China leads, the US follows, the "singularity moment" for humanoid robots has arrived!

```

The past month has seen more intensive developments in the global humanoid robot industry than the entire past year.

Leading Chinese OEMs are already running pilot projects with hundreds of units in logistics sorting and manufacturing lines, and are requesting expansion; Boston Dynamics is preparing to deliver Atlas to Hyundai's factory; Tesla announced the mass production of Optimus will start at the Fremont factory in 2026, with a second factory in Texas scheduled for 2027. The industry has already pivoted from “showcase robots” to “real-world deployments.”

According to Chasing Wind Trading Desk, JPMorgan's Asia-Pacific infrastructure and industrial analyst Karen Li stated in her latest report: The gap between winners and long-tail players is widening, and capital is increasingly concentrating on two types of companies—those with profitability and mass production readiness (platform enterprises), and those supplying high-quality components and AI/software 'brains'. The report team conducted joint field research at the Beijing Robotics and Auto Show from April 22 to 24, held a global investor webinar on robotics on April 16, and participated in the M+ Forum in Malaysia, meeting with UFactory's Malaysian authorized partner, Work E Robotics.

Among them, Tesla’s position in this race is more complicated than its market value suggests. The company plans capital expenditures exceeding $25 billion in 2026 for AI, robotics, and in-house chips; Optimus V3 design is nearing mass production, but the report keeps Tesla at an “underweight” rating with a $145 target price, while the current stock price is about $374.

In comparison, leading Chinese OEMs are pushing commercialization at a faster pace thanks to government procurement, supply chain advantages, and high-frequency hardware iteration; Boston Dynamics also holds a first-mover advantage in industrial integration.

The real bottleneck is not whether the hardware can move, but whether it can run stably

The most consistent feedback from the Beijing research: the main industry obstacle has shifted from “can prototypes complete tasks” to “can tasks be stably completed under mass production conditions.” Reliability, maintenance cycles, and integration time with production lines were recurring themes.

A dexterous hand supplier revealed 2025 shipments have already exceeded 10,000 units and are expected to double in 2026—this offers a reference point for how quickly commercial demand is shifting from data collection and showroom display to real-world deployment. However, hand component suppliers also stress that scaling challenges have become comprehensive tests of temperature, vibration, corrosion, and durability—not just whether “it can grip objects.”

Physical intelligence—the robot's “brain”—is generally seen as the core commercialization hurdle for 2026. VLA (Vision-Language-Action) models convert language and video understanding into robotic actions; world models handle reasoning, planning, and environmental understanding; both are considered the ultimate path to embodied AI. Sim-to-real transfer remains a common challenge. One leading humanoid robot company is tackling real-world data bottlenecks, especially authenticity in force, friction, and touch, through “control world” tools and data infrastructure layers.

The commercial logic is also diverging. Some OEMs sell the “brain” bundled with the robot body, some sell hardware only, and others provide SDKs for customers to develop the intelligence layer themselves. The key judgment here is: a significant portion of today’s business traction comes from big tech and industrial clients viewing robots as data collection tools—not purely as labor replacements.

Tesla is betting on the right direction, but its timeline may be a gift to rivals

The $25 billion capital expenditure plan is real, mass-producing Optimus is a true strategic priority, but Tesla’s management admits the initial ramp-up will be slow. The Fremont factory aims for a capacity of 1 million units/year, with a second Texas factory for further expansion, but all of this is for 2027 and beyond.

The company has deliberately limited public demos of Optimus 3 for IP protection and to avoid copycats—this explanation alone shows how fierce the competition is. At the same time, Tesla’s self-developed AI5 chip will power both Optimus and data centers; the vertical integration of chip manufacturing is its key defensive moat, but building this capability takes longer than ramping up robot mass production.

JPMorgan’s assessment of Tesla: it is catching up to leading Chinese OEMs and Boston Dynamics, not leading the pack. Core EV business competition and margin pressures remain; the current 187x 2026 forward P/E is almost entirely based on AI and robotics expectations.

Chinese manufacturers are advancing faster due to government support, supply chain, and rapid iteration, while Boston Dynamics is aligning the Atlas roadmap to Hyundai’s industrial application (with first applications possibly starting in 2028).

The logic in Southeast Asia is different from China: it’s not about saving labor, but keeping people out of dangerous areas

The Malaysia M+ Forum revealed an interesting perspective. Southeast Asian companies considering robot deployment do not prioritize labor costs—their low-wage environment invalidates this logic—but operational flexibility: 24/7 operation, replacing humans in hazardous environments, and stable work quality.

The oil and gas industry is currently the clearest buyer, with UFactory’s B2 quadruped robot (IP protection level, ~40kg payload) being considered for gas leak detection and perimeter patrol. Manufacturing is seen as the next adoption wave, with the UFactory G1 humanoid robot (upper body plus AMR base) targeting intra-factory logistics under controlled ground conditions. However, large-scale procurement may require two to three more years of solution maturity and further hardware price reductions.

The business model is mainly buy-out for now, but discussions around RaaS (Robots as a Service) and subscriptions are increasing. Work E Robotics’ management said bluntly, “Once hardware becomes standardized, solutions are key.” Integration and deployment—from site modeling to LLM and sensor integration—will become the real sources of profit and customer stickiness.

The funding window is narrowing, but the way it’s narrowing favors the leaders

Primary market financing in early 2026 remains active, but the landscape is changing: as valuations are pushed up, capital is focusing on a few platform companies and high-quality component suppliers. Smaller OEMs face steep challenges: the combined costs of VLA-scale training compute, data acquisition, and manufacturing ramp-up far exceed the funds they can raise.

This divergence will likely spur M&A, strategic partnerships, and structured finance, instead of long-tail OEMs developing independently. The IPO pipeline is a key catalyst; government procurement and public sector projects are becoming major order drivers, especially as local governments establish data collection centers and pilot zones, causing orders to cluster strongly toward the later stages.

On the stock side, over the past month the Chinese robotics sector has rebounded an average of 10%, but JPMorgan’s logic is clear: companies with commercial momentum, strong order visibility, and technical differentiation deserve heavy positions; for small OEMs lacking these traits, the environment will only get tougher.

Risk Warning and DisclaimerMarkets have risks, investment requires caution. This article does not constitute personal investment advice and does not take into account individual users' unique investment goals, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article suit their particular circumstances. Invest accordingly at your own risk. ```