To The Moon

NextFin News — At the 2026 World Artificial Intelligence Conference, exhibition halls filled with robots performing practical work. Galbot’s S1 lifted and stacked 30-kilogram crates. Other systems sorted household items or tended miniature production lines. Across the floor, the majority of machines handling continuous material movement used wheels rather than legs. Industry tallies of new humanoid-style products released in the first half of 2026 show wheeled platforms slightly outnumbering bipedal ones for the first time. One alliance tracking the sector counted more than 90 new full-machine releases, with wheeled designs accounting for roughly 48 percent and bipedal designs about 45 percent. Companies including Galbot, Pudu, Fourier Intelligence, Qianxun and Youibot introduced or highli

NextFin News —— Cambricon Technologies reported first-half 2026 revenue of 5.996 billion yuan, up 108 percent from a year earlier, and net profit attributable to shareholders of 23.11 billion yuan, up 123 percent. Profit excluding one-off items rose even faster, at 137 percent. The company has now posted seven consecutive profitable quarters. Gross margin held steady near 55 percent. Almost all of the revenue—99.98 percent—came from cloud and data-center chips sold mainly to internet and financial customers. Edge products contributed only a negligible amount. On the day the results were released, the shares closed near 1,200 yuan, giving the company a market value of roughly 754 billion yuan. The stock had earlier pushed the firm’s valuation above one trillion yuan before retreating. The h

TMTPOST -- China's robotics startup X2 Robotics has confidentially filed an initial public offering application with the Hong Kong Stock Exchange (HKEX). According to investment banking sources, a large number of embodied-intelligence companies are currently lining up at the HKEX, and X2 Robotics is the only robot maker positioning itself around a fully self-developed, full-stack, end-to-end embodied model. In the embodied-intelligence space, most “in-house models” are essentially open-source models that have undergone post-training and then been repackaged. Behind X2 Robotics’ “original” label, however, its model was pre-trained entirely from scratch. Mainstream VLA (Vision–Language–Action) models in embodied intelligence are, in essence, a patchwork of multiple independent modules.

NextFin News —For most of the past decade, autonomous driving systems improved by refining separate pieces of software. Cameras and sensors spotted objects. Another program guessed what those objects might do. A third program chose a path. The approach worked in many everyday situations, yet information was lost each time it passed from one piece to the next. The car could react well to familiar conditions and still struggle when something unexpected appeared. That design is now giving way to systems that try to understand and anticipate the physical world as a whole. Industry leaders call the shift physical AI: software that does not merely label what it sees but reasons about motion, cause and effect, and likely futures before it acts. At the 2026 Consumer Electronics Show, NVIDIA chief

NextFin News —NVIDIA has spent the past year expanding what a mobile base station can do. In October 2025 the company invested one billion dollars for a stake of about 2.9 percent in Nokia and agreed to collaborate on radio equipment that runs artificial intelligence alongside ordinary mobile traffic. Nokia plans to offer the resulting systems commercially from 2027, with early tests scheduled for later this year. Work already under way with operators such as T-Mobile has shown base stations that keep phones connected while also performing AI calculations on the same hardware. The practical reason is straightforward. Large AI models are still trained mainly inside centralized data centers. Once those models move into daily use—on phones, vehicles or factory equipment—sending every request

NextFin News — For several years after 2015, selling goods from China to overseas consumers through online marketplaces required little more than a computer, a supplier contact and the willingness to list products. Platforms were still expanding, competition was light, and free or low-cost traffic could carry even basic listings into profitable volume. That window produced visible success stories and a widespread belief that the business remained open to almost anyone. By 2026 the conditions that supported those early results have largely disappeared. Low-value parcel exports from China have declined for consecutive months. Major destination markets have removed or sharply limited duty-free thresholds that once allowed small packages to clear customs with minimal friction. Shipping rates h

NextFin News — In June more than sixty million visits were recorded across seventeen Chinese desktop AI office agents. One product alone, Tencent’s WorkBuddy, accounted for nearly twenty-one million of them. ByteDance’s TRAE IDE followed with almost thirteen million. Alibaba’s earlier offering trailed at under eight million. Those numbers mark a clear change in how people use AI at work: the conversation has moved from answering questions to finishing multi-step tasks. The same summer the three companies that dominate the traffic ranking each reorganized their agent efforts. Alibaba’s response arrived on August 3, when it opened public beta of QwenWork and released a new large model on the same day. The platform merges three earlier agents—one that controlled a user’s computer and files, o

TMTPOST — Xiaoyu Lin sat on the edge of her sofa on a weekday evening, the glow of her phone the only light in the small living room. She had decided that morning to begin training at a nearby gym, and the usual path—scrolling social feeds for starter lists, then opening three shopping apps to compare prices—felt suddenly exhausting. Instead she typed a single sentence into the chat window that now sat at the top of the largest marketplace app on her phone: “I’m a complete beginner at the gym. What do I actually need?” The reply arrived in under four seconds. It listed five categories, attached links, and asked whether she planned to train at home or in a facility. When she answered, the assistant added gloves, a towel, and a compact bag, each with a price and a short rationale. The first

NextFin News — On August 3, the weights for MiniMax’s H3 model appeared on Hugging Face under a community license. The same day, native support landed in ComfyUI. Developers could download the base checkpoints, run them locally through frameworks such as Diffusers, SGLang or vLLM, and generate short clips that combine video and native stereo audio from mixed text, image, video and audio inputs. The release followed an announcement on July 31 that positioned H3 as a general-purpose multimodal generator capable of producing 4-to-15-second sequences at up to 2K resolution. Independent rankings from Artificial Analysis place H3 first on the video-editing leaderboard, ahead of models from Google and other major laboratories. It also ranks near the top in text-to-video and image-to-video categor

Text | Shixiang, Author | Wang Tiemei, Editor | Gu Nian On July 1, 2026, the European Union’s sweeping new tariff regulations on cross-border small parcels officially took effect. The era of duty-free exemptions for packages valued under €150 came to a definitive close, replaced by a mandatory flat tariff of €3 on every low-cost item. Under this altered operating landscape, a significant portion of low-cost SKUs priced under €5 on Temu’s European platforms saw final consumer checkout prices surge by more than 70% over the past week. Driven away by these sudden tariff costs, a wave of consumers abandoned their shopping carts at checkout. Multiple merchants told ShiXiang that order volumes have plunged by at least 60% compared to pre-regulation baselines. For the past three years, Temu ancho