Reading time: 3 minutes
TLDR: Nvidia's board named CFO Colette Kress interim CEO after Jensen Huang resigned this week. Four US civil society groups sued Anthropic, Google, Meta, OpenAI and SpaceX's AI unit, asking a court to freeze the development and provision of advanced AI systems. Google rolled out Gemini 4 Hydrogen across its products with the same API price over a full one-million-token context. And a Microsoft 365 outage that reached more than 70 countries reminded teams what depending on one vendor costs. Four signals from a week in which the people and institutions around AI took the spotlight from the models.
Nvidia's CEO steps down
Jensen Huang resigned as Nvidia CEO, and the board named CFO Colette Kress as interim chief executive. The circumstances are still being reported and some details are contested, so this edition sticks to what matters for the AI supply chain. A Senate committee wants Huang and Commerce Secretary Howard Lutnick to testify, and South Korea's science minister said Nvidia confirmed it will still deliver the 260,000 GPUs it pledged for Korean data centres, with joint projects continuing unchanged. Nvidia supplies the chips most frontier labs depend on, so a leadership change matters to anyone planning around its roadmap, even if shipments are, for now, unaffected.
A lawsuit asks courts to freeze advanced AI
Four US civil society groups filed suit against Anthropic, Google, Meta, OpenAI and SpaceX's AI unit, seeking an injunction against further development and provision of advanced AI systems. The filing cites the labs' own public warnings about extreme AI risks. It is an early filing and nothing has been decided, and sweeping orders of this kind are hard to obtain, so the practical risk to your tools this month is low. What is new is the argument: it turns safety statements made in blog posts and testimony into legal exhibits. Companies that publish strong risk language should expect it to be quoted back to them.
Google makes its fast model widely available
Google rolled out Gemini 4 Hydrogen across AI Studio, the Gemini API, Gemini for business, the Gemini app and Search. It is positioned as the quick model for everyday coding and agent tasks, below the slower Carbon and Argon. The API price is $1.50 per million input tokens and $7.50 per million output tokens across the entire one-million-token context window, with no long-context surcharge. In Google's Antigravity tool, an orchestrator can switch between Carbon and Hydrogen within a session, and Google reports near-Carbon task success with 24% less quota use on one coding benchmark. That is a vendor claim, so treat it as a hypothesis to test on your own workloads.
One outage, many dependencies
A Microsoft 365 failure that began on November 12 left users in more than 70 countries without Outlook, Teams, SharePoint and Copilot. Microsoft attributed it to a network capacity update and a recent infrastructure change, not to a cyberattack or to AI, and reported broad recovery after more than 20 hours. The lesson for AI adoption is not specific to Microsoft. If your team's email, documents, chat and AI assistant all sit with one vendor, one failure takes all of them down together, and the fallback plan should say who works how while that happens.
Capital kept moving in the background. SpaceX said it plans to invest more than $74 billion in AI computing facilities in the United States, Anthropic reportedly began its IPO roadshow, and OpenAI reportedly reached about $10 billion in quarterly revenue while still posting heavy losses.
The quiet signal
A supplier changed its chief executive, a court was asked to slow the industry down, a vendor lowered the cost of long context, and an outage showed how concentrated the stack is. None of these is about model quality. They are about who controls the inputs, who can be sued, and how much risk a team carries when one provider fails. Those are the variables worth tracking in the months ahead.
Speech models ace the lab. Canto's built for your commute.
Speech recognition is great on clean studio audio. Real dictation happens at your desk, on a walk, or in a loud cafe. That's why Wispr Flow built Canto, its own speech model, trained on how people really talk.
On 10 hours of real dictations from 2,300+ speakers, it had the lowest word error rate of every model tested, ahead of models from Google, OpenAI, AssemblyAI and Deepgram. Background music, traffic, someone talking next to you, a whisper: on a challenge set full of exactly that, Canto had the lowest error rate among real-time models tested.
Talk naturally and Wispr Flow types clean, formatted text in any app. 4x faster than typing.
P.S. If you run a newsletter or are thinking about starting one, the platform behind AI Quiet Signal is beehiiv. It handles the infrastructure so you can focus on the signal.
