China

A Reuters review of more than 150 Chinese procurement notices, investment plans, contracts and company disclosures identified computing projects with a combined reported value above 23 billion yuan ($3.2 billion) built or planned around Nvidia's restricted H100, H200 and H800 processors, from a planned 6.2 billion yuan centre in Fujian to clusters already running in Wuxi. The total covers buildings, power and other infrastructure, not chip spending alone, and includes plans that may never be completed. Reuters does not establish how the hardware reached Chinese users, and the projects have not been linked to Megaspeed or the allegations around Jensen Huang.

Will DePue of Unemployment Labs published an analysis on November 12 arguing that six models released since April 2026 by DeepSeek, Moonshot AI, MiniMax and Ant Group were initialized on Nvidia Blackwell-class GPUs, which need a U.S. licence for export to China. The method reads leftover starting values in the published weights and matches them to each chip's processor-core layout; earlier models from the same labs read as Hopper. DePue confirmed two matches on physical Blackwell hardware. He says the evidence shows access to the chips, not where they were located, how access was obtained, or whether export controls were violated.

MiniMax released H3-Image on November 11, an open-weight text-to-image and image-editing model built on its H3 family, with a 33.1 billion-parameter diffusion transformer distilled to six sampling steps. MiniMax says it generates a 4096x4096 image in under 10 seconds and accepts up to 16 reference images; those are its own claims. The weights are on Hugging Face under a community licence whose territory excludes the European Union, the United Kingdom, South Korea and the United States, and which requires written authorization above $1 million in annual revenue or for hosted services.

Chinese AI companies are trying to rent Nvidia-powered computing capacity at data centres outside China, three people familiar with the matter told Reuters on November 12. They fear the allegations around former Nvidia chief executive Jensen Huang could add momentum to the Remote Access Security Act, a House-passed bill that would let the Commerce Department regulate cloud access to restricted chips. A major Chinese cloud provider has discussed locking in overseas capacity and smaller labs are approaching Southeast Asian rental providers, the people said. The companies were not named.

Northwestern Polytechnical University, a Chinese defence-linked university on the US Entity List since 2001, bought two Nvidia H200 chips in October for 874,700 yuan from a domestic supplier, according to procurement records reported by Reuters on November 11. Its cybersecurity school also rented access to servers with H200 chips this year. A Reuters review of 26 records found other Chinese universities buying or renting restricted Nvidia chips through local suppliers. Nvidia, the university and the US Commerce Department did not respond to Reuters.

Treasury Secretary Scott Bessent said on November 10 that next week's Trump-Xi meeting will advance a U.S.-China dialogue on 'superintelligence,' in which the two sides expect to exchange best practices for evaluating and testing advanced models before release and have agreed to a direct channel for alerting each other to serious incidents. Reuters and CNBC report that he made no commitment to limit AI development. No thresholds, agencies or procedures for the channel have been published.

Alibaba opened a preview of Qwen4-Max on November 11, the flagship of its next-generation Qwen family, with text, image and video input, reasoning on by default and a 1M-token context window. API access is through its Token Plan subscription only, with no pay-as-you-go option, and Alibaba says the preview model will be retired after November 25. Alibaba has not published benchmark figures in its model documentation, and Artificial Analysis did not list the model when we checked.

Microsoft has been pressing the Trump administration to halt a planned ban on imports of new Chinese optical transceivers, the components that carry data between servers inside AI data centers, Reuters reported on November 10, citing three people familiar with the situation. Industry groups argue the ban could leave purchased AI chips idle; Chinese firms hold about two-thirds of the market, per Counterpoint. The FCC is weighing a threshold that would spare current models. Microsoft declined to comment.

President Trump told CBS's 60 Minutes, in an interview aired November 8, that the risks of artificial intelligence can be managed and that he doesn't mind guardrails, signalling openness to working with Democrats on safeguards, Reuters reported. He said he did not want rules that put the United States in 'a deep second position' behind China. Reuters notes Trump has previously favoured letting tech companies police themselves; no specific proposal was described.

DeepSeek says Harness, its open-source agent application, had 1.54 million daily active users on its official API in the week ended November 4, up from about 271,000 at the end of September, and that users consume about 92 trillion tokens a day, according to Yicai and Bloomberg. Bloomberg puts that at about 60 million tokens per user per day. The figures are DeepSeek's own and have not been independently checked.

The New York Times reported on November 7 that the Commerce Department approved New Matrix, a 17-month-old Chinese company whose listed Beijing address is an empty office, to assemble Nvidia H200 chips into AI servers, and that corporate records tie two of its owners to Wise Road, a private equity firm entity-listed in 2024. It is the second approved H200 handler the Times found with no visible presence. Commerce's Bureau of Industry and Security says it reviews all applications for national security risks; Nvidia says China strictly limits H200 imports. These are reporting findings, not a government determination.

About 600,000 Nvidia H200 AI chips are stored in Hong Kong waiting to enter mainland China because Beijing has not approved large-scale imports, two people told Reuters. Several big Chinese internet companies have each received about 10,000, and an eight-card H200 server sold for more than 6 million yuan in October, about three times its price a year earlier, one person said.

Tencent released Hy4 on November 4 with image understanding and computer use for the first time, including training on Blender and CAD. It is live in Tencent's apps, on Tencent Cloud and through OpenRouter at the preview's prices. Tencent says the weights and system cards will follow "in the coming days". Its finance and hallucination figures come from evaluations Tencent ran itself.

Revolut says it cut its AI coding inference costs by roughly 90% in under a year by making self-hosted Z.ai GLM models, now GLM-5.3 on 56 Nvidia B200 GPUs, the default for about 2,000 engineers instead of relying mainly on services like Claude Code, The Register reported. Its infrastructure lead said cheap tokens led engineers to hand debugging and ticket work to AI. Commercial models remain for vision and other exceptions.

Shenzhen chipmaker Moffett AI said on November 3 that it has closed a Series C+ round of nearly RMB 1 billion, about five months after raising a similar amount, taking the two rounds to nearly RMB 2 billion. Moffett designs inference chips that skip unnecessary calculations in AI models, an approach known as sparse computing that China's industry ministry has said it will prioritise. The company says third-quarter revenue grew several-fold and its next SP200 card arrives before year-end.

Meituan released LongCat-2.5 on November 2, a natively multimodal model with 1.6 trillion parameters, about 48 billion active per token, built for long agentic coding tasks. Meituan says it trained and runs the model on domestic AI accelerators with under 80 GB of memory each. Its own benchmark table puts it close to Claude Opus 5 on Terminal Bench v2.1 but well behind on the newer v4.0. It is available by chat and API; weights are planned but not yet out.

Kuaishou said on October 30 that its AI video unit Beijing Kling has completed its fundraising, with investors paying about RMB19.8 billion ($2.9 billion). Two investors, Dakang Lanwu and BFI MIDCO, failed to pay within the deadline and dropped out, and others, including units of Kingdee and Shanghai-listed GPU maker MetaX, took over parts of the commitments. Kuaishou keeps control, with 66.68% of the equity and 52.45% of the votes.

Baidu launched Apollo World on November 3, an AI world model that simulates city traffic for training and testing its Apollo Go robotaxis. Baidu says it can generate up to 100,000 km of simulated driving a day, about three years of real-world testing by one car with a safety driver, and produces live camera and lidar data so the driving system reacts as it would on a real road. Other road users in the simulation are modelled as AI agents that respond to the robotaxi.

Beijing-based GenRobot AI has closed a $100 million Series A+ round, disclosed on November 2 after closing in the third quarter. Investors include SAIC Motor, Baidu, Malaysia's Eastworlds, Shoucheng Holdings and humanoid maker Galaxy General Robotics (Galbot). GenRobot sells equipment that records human demonstrations plus software that turns them into robot training data. It says it has raised more than $300 million in total. No valuation was disclosed.

Meta and Shanghai-based Fourier built a humanoid prototype known inside Meta as Max in about three months, The Information reported on October 31, and began work on Max 2 in September. Max reportedly stands about 1.5 metres tall, weighs about 48 kg and runs on Nvidia's Jetson Thor. Meta told the outlet the work is internal research and it has no plans to sell the units.

DeepSeek has published dsh-superagent on GitHub under the MIT licence: a proactive coding agent for one GitHub repository that listens to GitHub and, optionally, Feishu, takes requests from approved people and replies in the original thread, for example with a pull request. DeepSeek calls it a research prototype and not an official product. Bloomberg reported it is aimed at Anthropic's Claude Code.

CrowdStrike says the China-nexus group it tracks as VAPOR PANDA used the ARTEX AI agent framework, mainly with DeepSeek, against more than 200 organizations in at least 14 countries starting October 17. It confirmed post-exploitation at nine. One operator ran several intrusions at once, stepping in only at decision points. CrowdStrike says this is separate from the Storm-0940 activity Microsoft reported.

Updated Nov. 13 (evening): Senate Banking Committee chairman Tim Scott asked former Nvidia CEO Jensen Huang to testify on December 3 and Commerce Secretary Howard Lutnick on December 10, warning of subpoenas if they decline. Treasury Secretary Scott Bessent told CNBC that Treasury, War, USTR and State had backed adding Megaspeed to the entity list for months but that Commerce has the pen, after Reuters reported the White House was discussing Lutnick's future. House chairman Brian Mast says Huang agreed to testify before Thanksgiving. Attorney General Todd Blanche says the Justice Department has not concluded whether Alice Huang was a spy and the FBI is investigating. Earlier: Huang resigned on November 12 with CFO Colette Kress as interim CEO, and Democrats opened a records inquiry. The spy allegations remain unproven; China denies them.

Ant Group's inclusionAI lab published the weights of Ling-3.1-flash-1019 on October 27 under the MIT license, along with the Ling-3.1-flash-base model. The model card describes a mixture-of-experts model with 560 billion total parameters, 25 billion active per token and a 1-million-token context window, post-trained for reasoning and agent tasks. Artificial Analysis now scores it 29 on its Intelligence Index v5.0, a different scale from the 44 on v4.3.2 that the model card cites.