Picking up where last week left off, the man who built the chip era stepped down, super intelligence became the default brain of Search, and Anthropic took a new flagship model on the road to public investors.

The chip era lost its founder, in one abrupt week. On November 12, Nvidia announced that Jensen Huang had resigned as president, chief executive and board member, effective immediately, ending 33 years leading the company he co-founded in 1993. Chief financial officer Colette Kress, in that role since 2013, became interim chief executive while the board searches inside and outside the company, and Harvey C. Jones became the board’s first chair. Nvidia gave no reason in its announcement, which was filed with the Securities and Exchange Commission, and its third quarter results stay on schedule for November 17.

Why it matters for how we live and build: Nvidia sits at the centre of the super intelligence buildout, so who leads it, and how steady this handover is, shapes the price and supply of the computing power almost every builder rents.

Google made super intelligence the default answer in Search. On November 12, Google released Gemini 4 Hydrogen, the third model in its Gemini 4 series, and made it the default model in AI Mode in Search and the model answering the hardest questions in AI Overviews. Google prices it for developers at $1.50 per million input tokens and $7.50 per million output tokens across a one million token context window, and reports a score of 51 on the Artificial Analysis Intelligence Index, just ahead of the far pricier Gemini 4 Argon at 50.

Why it matters for how we live and build: when the default model in Search changes, the change reaches billions of everyday questions at once, and capable intelligence at this price quietly resets what every app, publisher and business built around search can assume.

Anthropic shipped its strongest model, and priced it lower. On November 9, Anthropic released Claude Fable 5.2, its most capable generally available model, alongside Claude Mythos 5.2, the same underlying model with fewer safeguards for organisations verified for cybersecurity and life sciences work. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1, with output up to 30 percent faster, a one million token context window and availability across Anthropic’s own platform and the major clouds.

Why it matters for how we live and build: the frontier is getting cheaper and faster at the same time, and a two tier release, open to everyone in one form and restricted in another, is becoming the normal way powerful and sensitive capability is shared.

Anthropic took its story to public investors. On November 10, Anthropic launched the roadshow for its proposed initial public offering, with an expected price range of $70 to $78 per share and a planned listing of its Class A shares on the Nasdaq under the ticker ANT. The pricing and trading dates reported so far are targets, not confirmed dates.

Why it matters for how we live and build: a public Anthropic gives everyday investors a direct way to own a frontier lab, and gives all of us public numbers, quarter after quarter, for what building super intelligence actually earns and costs.

Agents moved into the office, and into our own buildings. On November 10, Anthropic launched Claude Whiteboards and Claude Sheets in beta and made Claude Data generally available, while OpenAI opened Businesses in ChatGPT in beta, letting companies claim a page inside ChatGPT and, for eligible United States ecommerce and online service businesses, launch Business Agents. Two days later, on November 12, Anthropic opened a private beta of self hosted Claude Managed Agents, so enterprises can run the harness, sandboxes, credentials and session history on their own servers, with only the model itself staying with Anthropic.

Why it matters for how we live and build: super intelligence is moving deeper into the tools we already use, and the question for serious teams is shifting from whether agents do the work to where they run, whose credentials they hold and who can audit what they did.

Security agents learned to attack a site, then draft the fix. On November 12, at its Cyber at the Frontier keynote, OpenAI launched Daybreak Red Team, in early access, which sends fleets of agents to penetration test web applications using source code and application context, validate flaws reported by scanners and researchers, and retest patches. Its companion, Daybreak Defense Factory, built on GPT-6 Cyber and available through a plugin in Codex, scans codebases as they change and prepares fixes as merge ready pull requests with tests, for a human engineer to approve before anything deploys.

Why it matters for how we live and build: defence is becoming a continuous agent workflow rather than an occasional audit, and the human signature on the merge is the control that keeps that speed trustworthy.

Robots were graded on a real Mercedes line. On November 10, Google DeepMind released Gemini Robotics 2.5, its most capable vision language action model to date, and reported that Apptronik Apollo 2 humanoids running it sorted automotive parts into sequencing trays on active Mercedes-Benz production lines in Germany and Hungary with a 96 percent success rate. DeepMind evaluated the model across more than 200 tasks and is rolling it out to robotics partners. The company has not published how many trials sit behind the 96 percent figure, so we read it as a strong pilot result, not a production guarantee.

Why it matters for how we live and build: a humanoid doing one repetitive job well on a live line is the shape progress actually takes, and honest denominators are what will separate working robots from demonstration robots.

Quantum and classical machines simulated a protein together, at a new scale. In a preprint reported on November 12, a team from Cleveland Clinic, RIKEN and IBM described the largest heterogeneous quantum classical electronic structure calculation to date, simulating full protein complexes of 12,635 and 11,608 atoms, more than 40 times larger than the prior benchmark. The team paired two 156 qubit IBM Heron processors with leadership supercomputers, including Fugaku, ran 21,006 circuits over more than 239 hours of quantum processor time and collected 3.0 billion measurement shots. The work is a preprint and has not yet been peer reviewed.

Why it matters for how we live and build: drug discovery lives or dies on how accurately we can model molecules, and a hybrid workflow that handles proteins at this scale points to a practical division of labour, quantum machines for the hardest fragments and classical machines for everything around them.