NVIDIA Bets $1 Trillion, Anthropic Builds a Think Tank | Weekly Digest
Rogue AI agent at Meta, Microsoft vs OpenAI contract war, and tools of the week
Hey! Welcome to the latest Creators’ AI Edition.
The biggest AI conference of the year just wrapped in San Jose. Jensen Huang put $1 trillion on the table and called it conservative. Anthropic launched a think tank while suing the Pentagon. OpenAI quietly made its best reasoning model free. And a rogue AI agent at Meta triggered a company-wide security alert. Seven days. No boring moments. Let’s get into it.
Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
Keep your mailbox updated with practical knowledge & key news from the AI industry!
Featured Materials 🎟️
NVIDIA GTC 2026: Jensen Huang Says $1 Trillion — And That’s Just the Start 🟢
NVIDIA GTC 2026 ran March 16–19 in San Jose — 10,000+ people packed the SAP Center, and Jensen Huang delivered two hours of announcements that set the direction for the next phase of AI infrastructure.
The headline number:
Jensen Huang said he expects purchase orders between Blackwell and Vera Rubin to reach $1 trillion through 2027 — doubling earlier estimates of $500 billion. He attributed the surge to agentic AI driving a fundamental shift in computing requirements: more inference, more orchestration, more continuous compute.
What’s new:
Vera Rubin NVLink 72 — the new Vera Rubin CPU is designed for super high single-thread AI performance and represents 40 million times more compute in 10 years.
Groq 3 LPU — Huang unveiled the Nvidia Groq 3 Language Processing Unit — the company’s first chip from the startup it mostly acquired through a $20 billion asset purchase in December. The Groq 3 LPX rack holds 256 LPUs, designed to sit beside the Vera Rubin rack-scale system.
NemoClaw — NVIDIA’s enterprise-secure reference stack for OpenClaw, with 17 launch partners. Built to resolve the key problem with OpenClaw in production: no sandboxing, no PII filtering, no compliance controls.
NVIDIA Agent Toolkit — includes NVIDIA OpenShell open source runtime for building self-evolving agents. Leading partners including Adobe, Atlassian, Salesforce, SAP, Cisco, CrowdStrike, and ServiceNow are advancing enterprise AI agents with the toolkit.
Autonomous vehicles — Huang announced that Nissan, BYD, Geely, Isuzu and Hyundai are building level 4 autonomous vehicles on Nvidia’s Drive Hyperion program, declaring “the ChatGPT moment for self-driving cars has arrived.”
Disney’s Olaf robot — trained in NVIDIA simulation, walked on stage and held a conversation with Huang. The keynote closed with an AI-generated song performed by robots around a campfire.
For creators: The Groq LPU is the inference play. Training is settled — NVIDIA dominates. The fight now is for cheap, fast, scalable inference. The $1T number is the bet that agentic AI workflows will consume compute orders of magnitude beyond what chat-based AI did.
Source: CNBC | Tom’s Hardware | NVIDIA Blog | TechRadar
Anthropic Launches a Think Tank While Fighting the Pentagon 🧠
Anthropic launched the Anthropic Institute — an internal think tank that merges three of its current research teams into a unified operation focused on AI’s societal implications. The timing is pointed: the company is simultaneously suing the Trump administration over its Pentagon blacklisting.
What the Institute does:
The institute brings together three existing teams: the Frontier Red Team, which stress-tests AI systems to understand the limits of their capabilities; Societal Impacts, which studies how AI is being used in the real world; and Economic Research, which tracks its impact on jobs and the broader economy.
The institute will examine questions such as how powerful AI systems could affect jobs and economic activity, what risks they could create or amplify, how companies should determine the values reflected in AI systems, and how increasingly capable systems should be governed if recursive self-improvement begins.
The people:
The Institute starts with roughly 30 experts and is slated to double its staff each year. Jack Clark, previously head of public policy, will now serve as the Institute’s Head of Public Benefit. Founding hires include Matt Botvinick (ex-Google DeepMind), economist Anton Korinek (University of Virginia), and Zoë Hitzig (ex-OpenAI).
The DC play:
In parallel, Anthropic is expanding its Public Policy organization. The company plans to open an office in Washington, D.C. this spring. Sarah Heck (ex-Stripe) leads the team.
The bigger picture: Leaders connected to OpenAI, Google, and Microsoft have backed Anthropic in legal filings against the Pentagon designation. The Institute is Anthropic’s bet that public research credibility — not just product growth — is how you win the long game with governments.
Source: Anthropic Blog | eWeek | SiliconAngle | TechBuzz
GPT-5.4 Mini Is Now Free for Everyone 💸
On March 17, OpenAI rolled out GPT-5.4 mini in ChatGPT. GPT-5.4 mini is available to Free and Go users via the “Thinking” feature in the + menu. This is the first time OpenAI’s reasoning-capable model tier is available at no cost.
What changed:
Free and Go users: access GPT-5.4 mini via Thinking toggle
Plus and paid users: GPT-5.4 mini as rate-limit fallback during peak usage
Enterprise: option to default Auto routing to mini
GPT-5 Thinking mini will be retired in 30 days as a selectable option
The model picker was simplified to three tiers: Instant, Thinking, and Pro — with an Auto-switch feature and advanced controls for thinking effort.
Pricing:
GPT-5.4 nano is only available in the API and costs $0.20 per 1M input tokens and $1.25 per 1M output tokens — making it one of the cheapest reasoning-capable APIs on the market.
This is OpenAI’s clearest move yet to compete with Claude’s free tier momentum. Reasoning for free used to mean a degraded experience. GPT-5.4 mini on the Thinking tier is not a degraded experience.
Source: Releasebot | AI Pulse
News of the week 🌍
Claude 1M Context Now Free for Everyone 🧠 — The 1M token context window is now generally available for Opus 4.6 and Sonnet 4.6 at standard pricing on the Claude Platform — no multiplier, no extra fees. 1M context is also included in Claude Code for Max, Team, and Enterprise users with Opus 4.6. Fewer compactions, more conversation kept intact.
Rogue AI Agent Triggers Security Crisis at Meta 🚨 — A rogue AI agent at Meta triggered a severe security incident, inadvertently exposing massive amounts of sensitive company and user data to unauthorized personnel for two hours. Meta classified the event as a “Sev 1” — its second-highest severity level. The incident originated from a routine internal technical query that an autonomous agent mishandled. Timing note: it happened the same week Meta formally brought the Moltbook team into Meta Superintelligence Labs.
Microsoft Fears OpenAI’s AWS Deal Violates Azure Contract 🤝 — Microsoft has flagged that OpenAI’s newly announced partnership with Amazon Web Services may violate the Azure exclusivity terms in their existing agreement. Microsoft fears OpenAI’s AWS deal may violate Azure exclusivity contract. If true, this would be the biggest crack in the Microsoft-OpenAI partnership since the OpenAI board crisis of 2023.
Micron Earnings Beat — Stock Drops Anyway 📉 — On March 18, Micron disclosed earnings and guidance that significantly exceeded expectations, but its share price dropped as much as 6% in after-hours trading. Micron also revealed that capital expenditure for the current fiscal year will exceed $25 billion. Investors are asking whether the AI infrastructure buildout is ahead of actual demand.
Fed Holds Rates at 3.5–3.75% 🏦 — On March 18, the Federal Reserve held rates steady, maintaining the target range between 3.5% and 3.75%. AI infrastructure spending remains the biggest wildcard in the macro picture, with hyperscalers committing hundreds of billions in capex regardless of rate environment.
Bernie Sanders Interviewed Claude About AI Privacy — And It Went Viral 🤖 — On March 19, Senator Bernie Sanders sat down with a phone on a C-stand and asked Anthropic's Claude one question: what would surprise Americans about how their personal data is being collected? Claude said companies collect everything — browsing history, location, purchases, and even how long you hover over something before deciding not to click. When Sanders pressed on motive, Claude replied: "Money, Senator, it's fundamentally about profit." The 9-minute video hit 5.1M views on X within hours. The internet reacted with "Old Man Yells at Claude" memes — but the irony wasn't lost: a senator used a chatbot built by one of the most powerful AI companies to warn the public about the dangers of powerful AI companies.
OpenJarvis: Local-First AI Agents from Stanford 🏠 — Stanford researchers introduced OpenJarvis, an open-source framework for personal AI agents designed to run on-device by default, calling the cloud only when necessary. The “local-first” approach emphasizes privacy, latency, and efficiency. A direct counterpoint to every cloud-first agent framework announced this week.
Google Maps Gets Gemini: Ask Maps + Immersive Navigation 🗺️ — Google unveiled Maps updates powered by Gemini, including Ask Maps — conversational queries for complex, real-world questions — and Immersive Navigation, a more visual driving experience with 3D context.
Useful tools ⚒️
⭐ CreateOS — Build and deploy apps from any AI coding tool, in one place. Works with Claude Code, Cursor, Lovable, and v0. The missing “last mile” layer between vibe-coded output and actual deployment.
SuperX — All-in-one growth OS for serious X creators. Analytics, scheduling, engagement tracking, and AI-assisted content — all in one dashboard. For anyone building an audience on X in 2026.
Atoms — Turn your ideas into products that sell. Validates, builds, and launches product concepts with AI — from idea to landing page.
Hugo — The AI agent that doesn’t charge $1 per support ticket. Flat-rate customer support automation that handles complex queries, not just FAQs.
Umbrel Pro — 16TB home cloud server. Run OpenClaw, store files, and self-host your AI stack without any subscription. Privacy-first infrastructure for builders.
Know friends, co-founders or vibe-coders who need this content. Share!
Weekly Guides 📕
NVIDIA GTC 2026: Every Major Announcement in One Place — NVIDIA’s own rolling coverage from Jensen’s keynote through March 20, including all partner announcements and robotics demos
NemoClaw vs OpenClaw: What Changes for Enterprise — Why 70+ dependencies and zero sandboxing made OpenClaw a non-starter for Fortune 500, and how NemoClaw fixes it
The Agentic Web Explained: Moltbook, OpenClaw, and What Comes Next — The best breakdown of why Meta bought Moltbook and OpenAI bought Promptfoo on the exact same day
The Anthropic Institute: Full Charter and Founding Hires — Anthropic’s own post on what the Institute will research, who leads it, and why they’re opening a DC policy office
GPT-5.4 Mini: What Free Users Actually Get — Full changelog for the model picker update, Auto-switch behavior, and what gets retired in 30 days
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
NVIDIA GTC 2026 Full Keynote Replay — Jensen Huang’s two-hour address: Vera Rubin, Groq LPU, NemoClaw, Disney robots, and the $1T order projection
The Rogue AI Agent Incident at Meta: Full Report — Detailed breakdown of the Sev 1 security incident, how an autonomous agent exposed sensitive data for two hours, and what it means for enterprise agentic AI
OpenJarvis on GitHub — Stanford’s Local-First Agent Framework — The privacy-first alternative to every cloud agent launched this week
McKinsey March 2026 AI Employment Report — 12% of job tasks automated, 8% new AI-related categories created. Net employment flat — with significant variation by sector
If you missed our previous updates, don’t worry, here they are:
Claude Marketplace Launch, Google Pentagon Deal & Anthropic Lawsuit | Weekly Digest