Anthropic closes its first trading day 17% above the IPO price
Anthropic priced its IPO on 18 November 2026 at $83 a share, higher than the $70 to $78 range it had marketed. The deal was for 980 million Class A shares. Anthropic sold about 956.4 million of them and existing holders sold the other 23.6 million, so the offering was worth about $81.34 billion at the offer price. Goldman Sachs and Morgan Stanley ran the sale, and the underwriters can buy up to 147 million more shares within 30 days.
The stock started trading on 19 November under the ticker ANT. Its first trade on Nasdaq was at $111.50 and it closed the day at $97.17. Separately, a $26 billion private placement at the same $83 price includes Google, Nvidia, AMD and Microsoft among its buyers. The prospectus shows $17 billion of revenue for the quarter to 30 September, up from $1.4 billion a year earlier.
For builders: Teams that depend on Claude can now read quarterly public filings on their supplier's revenue, spending and risks. Those filings sit well next to your own benchmark results when it's time for a vendor review.
OpenAI's models cut area and power in AMD and in-house chip designs
OpenAI published a research post on 18 November about using its models on hardware design. AMD worked with OpenAI on its flagship GPU IP, and the models rewrote SystemVerilog code to gain up to 14% across power, performance and area, measured against baselines that were already heavily tuned. Some of those changes went into an AMD design that has been sent for manufacturing. AMD also confirmed single-digit percentage gains in key blocks of a coming flagship CPU.
JalapeƱo is OpenAI's first custom inference chip. On that design, the team estimates its models' suggestions cut standard-cell area by about 13% in the vector unit and about 10% in the matrix unit, and the changes held up when the silicon was brought up. In one case, a redesigned multiplier took roughly 15% less area and finished in one clock cycle where the old design took two. OpenAI is also working with Synopsys to train GPT-Synopsys, a model built to run that company's EDA tools.
For builders: Frontier models can now improve hardware description code too, as long as every rewrite goes through strict verification. Chip teams can try it on their own RTL and keep simulation checks as a hard gate.
Gemini for Mac can now operate desktop apps
On 19 November, Google added computer use to the Gemini app for Mac. You describe a multi-step task and Gemini works through local files, desktop apps and websites in its own background window. A live preview shows in the chat while your main screen and cursor stay free. You choose which apps it can use, you can pause it at any time, and you approve key actions before it takes them.
It's rolling out in English in the US to adults with personal Google accounts and to Google AI Plus, Pro and Ultra subscribers. It needs an Apple Silicon Mac running macOS Sequoia 15.0 or later.
For builders: Desktop agents are now reaching everyday users. If your Mac app has unlabelled controls or deep menus, test how Gemini copes with them.
GPT-Live-1.1 is available in the OpenAI API
GPT-Live-1.1 became available to developers on 20 November as gpt-live-1.1. It's a full-duplex voice model, so it can listen while it talks. OpenAI's documentation says it transcribes more accurately than GPT-Live-1 and keeps a steadier accent and voice across languages. Developers can describe in plain language when it should hand work to a backend model or tool. The same voice model has run ChatGPT Plus and Pro voice since 15 October.
Sessions are billed by the minute: $0.05 while active and $0.02 on standby. Standby starts after an idle timeout of 5 to 60 seconds, set by the developer, and keeps the WebRTC call open. Backend model and tool use costs extra. The model works only through the Live sessions endpoint.
For builders: Standby pricing makes long, mostly quiet voice sessions cheaper, which helps with support lines and in-car assistants. Plan for the separate endpoint, because existing Realtime code won't work with it unchanged.
US brings back chip licences and adds five countries
The US Commerce Department's Bureau of Industry and Security published an interim final rule that took effect on 19 November. It brings back licence requirements for advanced computing chips and related items going to China, Macau and other restricted country groups. It also adds new requirements for Indonesia, Malaysia, Singapore, Thailand and Brazil, with an exclusion for US-headquartered customers that comes with monthly reporting. A new control covers chips for any data centre that gives significant compute to companies headquartered in China or Macau. The rule withdraws the January 2025 AI Diffusion Rule, including its controls on model weights. Comments are due by 19 January 2027.
For builders: If you rent GPUs from providers in Southeast Asia or Brazil, expect more customer checks and possibly slower capacity growth.
About 879,000 AI chips were cleared for China in six months
On 19 November, the House Foreign Affairs Committee released licence data. It shows that between December 2025 and May 2026, the Commerce Department approved about 879,000 Nvidia H200 and AMD MI325X chips for Chinese buyers, out of 3.4 million requested. Tencent was approved for 150,000, Alibaba for 75,000, Baidu for 43,000 and BYD for 4,000. It isn't known how many of the approved chips were shipped. Nvidia says it has since stopped shipping H200s to China.
For builders: Chip access for Chinese labs can change quickly with policy, so check how far your plans depend on any single supply route.
Sources
- Anthropic prices initial public offering
- OpenAI: Advancing chip design with our models
- Google: Computer use in the Gemini app for Mac
- OpenAI API docs: GPT-Live-1.1
- OpenAI Developer Community: GPT-Live-1.1 has arrived in the API
- Morrison Foerster: BIS issues interim final rule
- TechRepublic: Anthropic IPO opens at $111.50 in record Nasdaq debut
- Al Jazeera: Anthropic starts trading in the US after blockbuster IPO
- Fox Business: Anthropic begins trading on Nasdaq
- NPR: Investors scramble for a piece of AI giant Anthropic in blockbuster IPO
- The Standard: US signed off on sales of nearly 900,000 AI chips to China in six months
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.