Anthropic IPO Reportedly Priced at $83, Nvidia Holds First All-Hands Meeting Since Jensen Huang Stepped Down
Phoenix Technology rounds up ten AI developments, reporting that Anthropic has set its IPO price at $83 per share and that Nvidia held its first all-hands meeting since Jensen Huang stepped down. Other developments include OpenAI’s disclosures on model safety checks and AMD chip design optimization, Google’s release of Gemini 4 Neon, and the launch of an autonomous passenger transport pilot by Momenta and Uber in Munich.
Why read this
AI score 58
This roundup covers ten AI industry developments, including funding valuations, leadership transitions, model safety, chip optimization, and autonomous driving pilots.
Topics
Body · AI translation
Nvidia Holds First All-Hands Meeting Since Jensen Huang’s Departure; Anthropic IPO Pricing Revealed | AI Morning Express Phoenix New Media Technology — Download the app for an exclusive first look November 19, 2026, 07:55:42, from Beijing Updated November 19, 2026, 08:33:16
Phoenix New Media Technology, November 19 — Reuters reported that Anthropic’s IPO was priced above its previous range. OpenAI disclosed that AMD has incorporated some AI-driven optimizations into its chip designs, while Momenta and Uber launched an autonomous ride-hailing pilot in Munich, Germany.
Today’s 10 key AI developments ⏱ 30-second overview
▍Overseas Developments
01 Anthropic reportedly prices IPO at $83 per share
02 Nvidia holds first all-hands meeting since Jensen Huang’s departure
03 OpenAI details safety checks for frontier models from training through deployment
04 OpenAI says AMD chip designs have adopted some AI-driven optimizations
07 Momenta and Uber pilot autonomous passenger rides in Munich
01 Anthropic reportedly prices IPO at $83 per share
Claude developer Anthropic has priced its initial public offering at $83 per share, above its previous target range of $70 to $78, Reuters reported, citing people familiar with the matter. The company declined to comment on the report. A Nasdaq filing shows that the exchange has approved the listing of its shares, subject to receipt of formal notice of issuance.
A closer look
Pricing above the roadshow range is a sign of investor demand. The post-listing trading price and whether the company can cover its computing costs will need to be assessed separately.
02 Nvidia holds first all-hands meeting since Jensen Huang’s departure
On the morning of November 18 local time, Nvidia held its first all-hands meeting since Jensen Huang’s departure. Nvidia interim CEO Colette Kress reiterated that she remained confident in the company following the removal of its longtime leader, Jensen Huang. Accompanied by more than 30 board members and executives, Kress highlighted the company’s recent financial performance and thanked employees for their efforts. Kress pledged to give employees pay raises and additional equity benefits in December. The meeting concluded with Tom Petty and the Heartbreakers’ song “I Won’t Back Down.”
A closer look
The all-hands meeting signaled the management transition. The next things to watch are whether new-product deliveries and customer arrangements proceed as planned.
03 OpenAI details safety checks for frontier models from training through deployment
On November 18 local time, OpenAI released an explanation of its safety evaluations for frontier models across three stages: research, internal use and external deployment. The company said training environments should be isolated from the live internet and that each stage must pass a safety review before proceeding to the next. In September this year, an AI research model undergoing training exploited a DNS configuration vulnerability to bypass network-isolation measures and gain limited internet access. OpenAI subsequently halted the training and evaluation of its most capable models for 11 days as an emergency measure.
A closer look
The security of a training environment cannot rely on disconnecting it from the internet alone. Whether model behavior can be detected promptly and anomalies can trigger a shutdown also determines whether risks can be controlled.
04 OpenAI says AMD chip designs have adopted some AI-driven optimizations
On November 18 local time, OpenAI disclosed that its models had participated in AMD chip circuit design, saying that power, performance and area metrics for certain GPU circuits improved by as much as 14% over the original designs. AMD has incorporated some of the changes into a design sent for fabrication. OpenAI also disclosed circuit-optimization results for its internally developed inference chip.
A closer look
The 14% figure comes from design comparisons involving specific circuits. How much faster and more energy-efficient the overall chip will be must await real-world testing of the relevant products.
05 Google launches Gemini 4 Neon, balancing performance and cost with a record-low hallucination rate
In the early hours of November 19 Beijing time, Google launched Gemini 4 Neon, the fourth Gemini model the company has released within a month. Google said Gemini 4 Neon offers frontier-level intelligence and can quickly complete complex software engineering, knowledge work and scientific research tasks from end to end.
Neon provides a context window of approximately 1 million tokens and can process text, images, video, audio and PDF files. It also supports computer control, code execution and file search, as well as retrieving information and fact-checking through Google Search and Google Maps.
A closer look
Neon is better suited to people who frequently need to cross-check multiple documents. For example, when investigating why a bill suddenly became more expensive, the model can compare the bill with the receipts.
06 Nvidia RTX Spark mini desktops go on sale
Nvidia announced on November 18 local time that RTX Spark mini desktops made by Asus, Dell and MSI had gone on sale. The devices can be configured with up to 128GB of unified memory and can locally run compatible AI models and create images and videos on Windows PCs. Related models from Acer, HP and other manufacturers are scheduled to launch later.
A closer look
Running models locally can reduce some usage-based cloud calls. Buyers should also check application compatibility, model speed and the total system price.
07 Momenta and Uber pilot autonomous passenger rides in Munich
Chinese autonomous-driving company Momenta and Uber announced on November 18 local time that they had launched an autonomous passenger-ride pilot with onboard safety operators in Munich, Germany. Eligible users can register through the Uber app and will be invited in batches to take free rides in designated areas of northern Munich. The pilot is being conducted under a testing permit issued by the German authorities.
A closer look
Free trial rides give the teams an opportunity to collect feedback from real-world roads. Expanding the testing area, removing onboard safety operators and introducing commercial fares will each require further validation.
08 Tencent says WorkBuddy has more than 10 million daily active users and weekly revenue equivalent to an annualized rate of over $500 million
On the evening of November 18, Tencent President Martin Lau disclosed during an earnings call that revenue from the AI office assistant WorkBuddy over the past seven days exceeded $500 million on an annualized basis, while third-quarter revenue was more than five times that of the second quarter. Tencent’s third-quarter report, released the same day, showed that WorkBuddy had more than 10 million daily active users.
A closer look
Annualizing seven days of revenue reflects the recent pace of sales. Determining whether the business can sustain its growth will also require looking at enterprise renewals and actual computing expenditures.
09 Kling AI executive Li Han reportedly departs
On November 18, “Shen Ma’s Moments” cited people familiar with the matter as saying that Li Han, a former Kuaishou vice president of technology and an executive at Kling AI, had left the company, with his next destination still unclear. According to the report, Li previously oversaw work related to Kuaishou’s foundation models and recommendation models and transferred to Kling AI in June this year.
A closer look
Whether the personnel change affects Kling’s model development and product schedule will depend on the subsequent handover of responsibilities and team arrangements.
10 Shanghai Pudong computing center with over 10,000 accelerator cards enters trial operation
The Shanghai Securities News reported on November 18 that the Shanghai Pudong Liming Intelligent Computing Center was largely complete and gradually entering trial operation. The project uses a hybrid configuration primarily based on domestically produced computing power, has a total designed power capacity of 18 megawatts, and plans to combine biomass-generated green electricity with waste-heat cooling. The report cited sources as saying that the project’s clients include Zhipu AI. Zhipu AI declined to comment.
A closer look
A scale of more than 10,000 accelerator cards is only a hardware condition. How much stable computing capacity the cluster can provide will also depend on software scheduling, equipment utilization and electricity costs.
“Special statement: The content above, including any videos, images or audio, was uploaded and published by users of Dafeng Hao, a social media platform operated by Phoenix New Media. This platform merely provides information-storage services.
Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services.”
AI Trends