AI news, 21 November 2026

Anthropic's record IPO and Nvidia's China halt lead the week

By DataForge team21 November 20264 min read

Anthropic raised US$81.3 billion in the largest IPO on record, and Nvidia has stopped licensed H200 shipments to China.

Anthropic lists on Nasdaq after the largest IPO on record

Anthropic priced its shares at US$83 on 18 November. That was above the US$70 to US$78 range it had marketed to investors. It sold 980 million shares, more than the 825 million it first planned, and raised US$81.3 billion. That beats SpaceX's float in June, which raised about US$75 billion. About 956 million of the shares came from the company and around 24 million from its seven co-founders. A fifth of the offering was held back for retail investors.

Trading began on 19 November under the ticker ANT. The stock opened at US$111.50, 34% above the offer price, and closed at US$97.17. That's a first-day gain of 17%, and CNN put the company's value at US$1.76 trillion at the close. The broader market fell sharply on the same day.

What it means for NZ organisations: Claude's developer now has to publish audited results every quarter. If you buy Claude, your procurement and risk teams will have much better information for vendor reviews.

Nvidia stops licensed H200 sales to China

On Nvidia's earnings call on 17 November (US time), interim CEO Colette Kress said the company has stopped shipping licensed Hopper data centre products to China. She didn't give a reason. Under a US arrangement that gave Washington 25% of the proceeds, Nvidia sold US$2.4 billion of H200s to Chinese buyers last quarter. The call came after Jensen Huang resigned as chief executive. Board chair Harvey Jones said the board doesn't intend to bring him back.

What it means for NZ organisations: US export policy keeps moving chip supply around Asia. If you rent GPU capacity from regional cloud providers, ask them where the hardware sits and whether new US licensing rules could affect it.

Microsoft releases seven in-house AI models at Ignite

Microsoft AI announced seven models on 17 November. The main one is MAI Code 2, which is now in GitHub Copilot. Microsoft says it delivers 98% of GPT-6.1 Sol's coding performance at about 40% of the price, and 96% of Claude Opus 5.5's at about 5%, averaged over four benchmarks. A smaller model, Code 2 Flash, is in Copilot and in a limited preview in Microsoft Foundry. The other releases:

  • MAI Image 2.7 and MAI Image 2.7 Flash, with Flash costing about US$10 per 1,000 images
  • MAI Realtime 1, a voice model from US$0.01 per audio minute
  • MAI Cyber 2 Flash
  • Dragon Copilot Transcribe, for clinical notes

All the performance and price figures come from Microsoft.

What it means for NZ organisations: Teams on GitHub Copilot can try Code 2 now. Test it on your own repositories before making it the default, because vendor benchmarks often don't match real codebases.

European Commission confirms AI-driven attack on staff systems

On 19 November, Commission spokesperson Thomas Regnier confirmed an attack he called an "AI-generated cyber campaign". It was first detected on 22 October. Personnel systems were out for more than 10 days, including payroll, leave requests, medical reimbursements and translation services. Investigators haven't found evidence of a broad breach of staff data, and the Commission isn't aware of any ransom demand. It's the Commission's third disclosed cyber incident this year.

What it means for NZ organisations: Automated attack tools cut the time between a flaw being found and it being exploited. Make sure your HR and payroll platforms are patched and monitored, and that you've tested manual fallbacks.

Gemini for Mac can now operate desktop apps

Google started rolling out computer use in the Gemini Mac app on 19 November. When a prompt includes "@Computer use", the agent clicks, types, moves through menus and fills in forms in a separate background window. Users choose which apps it can reach and approve sensitive actions before they happen. For now it's limited to US users working in English who are 18 or older and pay for AI Plus, Pro or Ultra. It needs an Apple silicon Mac running macOS Sequoia 15.0 or later.

What it means for NZ organisations: It's US-only for now, but desktop agents from several vendors are on their way to staff laptops. It's worth deciding now which apps and data they'll be allowed to touch.

Google opens its Genie 4 world model to developers

Since 19 November, developers can use Genie 4 through the Gemini API's Interactions API and Google AI Studio. It builds continuous 720p worlds at 24 frames per second, with matching audio, from a text prompt, an image or both. Users can steer the scene with keyboard controls and live prompts. Every output carries an invisible SynthID watermark.

What it means for NZ organisations: It could be useful for training simulations, tourism previews and game prototypes. Check Google's API documentation for pricing and session limits before you build anything.

OpenAI puts GPT-Live-1.1 in its API

OpenAI released its GPT-Live-1.1 voice model to developers on 19 November. It can listen and speak at the same time and copes with pauses and interruptions. OpenAI says it handles multiple languages, follows instructions and reproduces voices better than GPT-Live-1. A new opt-in standby rate of US$0.02 a minute for idle WebRTC sessions also works with the older model. GPT-Live-1.1 has been the default ChatGPT voice for Plus and higher tiers since 15 October.

What it means for NZ organisations: Lower idle costs make always-on voice agents for bookings and contact centres cheaper to run. Before you deploy one, test how it handles New Zealand accents and te reo Māori place names.

Want to know which of these changes matter for your organisation? Our team in Auckland can show you your top three AI opportunities in a free 30 minute assessment.

Get my free assessment How we help
Get the daily AI news by RSS