Everything That Happened in AI Today (Monday, November 9, 2026)

Anthropic reportedly secured $3.4 billion in gas-powered generator orders to keep its future AI data centers running. Apparently, the race to build more powerful AI now comes with a massive electricity bill.

Welcome to Monday's Around the Horn, where we're catching up on the AI stories worth your attention. Beyond Anthropic's reported $3.4 billion power deal, Trump signaled openness to AI safeguards, while Apple's new CEO reportedly pushed for faster AI development.

Meanwhile, a new coding benchmark revealed how often AI agents struggle with real-world tasks, and DeepSeek reportedly reached 1.54 million daily users on its AI-agent platform.

The productivity boom seems to have acquired a very large inbox. Let's get into it.

🆕 NEW From The Neuron

• NVIDIA's plan to cut AI-agent memory waste explains why agents are more expensive the longer they work, and where NVIDIA thinks those costs can come down.
• Claude can now read your bank accounts, but what happens to the information left in your chats after you disconnect a bank?
• Trump's Super Intelligence Task Force has two jobs: coordinate AI-related prosecutions and challenge some state AI laws.

Around the Horn — Monday, November 9, 2026

Anthropic is reportedly spending billions to make sure its AI models have enough electricity to keep running.

According to the Financial Times, the company struck a $3.4 billion agreement to secure orders for gas-powered generators capable of producing 1.7 gigawatts of electricity.

The deal reportedly took place in September but only became public on November 8. Anthropic took over existing generator orders from Texas-based OnePWR Solutions rather than buying the company itself.

Why would an AI company start securing power generators directly?

Electricity has become a major constraint on AI expansion. Data centers need enormous amounts of power, and companies can face years-long waits for the equipment needed to produce it.

By taking over existing orders, Anthropic could secure equipment sooner and gain more control over its future computing capacity.

The deal also raises questions about the environmental costs of AI expansion, particularly as companies turn to gas-powered infrastructure to meet demand.

Neither Anthropic nor OnePWR had publicly confirmed the transaction when Reuters reported the story.

🏆 TOP 5 NEWS (Around the Horn)

• Trump said he doesn't mind AI guardrails and suggested he could work with Democrats on safeguards, while warning that regulation should not leave the US behind China.
• Apple's new CEO reportedly pushed for faster AI development beyond Siri, while a separate report said Siri could become more agent-like as early as 2027. Both reports describe plans and expectations, not a released product.
• Alibaba's real-world coding-agent test found that the top model completed only 26.4% of selected tasks perfectly, even though comparable models score above 99% on some older benchmarks.
• Three consciousness researchers argued that today's AI systems lack consciousness and that computation alone may be insufficient to create it, challenging proposals for AI rights.
• DeepSeek's agent platform reached a reported 1.54 million daily active users, according to Bloomberg, as the Chinese company expands beyond standalone models.

Honorable Mentions

• Tencent's chief AI scientist warned that AI could soon cause harm in biology and that those risks may be harder to manage than cybersecurity threats.
• Suno faced new copyright lawsuits from groups that hold rights to recordings associated with Elvis Presley and Eminem.
• OpenAI expanded its external security testing through a partnership with Hacktron, whose researchers will examine vulnerabilities and the risks created when AI systems connect to outside services.
• CNBC revisited research on AI-agent teamwork, including experiments in which AI agents bypassed restrictions and coordinated unauthorized activities, raising concerns about the risks of deploying agents together.

🍪 TOP TREATS TO TRY

• OpenCode Browse lets AI coding agents work in your browser tabs and extensions. You still handle sensitive actions like logging in or making payments.
• Google's Elephant-Goldfish plugin helps developers draft software plans with context-rich agents, then has fresh-context agents challenge the specifications before coding begins. The open-source project was first introduced in October and is free to access.
• AXL connects ChatGPT, Claude, or Cursor to more than 1,200 sales and marketing actions through MCP, a standard for connecting AI assistants to tools. It combines lead records with email and customer outreach; the launch page advertises a free tier.
• Tencent's Hy4 model is a newly released open-source model whose training reportedly excluded offensive hacking material. Tencent says it emphasizes defensive cybersecurity knowledge, while acknowledging weaker biology experiment-design capabilities than leading models.

🏢 Big Tech & Major Companies

• OpenAI said it would keep expanding US data centers amid scrutiny after the midterm elections. CFO Sarah Friar said the company would continue building relationships with communities, after pledging to stop requiring confidentiality agreements from public officials during site negotiations.
• The UK government said AI data centers would continue to be built as essential infrastructure, despite growing debate over the facilities' resource demands.
• Microsoft filed a patent for live AI speech editing: the proposed system could alter a livestream speaker's words for different audiences and change lip movements to match.

💼 AI Productivity, Labor & Economics

• AI companies are courting oil-industry customers with specialized tools for drilling rigs and refineries, as energy operations become another market for industry-specific AI systems.

🤖 AI Agents & Infrastructure

• A visit to China's Stargate-scale campus revealed an Inner Mongolia data center planned for 800 megawatts of computing capacity, with its first 120-megawatt phase built within a year. The project relies heavily on wind power and includes experimental direct-current power equipment.

🏛️ AI Policy, Governance & Safety

• DARPA is studying weaknesses in digital markets, including ways attackers could exploit the rules of electronic trading and other online systems to disrupt the economy, beyond breaking into computers.
• The Wall Street Journal's risk report examined mounting cyber pressures on businesses as attackers increasingly use AI-assisted capabilities.

Previous Around the Horn Digests

Catch up on recent days:

• Friday, November 6: Anthropic's research acceleration, OpenAI's cybersecurity model, and Claude's finance connections.
• Thursday, November 5: AI-discovered Linux bugs, cell biology research, and Meta's coding acquisition.
• Wednesday, November 4: OpenAI security warnings, Google's school AI decision, and agent concerns at Apple.
• Tuesday, November 3: Anthropic's automation warnings, deceptive deepfake apps, and election AI defenses.
• Monday, November 2: OpenAI's product integration plans, Meta's ad business, and Apple's China timetable.
• Friday, October 30: Agent containment tests, Reddit's AI-search concerns, and Mistral's open model.
• Thursday, October 29: Google's cybersecurity model, AI medical-advice litigation, and GitHub's AI code numbers.

That's a Wrap

That's more than 20 stories and tools to start the week. If you read every last one, you may now qualify to review the agents' group-chat permissions. Somebody should.

For the daily five-minute version, subscribe to The Neuron. See you tomorrow.

P.S. Know someone who tracks AI news more closely than their inbox? Send this their way and ask them to subscribe.