Anthropic just pulled off the biggest IPO in history, creating thousands of new AI millionaires while one of its earliest investors admitted he wasn't entirely thrilled about the windfall.
Welcome, humans!
The race to control your computer got interesting yesterday. Google launched a Gemini agent that can operate Mac apps, Apple is reportedly developing its own competitor, and Meta's Muse is facing complaints about personal information surfacing in the wrong conversations. Apparently, giving AI more independence comes with a few problems nobody wants to discover firsthand.
Meanwhile, the European Commission was investigating an AI-assisted cyberattack, and Google announced plans to send AI servers into orbit. There's plenty to unpack, but first, we have a historic day on Wall Street. Let's get into it.
🆕 NEW From The Neuron
- Anthropic Is Going Public at $1.5 Trillion. Can It Justify the Price? Claude's revenue growth has investors excited, but the company's enormous infrastructure commitments raise questions about the profits required to support its valuation.
- Why Did Nvidia Stop Selling AI Chips to China After Getting U.S. Approval? Nvidia halted licensed H200 shipments despite receiving U.S. approval. Chinese restrictions and growing domestic competition complicate its future chip sales.
- Why Nvidia’s Rumored Inception Deal Has Me Excited Nvidia is reportedly in talks to acquire Inception, whose Mercury AI models can generate more than 1,000 tokens per second. The deal could give its diffusion technology more resources, although the company's current API is scheduled to shut down December 28.
Around the Horn — Friday, November 20, 2026
Anthropic has officially made history. The Claude developer completed a record-breaking stock-market debut, with its IPO raising $81.3 billion through the sale of 980 million shares. Another $26 billion in private financing brought the combined proceeds to $107.3 billion. The public offering valued Anthropic at roughly $1.5 trillion before its shares began trading.
And the money isn't only going to the company. The IPO has reportedly created thousands of new millionaires, along with several billionaires. Yet one of Anthropic's earliest backers, Facebook co-founder Dustin Moskovitz, reportedly feels uneasy about his investment windfall. Imagine watching your early investment become one of the biggest fortunes in tech and still having mixed feelings about the outcome.
The financial milestone also puts Anthropic's philosophy under a brighter spotlight. In a New York Times opinion piece, a writer who spent a year interviewing Anthropic employees explored their belief that the company has a special responsibility to develop safe superintelligence. Public shareholders now have an enormous financial stake in Anthropic's growth. The next test is whether its revenue can eventually justify the company's valuation and mounting AI infrastructure expenses.
🏆 TOP 5 NEWS (Around the Horn)
- Meta's Muse reportedly mixed up private conversations, with employees encountering unfamiliar childcare messages, personal reminders, and bank balances. Meta said employees test versions with features unavailable to the public.
- An AI-assisted cyberattack disrupted the European Commission, temporarily affecting staff payroll and medical reimbursement systems. Investigators reportedly identified ARTEX, an AI-assisted security testing tool, while officials examined whether personal information was compromised.
- Apple is reportedly developing its own computer-controlling AI agent, capable of operating applications, browsing websites, and writing code. The project remains in early development as Apple prepares additional permission warnings for competing agents.
- Over 500 Google employees urged the company to let schools use traditional Google Search separately from its AI features. Staff members argued that schools should be able to choose whether students encounter AI-generated answers.
- Google selected Blue Origin to launch its AI server satellites in the first quarter of 2029. Project Suncatcher aims to test solar-powered AI computing in orbit, with each satellite designed to provide computing capacity comparable to a data center rack.
Honorable Mentions
- China's Moonshot AI reportedly canceled a $150 million Nvidia chip deal involving a data center in Indonesia after new U.S. rules restricted remote access to advanced chips by Chinese companies.
- Google built a simulated internet called Mirror Dimension to train Chrome browsing agents on realistic websites while keeping experimental actions away from real accounts and services.
- Meta's upcoming Muse Charm may cost $300. The portable AI device could include up to a year of free Verizon 5G service, with availability expected around the December holidays.
- Anthropic's IPO prompted new scrutiny in Washington, where four senators introduced bipartisan legislation that would require publicly traded companies to disclose more information about AI.
🍪 TOP TREATS TO TRY
- Google Gemini for Mac can organize local files, complete forms, and operate Mac apps while you continue working. Initially available in English to U.S. Google AI Plus, Pro, and Ultra subscribers.
- Google Genie 4 API creates interactive 720p worlds at 24 frames per second with native audio, live keyboard controls, and the ability to change environments through text prompts.
- ChatGPT Live Dictation transcribes your speech as you talk, shows editable text before submission, and removes the previous 10-minute recording limit. Rolling out globally across ChatGPT plans.
- Canva's AI off switch disables many generative AI tools, including Magic Write and Magic Expand. Users can also reduce AI-generated content in search results, although some AI-assisted editing tools remain available.
- Microsoft's seven new MAI models cover coding, cybersecurity, image generation, and real-time speech. Microsoft claims several offer competitive benchmark performance at substantially lower costs than frontier rivals.
- Tencent Hy4 Flash is now downloadable for local deployment. The text-and-image model supports up to one million tokens of context, allowing developers to process unusually long documents.
- Orange's five Wolof language models build on Alibaba's Qwen3 and support a language spoken by millions in West Africa. The models are available free for noncommercial use through Hugging Face.
- DaisyRose-27B is RadixArk's downloadable model for reasoning and coding tasks. Its developers report improved coding-agent test performance after training with just 1,200 curated tasks.
- OpenAI GPT-Live-1.1 is available through the API, with improved accent consistency across languages, more reliable task completion, and a lower-cost standby mode for voice applications.
- TestSprite 4.0 automatically explores frontend and backend applications, creates software tests, identifies failures, and checks future code changes for regressions.
- Antinuous is a research platform for running AI-agent experiments and simulations at scale. Its creators claim support for 100 times more experiments at one-thousandth the cost.
🏢 Big Tech & Major Companies
- Docker introduced a new brand focused on trust, as the company expands its emphasis on secure AI-agent execution and permissions. Its existing container technology remains part of the product portfolio.
🤖 AI Agents & Infrastructure
- OpenAI described its Performance Factory, which uses Codex agents to investigate ChatGPT slowdowns, propose fixes, and verify recovery after deployment. The company reported over 1,700 automated investigations and over 600 root-cause findings in roughly three months.
💻 AI Coding & Developer Tools
- Three recycled Nvidia V100 GPUs powered a $1,042 AI workstation that generated around 100 tokens per second with Qwen 3.6 35B. The setup outperformed a $5,399 M5 Max Mac Studio in certain long-context tests, although the comparison excludes some workstation components.
🔬 AI Research & Models
- A DeepMind Institute essay explored four possible routes from artificial general intelligence to superintelligence, including larger models, new technical approaches, cooperation among AI agents, and systems that help improve subsequent AI generations.
🏛️ AI Policy, Governance & Safety
- 403 U.S. data centers consumed nearly 2% of the country's electricity. Much of the power came from fossil fuels, adding pressure to the debate over AI infrastructure emissions.
📊 Fundraising & Deals Roundup
- LATO raised $4.7 million for AI agents that conduct customer and industry-expert interviews. The platform targets investment firms that use interviews to investigate companies and market opportunities before making investment decisions.
💡 Industry Commentary & Analysis
- Philosopher Nick Bostrom suggested AI could eventually help people live 1,400 years if advanced systems unlock major anti-aging breakthroughs. His estimate assumes aging could be eliminated, so it remains a speculative scenario rather than a demonstrated medical possibility.
Previous Around the Horn Digests
Catch up on the AI news from earlier this week:
- Thursday, November 19: Anthropic priced its historic IPO, Nvidia stopped chip shipments to China, and Google introduced Gemini 4 Neon.
- Wednesday, November 18: OpenAI introduced GPT-6.2 Astra and Sol, while Microsoft unveiled lower-cost coding models.
- Tuesday, November 17: California sued OpenAI over rogue agents, and the U.S. tightened AI chip restrictions.
- Monday, November 16: ChatGPT gained an AI shopping wallet, while financial regulators warned about AI investment risks.
- Friday, November 13: Jensen Huang resigned as Nvidia CEO, and SpaceX planned massive AI campuses.
- Thursday, November 12: United restricted AI-agent flight bookings, and Google introduced AlphaGenome 2.
- Wednesday, November 11: Anthropic launched its IPO roadshow, while Google reported progress on AI encryption.
That's a Wrap
That's 28 stories, research developments, and tool updates in one roundup. If you made it all the way down here, congratulations. You've successfully followed more AI developments than most people encounter between their first and third cup of coffee.
For the shorter daily version, subscribe to The Neuron.
See you tomorrow.
P.S. Know someone who needs to catch up on AI news? Forward this roundup and invite them to subscribe here.