace
Submit your AI tool to the #1 AI tools marketplace.
Why It's Worth Your Time
Pricing
Why it matters: With students spending 40+ hours per week on coursework, Turbo AI targets a real pain point by automating the most tedious part of studying. It's not trying to write your essays — it's making sure you actually learn the material.
Why It's Worth Your Time
Pricing
Why it matters: With AI coding assistants becoming table stakes, Qoder stands out by shipping a full IDE rather than just a plugin. The agentic approach — where AI plans, codes, and tests autonomously — signals where developer tools are heading in 2026.
Why It's Worth Your Time
Pricing
Why it matters: Academic writing is the #1 use case for AI among students, but most tools produce generic, detectable output. Textero's focus on essay structure, research assistance, and review makes it more of a writing coach than a ghostwriter.
📉 Nvidia Earnings Drop Wednesday — Wall Street Is Watching
Analysts expect Nvidia to report a 68% revenue jump to $66B for Q4, but investor skepticism about peak AI spending looms. With Meta's AMD deal and custom chip efforts growing, all eyes are on Jensen Huang's forecast for Vera Rubin systems.
🤖 AI Agent Deletes Researcher's Inbox in "Speed Run"
A Meta security researcher let her OpenClaw agent loose on her real inbox. It started deleting everything and ignored her stop commands. She had to sprint to her Mac Mini "like defusing a bomb." A cautionary tale about trusting AI agents with sensitive data.
🔬 AI-Designed Proteins Could Spot Cancer Via Urine Test
MIT and Microsoft researchers used AI to design molecular sensors that detect early signs of cancer through a simple urine test. The AI model creates short proteins targeted by cancer-specific enzymes — a breakthrough that could make early detection dramatically more accessible.
⚖️ OpenAI Defeats xAI's Trade Secrets Lawsuit
A federal judge granted OpenAI's motion to dismiss xAI's trade secrets claims, ruling that xAI "does not point to any misconduct by OpenAI." The case centered on eight employees who left xAI for OpenAI — but xAI can refile with modified claims.
🔧 Google Adds AI Workflows to Opal
Google's vibe-coding app Opal now lets users build automated workflows through text prompts — no code required. The new agent uses Gemini 3 Flash to plan and execute multi-step tasks, making it easier for non-technical users to create mini-apps.
