新云社 update: Citigroup CEO said last week’s repricing of AI-related assets was “a bit overdone” and that demand for computing power has not disappeared. The comment pushes back on the market’s recent selloff in AI-linked names, arguing the underlying need for compute capacity remains intact.
新云社 reports: Suiyuan Technology rose nearly 10% on Monday as Chinese A-share compute chip stocks extended a rebound, with the company’s shares up more than 50% over five trading sessions and its total market capitalization returning above 200 billion yuan. Other chip-related names, including PinGao, Dashengda, VeriSilicon, Hesheng New Materials, Dongxin and Wantong Development, also advanced.
新云社 update: Ant Lingbo Technology open-sourced LingBot-Touch, a tactile foundation model for robots, on November 10. Pre-trained on nearly 5,000 hours of real tactile data, the model covers magnetic, visual and piezoresistive sensing and converts different signals into a unified tactile representation for use across dexterous hands and motion models, filling in the touch perception needed for fine manipulation.
新云社 update: Anthropic has notified bankers that the IPO filing documents previously sent to them contained errors, and the company will provide updated data to the banks. The AI developer did not disclose the nature of the errors or which figures are affected. The correction comes as Anthropic prepares for a public listing.
新云社 update: Anthropic plans to raise up to $69.3 billion in its initial public offering by issuing 866,666,664 shares of Class A common stock, with a price range of $70 to $80 per share. The filing sets the offering at the top end near $69.3 billion. The IPO would rank among the largest US listings.
新云社 reports: CSC Financial’s 2027 computer-sector strategy says China’s AI investment theme will shift from spending expectations to ROI delivery, with domestic chip adoption and application commercialization as the core variables. The brokerage flagged four directions to watch: domestic compute, AI applications and models, Xinchuang, and future industries. It said domestic AI chip penetration in training and inference is rising as smart-computing center construction accelerates.
新云社 update: Anthropic released its most powerful flagship model, Fable 5.2, which outperformed GPT-6 Astra across multiple benchmarks, while both Mythos 5.2 and the new model were made faster and cheaper. The launch comes as Anthropic’s roadshow is set to begin as early as Tuesday, with IPO pricing facing a test after the AI selloff. Nvidia fell over 4% even as the Nasdaq rose nearly 3%.
新云社 reports: Several listed Chinese PCB manufacturers say AI-related orders are full, with some flagging tight delivery schedules, as Ke Xiang shares announced capacity expansion for AI computing PCBs. A senior executive at a leading PCB listed company said orders from AI servers, automotive and communications segments have been strong this year, while consumer electronics demand remains steady. Electronic cloth, a key upstream material for copper-clad laminates, has risen sharply and repeatedly this year on stronger downstream PCB demand.
新云社 reports: Goldman Sachs said the Japanese yen remains the “best hedge” against the risk of a further selloff in AI-related assets. The call frames the yen as a defensive instrument for investors exposed to a potential deepening of the AI trade unwind.
新云社 update: SpaceX plans to build, own and operate a gas-fired power plant and supercomputing facility at its Starbase site in Texas, with an initial investment of $24 billion. The Starbase city commission will consider a tax abatement and development agreement at its December 9 meeting. The move points to vertical integration of SpaceX’s natural gas value chain.
新云社 update: OpenAI said its control room, operating alongside a misalignment monitoring system, can stop model output within 35 seconds of a decision to intervene and halt or pause a task. The company disclosed the 35-second window as part of its oversight setup for detecting and acting on misaligned model behavior.
新云社 update: Anthropic said its Claude model outperformed the best listed solutions on 72 problems from MIPLIB, a standard benchmark library for mixed-integer programming. The company also reported 45 optimality proofs and said it has submitted its solutions to MIPLIB, whose public records have not yet confirmed the tally.
新云社 reports: Anthropic plans to begin trading next week on Thursday and Friday. The AI company has not disclosed further details on the listing venue, pricing or shares to be offered. No official statement from Anthropic was included in the reports.
新云社 update: Senator Ed Markey is demanding that Google provide schools with an AI-free version of its Search product, after the company said schools can enable Search and Gemini separately but that AI Overviews remain built into Search. The push highlights growing scrutiny of how AI features are bundled into tools used in classrooms. Google’s current controls let schools toggle Gemini independently, but not remove AI Overviews from Search results.
新云社 reports: Nvidia shares fell 4.3% on the day, closing at their lowest level in seven months. The decline marks the chipmaker’s weakest close since the spring, with the stock extending losses as investors weighed pressure on the AI trade. Nvidia remains a key bellwether for crypto-linked equities and AI infrastructure demand.
新云社 update: Fireworks AI disclosed that a third-party cloud services provider hosting its research and development machines experienced a security incident on October 23. The company did not specify the scope of the breach or which data or systems were affected. Fireworks AI is an AI inference platform, and the disclosure concerns its internal R&D infrastructure rather than a crypto-specific system.
新云社 update: US Senator Ron Wyden said he will produce a comprehensive legislative draft addressing taxation of data centers. The announcement signals a push in the Senate toward a dedicated tax framework for data center operations, though no bill text, timeline or specific tax provisions were disclosed. The scope and details of the proposed legislation remain unclear.
新云社 update: Fireworks disclosed a security breach that exposed billing data, datasets and incomplete model weights, the company said. It disabled self-service dataset transfers for batch inference and dataset-based training, and reported no detected impact to its Inference service or Training SDK. The company did not say how many customers were affected or when the breach occurred.
新云社 update: Google has introduced two new AI chips, the TPU 8tx and the dual-chip TPU 8i Max, aimed at Nvidia’s data-center dominance. The TPU 8tx is built to protect sensitive AI workloads, while the TPU 8i Max targets memory-hungry AI agents. The launch positions Google’s in-house silicon against Nvidia’s GPU lineup for AI training and inference.
新云社 reports: Nvidia CEO Jensen Huang plans to attend Semafor’s “Silicon Valley and the World” event, people familiar with the matter said. The appearance would put the head of the AI chipmaker at a public forum hosted by the media outlet. No date or further details of the planned participation were provided.