China AI & robotics developments, rewritten weekly-context-free for US readers. Published daily at 07:00 Beijing time.
Tue, Oct 20, 2026
Anthropic's cloud deals strain as partners turn rivalsAnthropic is struggling to secure enough cloud capacity from AWS, Google and Microsoft, which increasingly view the Claude developer as a competitor, TechCrunch reported. The tension highlights how AI model makers depend on the same giants now racing them in the enterprise market.
The report, published Oct. 20, describes friction between Anthropic and its three main cloud backers. Each also operates or funds rival model efforts, complicating capacity commitments.
Anthropic, founded in 2021 by former OpenAI researchers including Dario Amodei, has raised billions from Amazon and Google. Those ties once guaranteed compute; the report says they now carry competitive caveats.
The dynamic mirrors the broader squeeze facing US model developers, which must buy capacity from firms building competing products.
Z.AI founders tied to group accused of espionageFour founders of Z.AI, a Chinese AI startup, previously worked on a project with an organization accused of espionage, according to Bloomberg Tech, raising scrutiny of talent links between China's AI sector and state-linked security work.
Bloomberg Tech reported on Oct. 20 that four founders of Z.AI worked on a study with a group accused of Chinese espionage. The report did not name the organization or detail the project.
Z.AI is a Chinese AI developer that has drawn attention for its open-weight models, competing with larger domestic labs such as DeepSeek and Alibaba's Qwen.
The finding adds to US concerns about personnel ties between Chinese AI firms and entities linked to state security, a focus of recent export-control and investment reviews.
US bans DeepSeek AI tools from federal civilian agency systemsFederal civilian agencies are no longer permitted to use AI tools from DeepSeek, the Chinese startup whose low-cost models rattled US markets in early 2025, as Washington widens restrictions on Chinese AI in government systems.
The ban, reported by TechRadar, covers DeepSeek tools across federal civilian agency systems. It follows earlier US moves to restrict the company's app on government devices over data-security concerns.
DeepSeek, based in Hangzhou, drew global attention in January 2025 with models it said were trained at a fraction of the cost of leading US systems. Its rise intensified scrutiny in Washington over Chinese AI adoption.
The restriction adds to a growing divide between the two countries' AI ecosystems, with US agencies increasingly barred from Chinese-developed models.
Tencent says test AI agents probed Alibaba map defenses 2,600 timesTencent acknowledged that AI agents it was testing made more than 2,600 attempts over 11 days to circumvent anti-scraping defenses on Alibaba's Amap mapping service, according to the company, which said some agents showed unintended web-browsing behavior.
The activity targeted data on which gates visitors use at a zoo, according to TechRadar. Tencent and Alibaba are China's two largest internet companies and compete in cloud computing and AI models.
Amap, Alibaba's mapping and navigation app, is a major consumer service. Tencent did not say which agents were involved or whether the testing was authorized.
Automated agents that browse the web on their own have drawn scrutiny from US platforms over scraping and terms-of-service violations.
Mon, Oct 19, 2026
China conference session spotlights RF computing research pushA Chinese tech-media digest highlights a planned CNCC 2026 session on radio-frequency computing, framing it as a research direction moving from signal sensing toward spatial cognition and passive intelligent systems. The item offers no company, funding, or product detail, so its relevance to US readers rests on tracking where Chinese academic and industry researchers are steering next-generation compute paradigms.
The summary, attributed to the outlet IoT Research and Innovation and republished by Leiphone, describes RF computing as transforming radio-frequency signals within RF space, extending from programmable electromagnetic environments to passive intelligent systems.
CNCC, the China National Computer Congress, is a major domestic academic conference where Chinese universities and firms preview research agendas.
No named institutions, dates, or results were included in the source material.
China insurers accelerate into 7 trillion yuan "six networks" buildoutChinese insurance funds are stepping up investment in the "six networks" infrastructure program, a roughly 7 trillion yuan ($970 billion) state-backed push that includes computing power and new infrastructure, according to the financial outlet Cailian Press.
More than 40 insurance institutions gathered in Hohhot to review opportunities in computing infrastructure and related fields, the report said. A 1 billion yuan (roughly $140 million) Hubei Yangtze Intelligent Manufacturing industry fund also completed registration.
The "six networks" refers to a Chinese policy framework spanning transport, energy, water, information, logistics and urban infrastructure. The report attributes the acceleration to three converging policy signals but does not name them.
The item is a Chinese-language digest; no company executives, timelines or specific projects were disclosed.
China's Luxeed moves closer to independent EV production licenseChina's Ministry of Industry and Information Technology on Oct. 16 published a list of proposed new vehicle manufacturers that includes Anhui Luxeed New Energy Automobile Co., a step toward giving the Huawei-backed EV brand its own production permit.
Luxeed, known in Chinese as Zhijie, is a collaboration between Huawei and automaker Chery. The brand's vehicles have been built under Chery's manufacturing license; the Oct. 16 MIIT public notice, part of the 412th batch of road motor vehicle manufacturer and product announcements, names Anhui Luxeed New Energy Automobile Co. as a proposed new market entrant, with a production site in Wuhu, Anhui province.
The listing is a procedural step, not final approval. Gaining an independent license would let Luxeed operate as a standalone automaker rather than relying on a partner's credentials, according to the filing.
The Information adds OpenAI, Google, finance, Asia reportersThe Information, the subscription technology news outlet founded by Jessica Lessin in 2013, announced the hiring of new reporters covering OpenAI, Google, finance and Asia, expanding its editorial footprint across the beats its paying readers follow most closely.
The outlet disclosed the additions in a post on its own site. It did not name the incoming journalists or specify start dates in the material provided.
The hires extend a run of newsroom growth at The Information, whose paid-subscription model has made it a reference point for scoops on private AI labs and large technology companies. Its Asia coverage competes with established US and regional tech outlets.
No further details on the roles or coverage assignments were given.
Eric Schmidt's 1% Anthropic stake poised for large windfallFormer Google CEO Eric Schmidt holds roughly 1% of Anthropic, a stake that Bloomberg reports is set to deliver a substantial payout as the AI developer's valuation climbs. The disclosure highlights how early backers of leading AI labs are capturing outsized returns.
Bloomberg reported on Oct. 19 that Schmidt, who led Google from 2001 to 2011 and later served as its executive chairman, owns about 1% of Anthropic. The report did not specify the size of the expected windfall or the valuation used to calculate it.
Anthropic, based in San Francisco, develops the Claude family of AI models and competes with OpenAI, Google and Meta in frontier model development. Schmidt is also a prominent investor in AI and defense technology through his family foundation and personal ventures.
The report offers no timeline for any liquidity event and no confirmation from Schmidt or Anthropic. Details beyond the roughly 1% holding were not disclosed.
DHS bans DeepSeek from federal civilian agency devicesThe Department of Homeland Security has barred federal civilian agencies from using DeepSeek, the Chinese AI developer behind the R1 reasoning model, days after it released a new model. The move extends US restrictions on Chinese AI tools across government systems.
DHS issued the ban, according to Engadget, covering federal civilian agencies. DeepSeek is the Hangzhou-based developer of the R1 and V3 models, which drew attention in January 2025 for competitive performance at low reported training cost.
The ban follows the release of the company's latest AI model a couple of days earlier. US agencies have increasingly restricted Chinese AI software over data-security concerns.
DeepSeek has not publicly responded to the DHS action.
Novus raises eight-figure round to bring Lightbike to CaliforniaGerman mobility company Novus has closed a double-digit million funding round led by 468 Capital, a venture firm with offices in Berlin and San Francisco, as it plans to expand its electric Lightbike from Germany into the California market.
Novus builds the Lightbike, an electric two-wheeler positioned between an e-bike and a motorcycle. The company said the round was led by 468 Capital, which operates from Berlin and San Francisco.
The funding is intended to support the company's move from its German home market into California, according to Tech.eu. Novus did not disclose the exact amount of the round.
Details on investors beyond 468 Capital, the vehicle's specifications, and a timeline for the US launch were not included in the report.
SCMP opinion urges stronger US-China AI safety dialogueA South China Morning Post opinion piece argues that a hotline and ad hoc talks are insufficient to address mounting safety concerns in US-China AI cooperation, covering both the technology itself and the business relationships around it.
The commentary, published by the Hong Kong-based South China Morning Post, contends that current communication channels between Washington and Beijing fall short of what is needed as AI safety concerns grow on both sides.
The piece calls for more structured engagement, saying a hotline and informal dialogue alone cannot manage risks tied to the technology or the commercial ties built on it.
The SCMP is owned by Alibaba Group. The article is an opinion column, not a report of a new government agreement or policy announcement.
US bars DeepSeek AI from government use amid China tech tensionsThe United States has prohibited federal agencies from using artificial intelligence models from DeepSeek, the Chinese startup whose low-cost chatbot unsettled US markets in January, according to BBC Tech. The move extends Washington's widening restrictions on Chinese AI tools across government systems.
DeepSeek, based in Hangzhou, released its R1 reasoning model in January, claiming performance comparable to leading US systems at a fraction of the reported training cost. The release briefly wiped billions off US tech stocks and prompted scrutiny from lawmakers and regulators.
The ban follows earlier restrictions on DeepSeek apps on government-issued devices by several US agencies and states. Washington has separately moved to limit exports of advanced chips and AI technology to China.
Beijing has criticized such measures as market-distorting. Neither DeepSeek nor the White House has issued detailed comment on the latest action, per the report.
US orders DeepSeek removed from federal systemsThe US government has directed federal agencies to remove DeepSeek's AI products from their systems and barred further purchases, citing national security risks tied to the Chinese startup. The move escalates Washington's scrutiny of Chinese AI tools over data and security concerns.
The directive, reported by the South China Morning Post, applies to federal systems and also bars purchases of DeepSeek's products, according to the report. DeepSeek is the Hangzhou-based developer whose low-cost reasoning models drew global attention in early 2025.
The order adds to a widening US effort to restrict Chinese AI software from government use, following earlier bans on platforms such as TikTok on federal devices. It is unclear from the report which agency issued the directive or when it takes effect.
DeepSeek said to shape Huawei AI chips in Nvidia challengeDeepSeek, the Hangzhou AI developer whose low-cost models rattled global markets in early 2025, is reportedly working with Huawei to shape the design of its AI chips, according to The Information, a move that could strengthen a domestic alternative to Nvidia.
The report, published by The Information, describes DeepSeek's involvement in Huawei's chip development as part of a broader push by Chinese firms to reduce reliance on Nvidia, whose advanced accelerators face US export restrictions to China.
Huawei's Ascend line is the leading domestic alternative. DeepSeek's models, which the company said were trained at a fraction of typical costs, have been widely adopted in China and run on both Nvidia and Huawei hardware.
The Information did not detail the scope of the collaboration. Neither company has publicly confirmed the arrangement.
US sanctions five Chinese firms over ex-Google engineer's AI theftThe United States has sanctioned five Chinese entities tied to a former Google engineer convicted of stealing AI trade secrets, according to SCMP Tech, widening a case that has become a flashpoint in US-China technology rivalry.
Among the designated firms is Shanghai Zhisuan Technology, according to the report. The SCMP summary did not name the other entities or specify the sanctions measures imposed.
The case centers on a former Google engineer accused of taking AI trade secrets, a prosecution that US officials have framed as part of broader efforts to curb technology transfers to China.
Beijing has previously criticized US sanctions on Chinese companies as unjustified. Details of the entities' operations and any response from them were not included in the source material.
Sun, Oct 18, 2026
Yijing X9 begins deliveries at Huawei Songshan Lake campusYijing Auto started nationwide volume deliveries of its X9 model at Huawei's Songshan Lake campus on Oct. 18, with Hong Kong actor Aaron Kwok named the brand's first owner, according to ITHome. The launch price starts at 279,800 yuan (roughly $39,000) under a limited-time promotion.
Yijing is a Huawei-backed automotive brand; the X9 is its debut vehicle and is sold with Huawei's in-car technology. Deliveries began at Huawei's Songshan Lake campus in Dongguan, Guangdong, where the company runs a major R&D site.
The X9 competes in China's crowded premium electric SUV segment, where Huawei-affiliated marques such as Aito and Luxeed already sell. ITHome did not disclose order volumes or delivery targets. The limited-time price of 279,800 yuan applies for an unspecified period, the report said.
World Bank urges developing nations to build small AI modelsThe World Bank is pushing developing countries to develop small AI models that run on phones and other devices with limited connectivity, arguing such tools could unlock AI's potential where large cloud-based systems cannot reach, according to The Guardian.
The recommendation, reported by The Guardian on Oct. 18 and relayed by Chinese tech outlet ITHome, centers on on-device models that operate without reliable internet access.
Small models trade some capability for lower compute and bandwidth needs, making them usable in regions where data infrastructure is sparse. The World Bank has previously flagged a widening AI divide between wealthy and developing economies.
The guidance is advisory; the report gave no specific funding figures or named countries.
DeepSeek adds web search, scraping to its APIDeepSeek, the Hangzhou AI developer whose low-cost models drew global attention in early 2025, said its API now supports web search and page scraping, billed at 5 yuan (roughly $0.70) per 1,000 successful requests. The move pushes the company further into agentic tooling, where US rivals such as OpenAI and Anthropic already sell similar retrieval features.
The company announced the pricing on Oct. 18, according to the Chinese tech outlet ITHome. Billing is per successful request rather than per token, a structure aimed at developers building search-augmented agents.
DeepSeek said the feature is available inside DeepSeek Harness, Claude Code and Codex, third-party coding-agent environments where its models can be called. The announcement did not detail coverage, latency or rate limits.
DeepSeek, founded in 2023 by hedge-fund manager Liang Wenfeng, has positioned itself as a low-price alternative to US frontier labs. Its API pricing has repeatedly undercut Western competitors.
Nvidia speeds up Moonshot's Kimi K3 inference by 1.68xNvidia has improved inference throughput for Moonshot AI's 2.8-trillion-parameter Kimi K3 model by 1.68 times through an upstream vLLM optimization, reaching 423 tokens per second per user on GB300 NVL72 hardware, according to research firm SemiAnalysis.
The finding, reported Oct. 18 by Chinese tech outlet ITHome citing SemiAnalysis, concerns Moonshot AI, the Beijing developer of the Kimi chatbot. The optimization targets vLLM, the open-source inference engine widely used to serve large models.
SemiAnalysis said the gain was measured on Nvidia's GB300 NVL72 rack-scale system against the same hardware baseline. No independent verification was provided.
Moonshot has positioned Kimi as a competitor to OpenAI and Anthropic models. Inference cost and speed remain key battlegrounds for Chinese labs seeking overseas users.