AOYii AI Daily - Nov 19, 2026

Daily Highlights

  • Anthropic prices IPO at $83, largest IPO in history
  • Jensen Huang steps down; Kress becomes Nvidia interim CEO
  • Nvidia halts H200 sales to China, risking billions in revenue
  • Google ships Gemini 4 Neon with 1M context and 4 thinking levels
  • Microsoft's agent-framework PyPI package poisoned for 55 minutes

TODAY'S HEADLINES

[Anthropic prices IPO at $83 per share, set to become the largest IPO in history] On the evening of Nov 18 (ET), Anthropic priced its IPO at $83 per share, roughly 6% above the top of its $70-$78 guided range, with shares set to begin trading on Nasdaq on Nov 19 under the ticker ANT. Institutional demand was about 6x oversubscribed; the offering raises at least $68 billion, and with a $26 billion private placement the total exceeds $94 billion, surpassing SpaceX's record $86.2 billion IPO from June. (Source: Bloomberg / CNN)

[Jensen Huang steps down as Nvidia CEO; interim chief halts H200 sales to China] Nvidia confirmed Jensen Huang has stepped down as CEO, with CFO Colette Kress taking over as interim CEO. On the Q3 earnings call she announced the company will stop selling H200 chips in China, unwinding the arrangement Huang brokered with the Trump administration that let the US collect a 25% tariff on each sale. Jefferies estimates the policy could cost Nvidia $44-62 billion in annual data center revenue, sending ripples through China's computing supply chain and lifting domestic chip-replacement stocks. (Source: AI Weekly / Caixin)

NEW PRODUCTS & TECH

[Google ships Gemini 4 Neon: 1M context with four thinking levels] Google released its frontier model Gemini 4 Neon, priced at $3 per million input tokens and $15 per million output tokens, with a 1-million-token context window and four configurable thinking levels. The model ranks #3 (score 61) on the Artificial Analysis Intelligence Index, #2 on the Vals Index and #2 on Agent Arena, escalating a US-China price-capability race among top models. (Source: AI Weekly)

[Tencent Q3: Hunyuan Hy4 tops Agent Arena among Chinese models, cloud revenue up 35%] Tencent reported Q3 results on Nov 18. Its latest Hunyuan large model Hy4 ranked first among Chinese models on Agent Arena at launch; the AI assistant WorkBuddy surpassed 10 million daily active users with over 100 trillion tokens consumed daily, the highest of any LLM product in China; AI demand drove cloud revenue up 35% year over year. (Source: Sina Tech / Tencent earnings)

IPO & FUNDINGS

[Goldman raises 2027 global AI project financing forecast to $545 billion] Goldman Sachs' credit strategy team raised its 2027 global AI project financing forecast by 82%, from $300 billion to $545 billion, including $320 billion for GPU and compute financing and $225 billion for data center construction. Chip financing would be nearly 7x the level year-to-date in 2026; Goldman also flagged rising financing costs and reliance on vendor guarantees for some chip deals. (Source: Cailian Press)

[OpenAI trims funding target by $10 billion, valuation lands around $1.25 trillion] OpenAI lowered its funding target by $10 billion in its ongoing round, with the recently completed portion implying a valuation around $1.25 trillion, below its earlier $1.4 trillion ambition, as a public-market AI selloff weighs on the deal and Anthropic's listing steals the spotlight. (Source: Sina Tech)

OPEN SOURCE & RESEARCH

[Microsoft's agent-framework PyPI package hijacked for 55 minutes to steal cloud credentials] Attackers published malicious versions 1.26.1 and 1.26.2 of Microsoft's official agent-framework Python package to PyPI on Nov 17, using .pth startup hooks to silently harvest Azure and GitHub cloud credentials and CI/CD secrets on every Python startup. The versions stayed live about 55 minutes before removal (the package sees roughly 41,000 weekly downloads). PyPI has quarantined the malicious releases and 1.26.0 remains the last clean version; OX Security urged affected teams to rotate cloud and CI credentials immediately. (Source: The Hacker News / OX Security)

[China Telecom unveils Xing4: agentic LLMs trained entirely on Ascend chips] China Telecom posted a technical report on Nov 18 detailing Xing4, an agentic LLM series trained entirely on 8,192 Ascend 910C NPUs using MindSpore, with no GPUs or CUDA. The series also open-sources a 29B-total/4B-active sparse MoE model supporting 256K-1M context, offering the first complete public example of frontier-scale training on a full domestic stack. (Source: AI Weekly / technical report)

COMPANY & PEOPLE

[Masayoshi Son calls missing out on Anthropic a mistake] SoftBank founder Masayoshi Son said he regrets not investing in Anthropic before its high-profile IPO, explaining he passed because SoftBank was already committing capital to rival OpenAI. His remarks came as OpenAI cut its funding target and Anthropic priced higher, read by many as reflection on SoftBank's AI portfolio positioning. (Source: Sina Tech / Bloomberg)

[Microsoft and Intel expand Azure cloud and AI supply partnership] Microsoft and Intel announced an expanded cooperation agreement on Nov 19 covering Azure cloud and AI supply, confirming multi-year Intel Xeon server CPU supply capacity including instances based on next-generation server CPUs, supporting Azure's AI inference and general compute expansion. (Source: Cailian Press)

POLICIES & DATA

[China and Singapore hold AI governance exchange meeting in Beijing] China's Cyberspace Administration and Singapore's Infocomm Media Development Authority held an AI governance exchange meeting in Beijing on Nov 18, discussing governance policies and standards, safety and risk, and international cooperation, and reaching consensus on deepening AI governance collaboration. (Source: Cailian Press)


Powered by AOYII.com

Related News