The AI build-out hit a political wall this week. US voters turned against data centers at the ballot box, and Nvidia’s board opened an investigation into its own CEO. Meanwhile companies started counting what their AI tools cost, and security firms lined up about a billion dollars in deals for startups that protect AI agents. Here are the ten stories that mattered most, counting down to the biggest.
10. Assistants get voices, faces and a faster model
Google released Gemini Omni 1.2 Flash, a cheaper and faster multimodal model that handles longer clips. Meta gave its Muse agent a real-time voice mode and an animated avatar that US users can video chat with.
Apple is feeling the pressure. According to Bloomberg, CEO John Ternus is pushing Craig Federighi to speed up the Siri roadmap so it can compete with OpenAI’s and Meta’s assistants.
9. A humanoid robot maker lists on Nasdaq
Agility Robotics became the first US maker of legged humanoids to trade on Nasdaq, after shareholders approved its $2.5B SPAC. The stock closed its first day only 2% up, which raised questions about how big the humanoid market really is.
The models behind the robots kept coming. NVIDIA released Cosmos 3 open models for interactive world simulation, and LiberAI launched Liber-0, trained on 2.2 million hours of human experience for robot manipulation.
8. Agents arrive at work, and regulators take notice
Perplexity launched Teammates, always-on agents for the workplace. AWS refitted Kiro to orchestrate agents, and Airbnb plans an AI travel agent for early 2027.
Some are less keen. Cloudflare’s CEO said agents are a bane for some website operators, and UK ministers are seeking powers to investigate when agents cause harm.
7. AI’s appetite for memory squeezes PCs and Macs
High RAM prices have stalled Microsoft’s AI PC push and forced Windows 11 tweaks for 8 GB machines. Apple’s Mac sales are at a record, but memory costs threaten its margins.
Raspberry Pi raised prices again, a month after its last increase. Sam Altman spent election night defending his comments on PC memory.
6. Companies put limits on AI coding tools
Goldman Sachs is reportedly pulling Claude Code from its bankers and traders, and Apple is said to cap its engineers at $200 a day in AI spending. Revolut cut its AI coding bill tenfold by moving from Claude to self-hosted models, and Gartner says open models handle 90% of agentic work at a twentieth of the cost.
Open source is feeling it too. Linus Torvalds says the flood of AI-prompted kernel patches may have peaked, with fixes now arriving faster than new bug reports.
5. Security giants pay up for AI agent startups
CrowdStrike is paying nearly $500M for Fiddler AI and Dreadnode, two startups that monitor and test AI agents. Two days later, Palo Alto Networks was reportedly in talks to buy Clutch for $500M.
Recent incidents explain the demand. Thieves stole environment variables from the Trigger workflow tool, and Anthropic declined to fix a Claude Dispatch flaw that researchers say can be chained into remote command execution.
4. OpenAI ships a bug-hunting model as open models catch up
OpenAI launched GPT-6 Cyber, which finds exploitable vulnerabilities and drafts fixes. OpenAI says it stayed within scope more than 99% of the time during testing, though some ask whether a model alone is enough.
Free models are close behind. In an experiment where models attacked simulated networks, DeepSeek and GLM came not far behind OpenAI’s most advanced cyber model, and anyone can download them.
3. The money at Anthropic and OpenAI gets bigger and harder to read
Anthropic’s revised IPO filing shows Q3 revenue near $17B but compute commitments that could top $1T, with margins getting worse instead of better. TechCrunch called it a company still grappling with the economics of compute.
OpenAI went the other way and kept raising. It reportedly added another $11 billion at a $1.25 trillion valuation, in a round expected to close in the coming weeks.
2. Nvidia’s board investigates Jensen Huang
Nvidia opened an investigation into its CEO over his relationship with Megaspeed executive Alice Huang. A US intelligence memo describes her as recruited by China’s Ministry of State Security to spy on the company. In a memo to staff, Huang admitted to an affair.
Congress has summoned Huang over espionage concerns, and Yale’s Jeffrey Sonnenfeld puts the odds that he keeps his job at about 50-50. He runs the world’s most valuable AI chip company.
1. Data centers lost the US midterms
Voters backed candidates who favor building moratoriums and passed ballot measures that put new obstacles in front of data centers. Wired summed it up: voters hate data centers. Texas, the fastest-growing market, may lose its lead, and in Michigan a candidate who ran against data centers won a House seat and plans to push for a federal moratorium.
The industry is already adjusting. Meta warned investors that the backlash could slow its AI expansion, and OpenAI said it will stop using NDAs with public officials on data center deals. Protests calling for a moratorium are planned across Europe.
That is the week. Most of the top stories were about where AI gets built and who pays for it, not about new models. See you next Monday.