The global humanoid robotics sector has witnessed more activity in the past month than in the entire previous year. Leading Chinese OEMs are already piloting projects involving hundreds of units in logistics sorting and manufacturing production lines, with requests for expansion. Boston Dynamics is preparing inventory to deliver its Atlas robots to Hyundai's automotive factories. Tesla has announced the start of mass production for its Optimus robot at its Fremont facility in 2026, with a second plant in Texas scheduled for 2027. The industry has already transitioned from "showpiece robots" to "real-world operational deployment." According to analysis, the gap between winners and the "long tail" of smaller players is widening, with capital increasingly concentrating on two types of entities: profitable, platform-based companies ready for mass production, and suppliers of high-quality components and AI/software "brains." An analyst team completed a joint research tour of robotics and auto shows in Beijing from April 22-24, hosted a global investor webinar on the robotics industry on April 16, and also attended a forum in Malaysia, engaging with a local authorized partner. Tesla's position in this race is more complex than its market capitalization suggests. The company plans capital expenditures exceeding $25 billion in 2026 for AI, robotics, and in-house chip development. The Optimus V3 design is nearing mass production readiness. However, the analysis maintains an "underweight" rating on Tesla with a $145 price target, compared to its current stock price of approximately $374. In contrast, leading Chinese OEMs are advancing commercialization at a faster pace, leveraging government procurement, supply chain advantages, and rapid hardware iteration. Boston Dynamics also holds a first-mover advantage in industrial integration pathways. The primary bottleneck is no longer whether hardware can perform tasks, but whether it can operate reliably under mass production conditions. The most consistent feedback from the Beijing research was that the main industry challenge has shifted from "can a prototype complete a task?" to "can it perform stably under mass production?" The terms reliability, maintenance cycles, and integration time with production lines were repeatedly emphasized. A dexterous hand supplier revealed that its 2025 shipment volume has already exceeded 10,000 units, with an expectation to double that in 2026. This figure provides a reference point for how quickly commercial demand is moving from data collection and showroom displays to real deployment scenarios. However, component suppliers also stress that the challenge of scaling has become a comprehensive test of temperature resistance, vibration, corrosion, and durability, moving beyond simply "can it grasp an object." Physical intelligence—the robot's "brain"—is widely seen as the core bottleneck for 2026 commercialization. VLA models, responsible for mapping language and video understanding into robot actions, and world models, handling reasoning, planning, and environmental understanding, are considered the ultimate path for embodied AI. The "sim-to-real" gap remains a universal challenge. A leading humanoid robotics company is addressing the real-world data bottleneck, particularly for force, friction, and tactile authenticity, using its "control world" tools and data infrastructure layer. Commercialization logic is also diversifying. Some OEMs are bundling the "brain" with the robot本体, others sell only hardware, and some provide SDKs for customers to develop their own intelligence layer. Underpinning this is a key observation: a significant portion of current commercial traction comes from large tech companies and industrial clients using robots as data collection tools, rather than purely for labor substitution. Tesla is betting on the right direction, but its timeline might be giving opportunities to competitors. The $25 billion capital expenditure plan is real, and Optimus mass production is a genuine strategic priority. However, Tesla management itself acknowledges that initial production ramp-up will be slow. The Fremont factory targets an annual capacity of 1 million units, with the second Texas plant intended for further expansion, but these are goals for 2027 and beyond. The company is deliberately limiting public demonstrations of Optimus 3, citing the official reason of protecting intellectual property and preventing competitor imitation—an explanation that itself highlights the intensity of competition. Meanwhile, Tesla's in-house AI5 chip is designed to power both Optimus and data centers. This vertical integration in chip manufacturing is a key part of its defensive moat, but building this capability has a longer timeline than robot mass production. The qualitative assessment of Tesla is that it is catching up to leading Chinese OEMs and Boston Dynamics, rather than leading the pack. Competitive pressures and margin issues in its core EV business persist. With a forward 2026 P/E ratio of around 187, the valuation logic appears almost entirely dependent on AI and robotics. Chinese manufacturers are advancing faster due to government support, supply chain advantages, and rapid iteration. Boston Dynamics is also advancing the Atlas timeline for industrial applications at Hyundai, with initial deployments potentially starting from 2028. The rationale in Southeast Asia differs from China: the focus is not on labor savings, but on keeping humans out of hazardous areas. A forum in Malaysia provided an interesting perspective. For Southeast Asian enterprises considering robot deployment, the primary driver is not labor cost—low wages negate that logic—but operational flexibility: 24/7 operation, human substitution in dangerous environments, and consistent work quality. The oil and gas industry is currently the most clearly identified buyer. Quadruped robots are being considered for gas leak detection and perimeter patrol. Manufacturing is seen as the next adoption wave, with specific humanoid robot models targeted for factory logistics in controlled ground conditions. However, large-scale procurement likely requires another two to three years for solution maturity to improve and hardware costs to decrease further. The prevailing business model is currently outright purchase, though discussions around RaaS and subscription models are increasing. Integration deployment— involving site modeling, and combining LLMs with sensors—is expected to become the key profit source and driver of customer loyalty. The funding window is narrowing, but this trend favors leading players. Early 2026 private market financing remains active, but the landscape is changing: while valuations are being pushed higher, capital is increasingly concentrated on a few platform companies and high-quality component suppliers. Smaller OEMs face the challenge that the combined funding required for VLA model training compute costs, data acquisition costs, and manufacturing capacity ramp-up far exceeds what they can raise. This divergence is likely to spur mergers and acquisitions, strategic partnerships, and structured financing, rather than the independent development of long-tail OEMs. The IPO pipeline is a significant catalyst. Government procurement and public sector projects are becoming major order drivers, especially after local governments establish data collection centers and pilot zones, leading to orders being concentrated in the later stages. Regarding stocks, the average Chinese robotics sector stock has rebounded about 10% over the past month. However, the investment logic is clear: companies with commercial momentum, strong order visibility, and technological differentiation warrant significant investment. For small OEMs lacking these characteristics, the environment is only becoming more challenging.