跟 Serenity 学美股：美国和中国的一场重大前沿竞赛 白毛股神 Serenity 转述了 SVRC Research 今年 4 月发布的《State of Robotics 2026》年度报告中的核心发现，并在此基础上给出了自己的判断： 这将是美国和中国之间的一场重大前沿竞赛 她本人持有的 Agility Robotics $CCXI 被列入了该报告评出的美国机器人"国家冠军队"名单 --- 1️⃣SVRC 列出了美国机器人领域的十大国家级冠军企业： 1. Figure AI — 估值 390 亿美元（2025 年 9 月 C 轮），第三代 Figure 03 已在 BMW 南卡工厂部署，累计融资约 23 亿美元，未上市 2. Agility Robotics (CCXI.US) — Digit 机器人已累计超 6.5 万小时真实运营，横跨 9 个客户现场（包括 Amazon、GXO、Schaeffler、Toyota 加拿大工厂），已签约超 3 亿美元多年期订单，借壳 $CCXI 上市中 3. Apptronik — 估值约 55 亿美元（2026 年 2 月 A-X 轮），累计融资近 10 亿美元，Apollo 2 已发布，与 Google DeepMind 共建 Gemini Robotics 联合训练体系，未上市 4. Tesla (TSLA.US) — Optimus Gen 3 正将 Fremont 工厂原 Model S/X 产线改造为专用机器人产线，搭载 AI5 芯片，预计 2026 年夏末启动低速量产 5. Boston Dynamics — 现代汽车已全资收购（2026 年 6 月买下 SoftBank 最后 9.65% 股份），全电动 Atlas 量产版刚在 FIFA 世界杯赛场完成首次公开亮相，长期目标年产 3 万台 6. Physical Intelligence (Pi) — 专注机器人基础模型（π 系列 VLA 模型），不造硬件，B 轮估值 56 亿美元，据报道正以 110 亿美元估值推进新一轮融资 7. 1X Technologies — NEO 家用人形机器人定价 2 万美元或月租 499 美元，首年产能 5 天售罄，2026 年底开始发货，加州 Hayward 工厂初始年产能 1 万台 8. Amazon Robotics (AMZN.US) — 物流网络已部署超 100 万台各类机器人，通过投资 Agility 并深度测试 Digit 切入人形赛道，2024 年 8 月还以"反向收购雇佣"方式获取了 Covariant 的基础模型技术授权和核心团队 9. Covariant — 2024 年 8 月被 Amazon 以反向 acqui-hire 模式获取了三位联合创始人及约 25% 员工，同时 Amazon 取得其机器人基础模型的非独家授权，公司仍独立运营 10. Skild AI — 开发通用机器人大脑"Skild Brain"（全身型基础模型），2026 年 1 月 C 轮融资 14 亿美元，估值超 140 亿美元，SoftBank 领投，NVIDIA、Bezos 参投 报告核心判断：美国赢在未来，输在当下 SVRC 的原话是： "The United States leads the world in where robotics is heading: Foundation models, OpenAI-style scaling laws applied to action, autonomous vehicles. While losing the race on where robotics is shipping today." 翻译过来就是：美国在基础模型、具身智能缩放定律、自动驾驶这些"下一代方向"上处于全球领先。但在当下实际出货和量产部署的竞赛中，正在落后 --- 2️⃣报告将以下两个环节定性为美国机器人产业的结构性脆弱点： 软肋一：稀土材料暴露 钕铁硼（NdFeB）磁体是高性能伺服电机的核心——没有它，机器人关节就失去了力量密度。钐钴（SmCo）磁体则用于高温极端工况下的特种应用。这两种材料的供应链命脉几乎完全掌握在中国手中：中国占全球稀土开采的 60-70%，精炼加工的约 90%，永磁体制造的约 94% 这就好比一栋楼的钢筋全从一个供应商进货——平时没问题，一旦断供，连打地基都做不了 软肋二：执行器依赖 弹性串联执行器（SEA）、准直驱电机（QDD）和精密减速器是人形机器人关节的核心传动系统——相当于人体的肌腱、韧带和关节软骨。这些部件的全球供应链高度集中于三个国家： 1. 日本——Harmonic Drive Systems (6324.T) 掌握应变波谐波减速器原创技术，Nabtesco 垄断 RV 摆线减速器，Nidec-Shimpo 在扩张应变波产品线 2. 德国——Schaeffler 的行星齿轮执行器平台在 2026 年汉诺威工博会上获赫尔墨斯奖，Bosch Rexroth 的 ctrlX 自动化平台持续渗透 3. 中国——绿的谐波 (688017.SH) 以性价比切入国产替代，三花智控 (002050.SZ) 与 Tesla 联合开发执行器，拓普集团 (601689.SH) 切入 Optimus 关节组装 报告同时将制造速度不足、数据收集成本高昂（虽然遥操作示教数据成本已从 2024 年初每小时约 340 美元降至 2026 年初约 118 美元）和监管法规列为其他主要制约因素 --- 3️⃣2027 年整合预警与首批落地场景 SVRC 的判断是：至少六家资金充足的美国人形机器人公司正在争夺一个仍处于早期的市场。预计 2027 年将发生至少两起重大并购或合并事件 报告认为首批规模化部署将发生在两个场景： 第一，物流与电商——Amazon (AMZN.US) 已在仓库中深度测试 Agility Digit，FedEx (FDX.US) 同样是核心买家 第二，汽车制造——GM (GM.US) 与 Ford (F.US) 的标准化产线，天然适配第一代人形机器人有限但可靠的操作能力边界 --- Serenity @aleabitoreddit 的个人判断：这将是美国与中国之间的一场前沿竞赛 美国阵营代表： Agility Robotics（Serenity 持有仓位）、Tesla、Figure AI、Apptronik 中国阵营代表： 宇树科技 Unitree（已获证监会科创板 IPO 批文，拟募 42 亿元，估值约 420 亿元/60 亿美元） 智元机器人 AGIbot（计划港股 IPO，目标估值 51-64 亿美元，已产超 1.5 万台） 优必选 Ubtech (9880.HK，当前市值约 70 亿美元) 以及其他后续跟进的中国参与者 Serenity 特别强调，她一直在讨论的许多观点现在正被研究机构所验证——这种来自独立第三方的交叉确认，强化了其对供应链卡点分析方法论的信心 感谢Serenity @aleabitoreddit 提供的信息及思路🫡 仅供学习交流，请DYOR

Massive week in robotics. I went through everything from Weave Robotics, Apptronik, Flexion, and even more. Get your coffee and see what happened in robotics space:

[INTERNET EXPLORER NEWS] Agility Robotics is going public via SPAC merger at a $2.5B valuation, one of the first pure-play humanoid robotics companies to pursue a public listing Digit is deployed across 9 customer sites including Schaeffler, GXO, Toyota, and Mercado Libre. $ 300M in multi-year orders secured with 30 customers in the pipeline The deal is expected to generate $ 620M in proceeds for production scaling

The robotics moat is not the model. Apptronik has Google DeepMind. X Square Robot has WALL-B. Figure has Helix 02. All different model architectures. All roughly equivalent performance on standard benchmarks. The moat is data. Who has the most real-world hours? Who has the most failure cases? Who has the most edge conditions logged? Models converge. You can fine-tune any architecture on good data and get similar results. But data compounds. Every hour of real-world deployment makes the next hour safer. The robotics companies worth investing in are not the ones with the best demos. They are the ones with the most logged hours.

Then @FlexionRobotics introduced Reflect v1.0, a robotics intelligence platform achieving 90% success on complex multi-step missions vs 38% without reinforcement learning In a demo, a humanoid retrieved a parcel, navigated stairs and an elevator, unpacked the box, and stored items in a drawer, all from a single natural-language instruction Users can modify missions or update instructions mid-execution just by changing the prompt

Austin-based @Apptronik launched Apollo 2 in both bipedal and wheeled configurations The robot collects training data across logistics, manufacturing, and retail tasks inside Apptronik's newly expanded 90,000 sq-ft Robot Park in Austin That data directly powers Google DeepMind's Gemini Robotics foundation models. Everything proven on Apollo 2 feeds into the commercial Apollo 3

Apptronik just opened a 90,000 square foot robot training facility in Austin. They are not building robots there. They are building a data factory. Every robot in that facility exists to collect training data. Every movement, every grasp, every error becomes a training example. The company with the most data wins. The facility is the moat. Figure has BMW data. Tesla has Fremont data. Apptronik now has 90,000 sq ft of purpose-built data collection infrastructure. Models converge. Hardware converges. Data is the only defensible advantage in robotics.

You're missing Apptronik and Agility Robotics. But WLTH users aren't.

Here is the most honest assessment of the humanoid robotics market you will read this year. The companies that matter: Figure AI: Best-in-class deployment. 11 months at BMW. Helix 02 is a genuine breakthrough in pixels-to-actions learning. Funded well. Shipping to real customers. Tesla Optimus: Biggest manufacturing advantage. 1,000 units already in Fremont. Can scale to 100,000/year faster than anyone because they already own car factories. The robot is not the best. The factory is the moat. Boston Dynamics: Best brand, worst business model. Atlas Gen 5 is finally manufacturable (10x fewer parts), but Hyundai's 30,000 units/year target is 2 years away. They are late. UBTech: The dark horse. $17,650 consumer price point. 13,361 pre-orders. China-based manufacturing. If anyone makes humanoids a consumer product, it is UBTech, not Tesla. Apptronik: Best data moat. 90,000 sq ft training facility. Google DeepMind partnership. They are not winning the deployment race, but they are winning the data race. And data compounds. X Square Robot: Best valuation story. $2.8B valuation on the WALL-B foundation model. But more robots than employees, which is either impressive or concerning depending on your perspective. The losers: Everyone making demo videos instead of shipping products. You know who you are. The humanoid market in 2026 is where the smartphone market was in 2007. The iPhone just shipped. Everyone else is still making BlackBerries.

"Yo diría que la industria actual se basa, prácticamente, solo en prototipos", me comentó ayer Jeff Cardenas [ @jeffcardn ], CEO de @Apptronik, al anunciar Robot Park, una enorme fábrica física de datos de IA @forbesargentina forbesargentina.com/innovaci…

Apptronik calls Apollo 2 a "data collection and training platform," and has run it as one at Robot Park for over a year: 90,000 sqft (8361 sqm in non-retarded units), plus task data off Mercedes and GXO floors. The humanoid becomes a sensor for the flywheel, while data pipeline is the acutal moat.

SVRC Research’s “State of Robotics 2026” report is worth paying attention to. It listed Figure AI, Agility Robotics / $CCXI, Apptronik, $TSLA Optimus, Boston Dynamics, Physical Intelligence, 1X, $AMZN Robotics, Covariant and Skild AI as the key U.S. robotics “National Champions.” The most important line: The U.S. leads where robotics is heading. Foundation models. OpenAI-style scaling laws applied to action. Autonomous vehicles. Robot intelligence. But the U.S. is still losing where robotics is shipping today. That is the real U.S. vs China robotics gap. America has the capital, software, AI labs and enterprise pilots. China has manufacturing velocity, lower cost hardware, faster iteration and real shipment volume. SVRC also highlighted the weak points: Rare earth exposure. Actuator dependency. Precision reducers. Manufacturing speed. Data collection cost. Regulatory pressure. This is why I think robotics becomes one of the next major national security races. For the U.S., the leaders look like Agility, Tesla, Figure and Apptronik. For China, the pressure comes from Unitree, AgiBot, UBTECH and others. The market is still early, but the direction is clear: Robotics is moving from demo videos to deployment economics. Disclosure: long Agility / $CCXI.

Everyone keeps asking which company will win the humanoid robot race. Tesla, Figure, Unitree, Apptronik, and others dominate the conversation. But that framing misses where the real competitive advantage sits. The finished robot is only the visible layer of a much deeper industrial ecosystem. Every humanoid depends on technologies that relatively few companies in the world can manufacture at scale. Precision reduction gears, bearings, servo motors, magnetic materials, embedded controllers, industrial sensors, and advanced composites determine whether a robot is accurate, reliable, and manufacturable. Japan occupies an outsized position across many of these foundational technologies, with companies like Nabtesco, Harmonic Drive Systems, NSK, NTN, JTEKT, Nidec, Yaskawa, Renesas, Omron, Keyence, Shin-Etsu Chemical, TDK, and Toray holding leading positions in their respective markets. This is why focusing exclusively on humanoid OEMs can be misleading. Regardless of whose logo is on the robot’s chest, many of the highest-value components originate from the same small group of industrial technology leaders. The companies building the robots compete with one another. The companies enabling the robots often supply them all. The Physical AI race will not be won by software alone. It will be shaped by the industrial ecosystem capable of delivering the precision components that every humanoid requires. That ecosystem is where Japan continues to exercise disproportionate influence. #HumanoidRobots #PhysicalAI #Japan #Robotics #Manufacturing #IndustrialAI #Automation

Serenity 7/5 ê 重點消息 美國 RO-BOT-IKS「國家隊」報告 「State of Robotics 2026」 報告（SVRC Research 出 ê）内底 ū Figure AI、 Agility Robotics（透過 $CCXI）、 Apptronik、 Tesla、 Boston Dynamics 等美国上重要 ê RO-BOT-IKS kap 人形 RO-BOT 公司。