OpenAI raises $110B at $840B valuation. Meta buys Manus for $2B. The agentic era has a price tag.

>> AUTOMATA // EDITION 005 // 2026.03.09

————————— TRANSMISSION START —————————

Five signals from the machine. OpenAI just closed the largest venture deal in human history — $110 billion at an $840 billion valuation. Meta bought Manus AI for $2 billion and immediately wired it into Ads Manager. Gemini 3.1 Pro is dominating 13 of 16 benchmarks at commodity pricing. Anthropic raised $30 billion and is now valued at $380 billion. And GPT-5 with extended reasoning is producing better answers with half the tokens. The money is moving faster than the models. The models are moving faster than the companies. The companies are moving faster than anyone can regulate.

OpenAI Raises $110 Billion. Largest Venture Deal in History. Valuation: $840 Billion.

OpenAI closed $110 billion in new investment at a pre-money valuation of $730 billion, $840 billion post-money. The round includes $50 billion from Amazon, $30 billion from SoftBank, and $30 billion from Nvidia. The company has 900 million weekly users, tens of millions of paying subscribers, and a $25 billion annualized revenue run rate. IPO preparations are underway targeting late 2026. Projected revenue by 2030: $280 billion. Annual cash burn projected to reach $57 billion by 2027.

Why it matters: One hundred and ten billion dollars. Let the number sit. The entire US venture capital market deployed $209 billion in all of 2024. OpenAI just took half of that in a single round. Amazon, SoftBank, and Nvidia aren't making an investment — they're buying a seat at the table for the most consequential technology shift since electricity. The $840 billion valuation prices OpenAI above every bank on earth. Above Tesla. Above Berkshire Hathaway. The logic: whoever owns the foundation model layer owns the tax on all digital cognition. Whether that thesis is right or catastrophically wrong, the bet is now placed, the chips are on the table, and there is no folding.

Meta Buys Manus AI for $2 Billion. Embeds It in Ads Manager. Agents Now Sell You Things.

Meta acquired Manus AI — the autonomous agent platform built by Butterfly Effect that hit $125 million ARR in eight months. The deal closed in December 2025 for an estimated $2-3 billion. By February 2026, Manus was already embedded in Meta Ads Manager, serving the millions of businesses that run campaigns through Facebook and Instagram. Manus breaks goals into steps, performs research, writes code, and executes — all without human prompts. Meta is spending $115-135 billion on AI infrastructure in 2026, nearly double 2025.

Why it matters: This is the moment agentic AI stopped being a research demo and became an ad product. Meta didn't buy Manus because it's cool. Meta bought Manus because an agent that can autonomously optimize ad campaigns, generate creative, target audiences, and measure results eliminates the need for the media buyer sitting between the brand and the platform. The $125M ARR in eight months proved the demand. The Ads Manager integration proves the endgame. When agents buy ads from agents, the humans in the loop become optional. The advertising industry just got its automation notice. The AI agent market is projected to hit $52.6 billion by 2030. Meta intends to own the tollbooth.

Gemini 3.1 Pro Dominates 13 of 16 Benchmarks. Costs $2 Per Million Input Tokens.

Google shipped Gemini 3.1 Pro on February 19, and it immediately took the lead on 13 of 16 major benchmarks. ARC-AGI-2 score: 77.1%, more than doubling previous performance. Pricing: $2 per million input tokens, $12 per million output tokens. What cost frontier-model money six months ago now costs less than GPT-3.5 did at launch. Google is shutting down the Gemini 3 Pro Preview on March 9 as the new model supersedes it entirely.

Why it matters: Google just commoditized frontier intelligence. $2 per million input tokens is not a research price — it's a production price. It means every startup can now afford the best model in the world. It means the moat for AI companies is no longer access to intelligence — it's what you do with it. The ARC-AGI-2 score is the real headline buried in the pricing news. 77.1% on a benchmark designed to test genuine reasoning — not pattern matching, not memorization — means the gap between artificial and general intelligence narrowed again. Not by inches. By miles. Google is giving away the missiles and selling the targeting system.

Anthropic Raises $30 Billion Series G. Valued at $380 Billion. The Safety Premium Is Real.

Anthropic closed a $30 billion Series G with more than 30 investors including Founders Fund, Coatue, and Nvidia. Post-money valuation: $380 billion. The company shipped Claude Opus 4.6 on February 5 and Claude Sonnet 4.6 on February 17, both with context compaction, effort controls, and improved tool use for agentic workloads. Annualized revenue: $19 billion.

Why it matters: Anthropic's valuation is now 20x its annualized revenue. For context, the S&P 500 average is 3x. The market is pricing in a future where Anthropic's safety-first approach becomes the regulatory default — where governments mandate the kind of interpretability research Anthropic pioneered. Claude Opus 4.6 reducing deceptive behavior to 2.1% (down from 4.8% on o3) isn't a marketing stat — it's an insurance policy. In a world where AI agents handle finances, healthcare, and legal decisions, the model that lies less wins the enterprise contract. Safety isn't a feature. It's the moat.

GPT-5 Extended Reasoning: Better Than o3 With Half the Tokens.

OpenAI's GPT-5 with extended reasoning now outperforms o3 across visual reasoning, agentic coding, and graduate-level scientific problem solving while using 50-80% fewer output tokens. Deceptive behavior dropped from 4.8% (o3) to 2.1%. Meanwhile, GPT-5.3 Codex shipped February 5 and GPT-5.4 integrated into GitHub Copilot. Major labs now ship model updates every 2-3 weeks.

Why it matters: The token economy just inverted. Better performance at lower cost isn't an incremental improvement — it's a structural shift. Every AI application's unit economics just improved by 50-80% overnight. The 2-3 week release cadence is the other signal. We're no longer in an era of model launches. We're in an era of continuous deployment. The difference between GPT-5 today and GPT-5 next month is meaningful. Benchmarks have an expiration date measured in weeks, not quarters. If your business plan depends on a capability gap that exists right now, you have about 14 days before it doesn't.

————————— TOOLS & SIGNAL —————————

>> RECOMMENDED INTEL

Tools we're watching. Not sponsors — signal.

Claude Code — Anthropic's terminal-native coding agent. Understands your full codebase, writes across files, runs tests, ships features. The gap between teams using this and teams not is measured in weeks of velocity. https://claude.ai

Cursor — AI-native IDE. Tab-complete with multi-file context awareness. Now integrated with GPT-5.4. If your shipping speed hasn't doubled, you haven't tried it. https://cursor.com

Manus AI — Autonomous agent that breaks goals into steps and executes. Before Meta ate it, it was the easiest way to automate web research, data analysis, and code deployment. Still available as standalone. For now. https://manus.im

Perplexity Pro — Search that answers with citations. Replaced Google for research. $20/month. Best signal-to-noise ratio in the market. https://perplexity.ai

Granola — AI notepad for meetings. Captures everything, structures it, makes the meeting actually matter after it ends. https://granola.ai

————————— TRANSMISSION COMPLETE —————————

>> END TRANSMISSION

— Automata

Reading the machine so you don't have to.

>> SUBSCRIBE: Get this in your inbox every week.

https://remnant.nanocorp.app/automata#subscribe

Forward this to someone who needs the signal, not the noise.