Moonshot AI

RUNTO (洛图科技), which tracks smart-glasses retail across JD.com, Tmall and Douyin, published July 2026 figures showing the category fall for the first time: 74,000 units, down 15.3 percent year on year, for 130 million yuan, down 25.8 percent, at an average price of 1,751 yuan, down 12.4 percent. RUNTO attributes the fall to the calendar, because June was the 618 shopping festival and last July was the second-highest month in thirteen. The composition underneath it is the part that is not seasonal. AR glasses, the ones with a display in the lens, sold 33,000 units for 98.28 million yuan, up 72.7 percent by volume and 70.0 percent by revenue, at an average price of 2,990 yuan that did not move all year. Audio-only glasses sold 13,000 units for 8.7 million yuan, down 33.5 percent by revenue at 656 yuan. Camera glasses, the no-display design Meta made familiar with Ray-Ban, sold 28,000 units for 23.1 million yuan, down 46.6 percent by volume and 77.9 percent by revenue, with the average price down 58.5 percent to 819 yuan. The three segments sum to 130.08 million yuan, matching the reported total, and the prior-year segments sum to 175.41 million, which returns the 25.8 percent decline to the digit, so the shift is not a rounding artefact. AR went from 33.0 percent of category revenue to 75.6 percent in twelve months while camera glasses went from 59.6 percent to 17.8 percent. The two leaderboards have come apart: the largest volume share in July, 17.0 percent, went to 清野, a cheap camera-glasses newcomer, while by revenue the order is Rokid, RayNeo, XREAL, Qwen and Huawei, with the top five taking 65.8 percent. Against that, Alibaba previewed a displayless pair called the Qwen N1 at the Bund Summit, with iris recognition at a claimed false-accept rate below one in a million and eye tracking meant to authorise payment by looking at a terminal. In the Briefing: Bank of China put at least 300 billion yuan behind the compute industry at the 2026 China Computing Power Conference, with China Telecom, China Mobile, China Unicom, PICC, China Orient Asset Management and CDB Financial Leasing, targeting at least 100 intelligent computing centre projects and at least 3,000 compute-consuming firms across the 15th Five-Year Plan, while ICBC took its compute-token loan national; China Telecom’s research institute forecast ten quintillion tokens consumed this year rising to more than 3,500亿亿 by 2030, two endpoints its own stated compound rate does not fit, and put inference at 80 percent of China’s compute market by 2029 with 2026 AI capex at leading Chinese tech firms near 600 billion yuan; XPeng’s humanoid production line began operating and the first IRON walked off it under its own power, carrying 76 degrees of freedom and three Turing chips for a claimed 2,250 TOPS; and 25 Fields medallists including Terence Tao and Yu Deng signed a statement arguing that using mathematical problems as capability benchmarks damages the field.

硬氪, the hardware desk at 36Kr, walked through Beijing’s first humanoid-robot data training centre in Shijingshan and found a 3,000 square metre hall with the zone signs still hanging, the display stands still holding their plants, and the robots gone. It stopped seventeen months after it opened, and the third phase that was to run 108 robots across a thousand real tasks in 10,000 square metres was never built. The structure behind it is the story. A robot maker and a local government co-found a data-collection company, usually at a 1:9 or 2:8 split with the maker putting in the small share; the joint venture buys robot bodies, arms and supporting hardware from the maker; a third-party operator sells the data on to companies training models. So the manufacturer contributes one or two yuan in ten, books an order for all ten paid mostly with public money, and the buyer is a company it partly owns. 硬氪 reports more than a hundred such centres went up across 23 provinces in a year, that one robot company heading for a ChiNext listing had as much as 80 percent of its orders from the model, and that another valued above 20 billion yuan shipped 500 robots this year into data factories it had co-funded, against total expected shipments of 1,000 to 2,000. Figures from Interact Analysis cited in the piece put state-capital participation in these centres at 53.8 percent against 88.4 percent a year earlier, and an official at a provincial capital investment-promotion office told 硬氪 they had just turned down two data-collection proposals and that internally some of these projects look like a new kind of Ponzi scheme. The economics were thin even while the money flowed: an operator told 硬氪 that a collector wearing a rig for a full eight-hour shift yields roughly one hour of usable data once idling, out-of-spec motions and broken arms are subtracted, and collection has since gone bodiless to head-mounted egocentric video and UMI-style hand-worn grippers that need no robot present, yielding five to six usable hours out of eight at a crowdsourced rate that has fallen from 45 yuan an hour to 35. Real-robot data still matters, and a recent Tsinghua experiment found that stripping it out drops task success from 81 percent to 40 percent, but it is the seasoning now rather than the main ingredient. The listed-market version surfaced the same day: ASD, a Zhejiang cookware and small-appliance maker, filed an abnormal-fluctuation notice disclosing that its embodied-intelligence robot business has booked 137,200 yuan of sales to date, about 20,500 dollars, against 1.351 billion yuan of first-half revenue and a 93.4 million yuan net loss, and Sinovatio, filing after two straight limit-ups on AI-security enthusiasm, said its AI application products come to under 2 percent of first-half revenue.

Counterpoint has Chinese phone vendors shipping 1 percent fewer units in the second quarter than a year earlier, while sell-through, the count of handsets a person actually paid for, fell by double digits over the same three months. A person working in the sector told the Chinese outlet Dingjiao One that manufacturers expected memory to keep getting more expensive, bought components ahead of need and pushed more finished phones into the channel than the channel was moving, so part of the shipment figure measures warehouse space. Omdia put global second-quarter shipments down 6 percent, with Xiaomi down 26 percent, vivo 18 and OPPO 17. Counterpoint also has more than 40 percent of smartphone models worldwide taking a price increase during 2026 and global retail prices up about 15 percent on average, spread unevenly: India 21 percent, the Middle East and Africa 18, China 10, Europe 7, the United States 5. TrendForce tracked CPU, DRAM and SSD going from about 45 percent of a 900 dollar notebook bill of materials in the first quarter of 2025 to a projected 68 percent in the third quarter of 2026. Chinese memory makers are on the other side of the same trade: Counterpoint puts CXMT at 10 percent of global DRAM revenue in the second quarter, up from 4 percent a year ago, and YMTC at 14 percent of NAND revenue.

The Cyberspace Administration published a phase-two note on its Qinglang campaign against chaos in AI applications on 2 September, and it places obligations at three points in the pipeline rather than one. Platforms including Douyin, Weibo, Tencent and Xiaohongshu are named for expanding face, voiceprint and violation sample libraries, with 46 governance announcements published between them. Four model products, Doubao, Yuanbao, Qwen and ERNIE, are told to guard the content review pass on original training corpora and to label AI-generated synthetic content so as to prevent illegal information at source, which is an obligation on the pre-training set rather than on a feed. The third layer is distribution: Huawei, Xiaomi, OPPO and vivo are named as setting developer admission standards for AI applications and spot re-testing apps already listed, so four handset makers become the admission gate. The note landed one year and one day after the labelling Measures took effect on 1 September 2025. No published standard, audit or methodology sits behind the corpus clause, so what changed this week is scope rather than enforcement capability.

CXMT’s over-allotment option was exercised in full without the bookrunner spending any of it, because the stock never traded below its 8.66 yuan issue price during the window, taking the raise to 66.607 billion yuan gross, about 9.9 billion dollars. YMTC’s 33 billion yuan Shanghai application was accepted on 24 August, and Enflame set 2 September for subscription against a 6 billion yuan target. That is 105.6 billion yuan asked of China’s domestic equity market for DRAM, NAND and AI silicon inside roughly five weeks. CXMT’s first quarter of 2026 alone earned 24.76 billion yuan on 50.80 billion of revenue, at a 79.2 percent gross margin.

Harvey, the OpenAI-backed legal AI company, published its first post-trained model on 20 August. Harvey Tenet is a Kimi K3 base, post-trained with a rank-64 LoRA over the full network on a dataset of about 1,750 tasks. It completes almost twice as many held-out tasks on the Legal Agent Benchmark as base Kimi K3, and places second on that benchmark overall. Harvey ran ablations on candidate judge models against heavier frontier models and settled on Kimi 2.6 as the grader. Its M&A diligence work uses a GLM-5.2 orchestrator, taking rubric criteria pass rate from a 43.8 percent best baseline to 60.1 percent after self-distillation post-training.