Jensen Huang leaves Nvidia's top job

On 12 November, Nvidia announced that Jensen Huang had resigned as president, chief executive and board member, effective immediately. CFO Colette Kress is now interim CEO and also stays on as CFO. Harvey C. Jones, formerly chief executive of Synopsys, becomes the company's first board chair. The board says it'll consider internal and external candidates for the permanent role.

Al Jazeera reports that he left after reports of affairs, one of them with a woman suspected of links to Chinese intelligence. Nvidia's market value fell below US$4 trillion for the first time since July 2025. Huang has agreed to appear before the House Foreign Affairs Committee, and Democrats in both chambers are investigating plans to approve AI chip sales to Chinese firms. Third-quarter results are still due on 17 November, US time.

For builders: Nvidia's leadership says customer commitments and the product roadmap won't change. Teams planning GPU purchases should watch the earnings call and any new export-control moves, since either could affect supply and pricing.

Anthropic sets terms for its Nasdaq listing

CNBC reported on 16 November that Anthropic is offering 825 million Class A shares at US$70 to US$78 each. At that range it would raise more than US$64 billion at a valuation of about US$1.405 trillion. The shares are due to be priced on the evening of 18 November and to start trading on Nasdaq as ANT on 19 November. Up to 20% of the offering is reserved for retail investors.

For builders: As a listed company, Anthropic will have to report results every quarter. That gives teams that depend on Claude a much clearer view of the vendor's finances and spending.

OpenAI gives ChatGPT agents a wallet

On 13 November, OpenAI launched ChatGPT Wallet, which lets its agents complete purchases for users. For now it's limited to US ChatGPT Pro subscribers aged 18 or over who have eligible US-issued cards. American Express, Mastercard and Visa provide a single-use token with its own card number, security code and expiry date, so neither the agent nor the merchant sees the real card. Each purchase has to be approved with Face ID or a fingerprint. OpenAI plans to add pre-approval within limits on spending, product category or time. It hasn't disclosed merchant fees.

For builders: Merchants should expect checkouts started by agents and paid with network tokens. Fraud rules may need tuning so legitimate agent orders don't get declined.

The Fed ranks AI as the top financial risk

The Federal Reserve released its Financial Stability Report on 13 November. In its survey of market contacts, 87% named AI as a salient risk, up from 50% in spring 2026. The next most cited concern was a broad fall in risky assets, at 52%. Respondents pointed to stretched equity valuations, cyber threats from state actors or frontier models, and complex AI-related financing.

On 16 November in Milan, Governor Michael Barr warned that AI agents reacting to yield gaps or signs of stress could move deposits fast enough for a bank run to play out in minutes. He cited the report's estimate that on-balance-sheet debt at the seven largest AI firms will pass US$800 billion by year-end. He also said banks already see agents trying to act in retail accounts where consent can't be verified.

For builders: Barr called for strict authentication, explicit consent for each action and technical limits on what agents can do. Teams building agents that touch bank accounts should plan for all three.

Google adds Gemini 4 Hydrogen

On 12 November, Google released Gemini 4 Hydrogen, the fast model in its Gemini 4 family alongside Argon and Carbon. At its default setting it scores 51 on the Artificial Analysis Intelligence Index, one point above Argon. It scores 57% on Terminal-Bench 4.1, against 18.2% for Gemini 3.8 Flash. Google says it finishes tasks 2.5 times faster. It's now the default model for AI Mode in Search and is available in the Gemini app.

For builders: API pricing is US$1.50 per million input tokens and US$7.50 per million output tokens, with a 1 million token input window. In Antigravity, Carbon can hand routine steps to Hydrogen to use fewer tokens.

Australia plans a frontier AI regulator

On 10 November, Prime Minister Anthony Albanese and three ministers said frontier AI companies operating in Australia will face mandatory safety and security obligations under proposed AI Standards legislation. Firms will need to:

  • keep security controls in place through development and deployment
  • manage serious risks from their models
  • allow independent evaluation
  • report incidents promptly

A new regulator, working with the Australian AI Safety Institute, would enforce the rules and could impose significant penalties. No penalty amounts or start date have been announced.

For builders: The obligations apply to frontier developers. Teams deploying those models in Australia or New Zealand should watch for incident-reporting and evaluation terms appearing in vendor contracts.

Meituan opens LongCat-2.5 weights

Meituan published the LongCat-2.5 weights on Hugging Face on 9 November under the MIT licence. Its model card lists 1.6 trillion total parameters in a mixture-of-experts design, with about 48 billion active per token. In Meituan's own comparison, it scored 91.9 on BrowseComp, the highest of the four models listed, but trailed on several coding benchmarks. The hosted API costs US$0.30 per million input tokens and US$1.20 per million output tokens at a limited-time rate.

For builders: The MIT licence allows commercial use and modification, and Meituan has quantised the expert weights to four bits for domestic accelerators. Running a model this size yourself still takes a large cluster.

Sources

Talk about it with people building with AI

Our Auckland meetups are open to everyone. Members get a ForgeBox and the members' projects area for NZ$599 a year.

Become a member

Get the daily AI news by RSS