Good morning. The Jensen Huang story that broke two days ago has gone from “board investigation” to “preparing to oust the CEO” with remarkable speed, and a second affair has entered the picture. Elsewhere, Google is shipping a decision model and a new genomics model, Apple is finally giving local LLM tooling a proper home, and OpenAI’s quarterly revenue crossed ten figures for the first time.
Huang’s situation goes from bad to worse. The Wall Street Journal reported Wednesday that Nvidia’s board is preparing to oust Huang, with the stock down roughly 11% over five days. The accelerant: a New York Times story alleging a second affair, this time with Kimberly Powell, Nvidia’s VP of Healthcare and a direct report. TechCrunch’s timeline ties both threads together and notes Nvidia has engaged an outside firm to identify replacements — a tall order for a CEO with around 60 direct reports and no succession plan.
The governance angle is what actually matters. Several HN commenters pointed out that Huang holds only 3.4% of Nvidia with no dual-class voting shares, so unlike Zuckerberg or Musk, he has no structural defense if the board moves. Others flagged that his personal security reportedly detained someone who filmed the Powell encounter and offered money for an NDA, which doesn’t feel like a first-time response. The second-affair writeup is the cleanest summary if you’re catching up.
OpenAI crosses $10B in quarterly revenue. Per TechCrunch, Q3 2026 came in around $10B, up roughly 50% from Q2’s $6.7B, driven by enterprise products and agents. That’s a $50B annualized run rate — still well behind Anthropic’s $81B, though the two companies account differently (Anthropic includes cloud partner sales, OpenAI reports only its direct share). Context for yesterday’s $1.4T IPO pricing.
Google launches the Gemini Decisions API. The new Gemini Decision 1 model runs up to 1,000 yes/no, choice, or rubric questions in a single forward pass with 150-250ms latency, returning confidence scores without token-by-token decoding. At $0.10 per million input tokens it’s about 5x cheaper than Gemini 3.5 Flash-Lite for decision tasks, and multimodal input (including audio and video) is the real differentiator. HN reception is mixed: one commenter notes Liquid offers a comparable text model at $0.02/M, and respan/span-02 keeps getting cited as the quality bar to beat.
Apple’s Harbor brings local models to the Mac. 9to5Mac reports Apple is readying a Mac app under its Claris brand that simplifies running local LLMs, with Thunderbolt 5 clustering that pools unified memory across multiple Macs for larger inference. One HN commenter did the math: a 512GB Mac Studio amortizes to about $1,000/month over three years versus the same amount currently spent on Claude Ultra for five developers. Ollama and LM Studio veterans welcome a vendor-supported option but worry about bloat.
DeepMind’s AlphaGenome 2 goes cell-specific. Google’s new DNA model is trained on over 100 million single-cell profiles across 68 tissues and identifies nearly twice as many expression-altering variants as its predecessor. The conditional architecture accepts specific cell type inputs — including from researchers’ own single-cell experiments — rather than averaging across broad tissues, which is the part that actually changes what’s possible. The Verge has more on the disease-mechanism implications. It’s available via API for non-commercial research now, commercial access in private preview.
That’s the morning. Watch Nvidia’s open today, and if you’re a Fable 5.2 customer with a budget problem, Harbor’s private preview may be worth getting on the list for.