Anthropic cuts its roadshow valuation to $1 trillion.

Anthropic began its IPO roadshow today. It plans to issue 825 million shares of Class A common stock at $70 to $78 apiece. Its target valuation tops out at $1.02 trillion. The public offering could raise up to $64.4 billion; with a $26 billion private placement, the total reaches $90.4 billion. The shares will trade on Nasdaq under ANT.

The S-1 carries a heavier line: Anthropic has committed to nearly $900 billion in compute spending. Some future infrastructure depends on financing outside its control. So this roadshow is less a finish line than a way to open up funding channels.

Each of the seven co-founders is selling 3,378,378 shares. At the range’s $74 midpoint, each would cash out nearly $250 million. Afterward, their supervoting shares will still give them a combined 51.1% of voting power. The founders take some money off the table without loosening their grip. Standard playbook.

More revealing is where the valuation starts. Wall Street was discussing $1.5 trillion last week, and $2 trillion a few months earlier. Now the range tops out at just $1.02 trillion, and some institutions want it lower. OpenAI raised $11 billion last week at a $1.25 trillion valuation. Anthropic is clearly testing the market with a discount.

Gemini 4 Carbon opened to everyone today. DeepMind positions it as its strongest frontier model, aimed at long-running agent workflows in cybersecurity, complex knowledge work, software engineering, and scientific reasoning. It is already available through the API, AI Studio, and Antigravity.

Remember two prices: Simon Willison found it currently costs $2/million input and $10/million output. DeepMind says those rates will double to $4/$20 on December 1. Many model launches show only a tempting introductory price. This one puts the price hike on the calendar: use it now, pay more later.

Carbon’s starting price also matches Argon’s, but Google says it uses more tokens for deeper exploration and verification. So the price per token has not risen, but the total cost of a task might.

Alongside the roadshow, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 the same day. Fable 5.2 ranked first on Artificial Analysis Intelligence Index v5 with 65 points, while its per-token price fell 20%. But its GDPx cost per task rose from $67.16 to $88.15. A lower unit price does not guarantee a lower bill. Same story as Gemini 4 Carbon.

llm-anthropic 0.31 already supports both models. In Simon Willison’s test, Fable 5.2 on max exhausted its tokens without producing an image. The previous two generations completed the same task. It is a small detail, but the reliability limits on long tasks remain unclear.

On the product side, Claude launched Whiteboards and Sheets in beta. Dashboards became Claude Data and is now generally available. All three are available on every paid plan. Add Claude Meetings, and teams can create plans, organize whiteboards, and build spreadsheets during meetings. That means less switching to another tool afterward.

One developer has already used Claude Code with Fable 5.2 to turn a design image into a playable 12-level side-scrolling action game. It rewrote most of the code, made 175 images and 140 audio clips, then wrote a bot to run all twelve levels in a headless browser and check screenshots.

Nvidia CEO Jensen Huang admitted on November 8 that he had been unfaithful in his marriage, and the board opened an independent investigation. What made the story bigger was that US officials believe Megaspeed executive Alice Huang was providing intelligence to China. Jensen Huang did not name her in his public letter. The US and Singapore were already investigating whether Megaspeed had improperly transferred chips to China.

These issues need separating. Nvidia has long pushed to loosen export restrictions on China, and it has a clear commercial interest in doing so. Its lobbying cannot simply be attributed to the espionage allegations. A researcher pointed to possible intelligence targets: export-control plans, government communications, chip customers, and supply priorities.

Jensen Huang remains CEO during the board investigation. The company says it has no reason to believe national security or company secrets were compromised. But the governance and export-control disputes are now intertwined. Next, we’ll see whether regulators step in.

Notion also disclosed that a security incident at third-party provider Fireworks AI may have affected fewer than 1% of Notion workspaces. Some Notion AI chat transcripts may have been exposed. They had no user or workspace identifiers and no tool-call results, but their text could still contain sensitive credentials. Notion has disconnected Fireworks and disabled Insights.