Humanoids will have their ‘GPT-3.5’ moment within 5 years, says AGIBOT co-founder Yao Maoqing

AGIBOT co-founder Yao Maoqing expects embodied AI to reach a GPT-3.5-like level of general-purpose capability within 3-5 years, though reliable industrial deployment remains a major challenge. AGIBOT shipped 9,700 humanoid robots in the first half of the year, leading the market according to Counterpoint Research, and is considering a Hong Kong IPO. The company reported a 99.99% success rate across more than 64,000 manufacturing tasks in a six-day June demonstration, supporting optimism around scalable industrial robotics.

Analysis

The investable signal is not a near-term humanoid revenue inflection; it is a widening valuation gap between robotics suppliers with current automation cash flows and private humanoid developers whose economics remain unproven. Factory buyers optimize for uptime, throughput and fully loaded labor substitution, making conventional automation vendors such as FANUY, ABB, ROK and TER more likely to capture the first spending wave than general-purpose humanoid platforms. Chinese component scale could also pressure actuator, servo and vision-system pricing, creating margin risk for higher-cost Western pure plays before it creates a large new end market.

Claims of near-perfect task reliability should be treated as a controlled-demonstration datapoint, not evidence of production economics. The key bottleneck is tail-risk handling: a robot can appear reliable in repeated workflows while failing economically when parts vary, stations change or human interaction is required. Over the next 1-3 months, prospective Hong Kong listings could attract scarcity-driven AI capital; over 6-18 months, procurement contracts, recurring maintenance revenue and independently measured utilization will matter far more than demo performance.

Consensus is likely over-indexing to humanoids as a replacement for dedicated industrial automation. The more probable adoption curve is complementary: mobile manipulators and humanoids fill low-volume, variable tasks, while fixed robots retain high-speed repetitive work. That favors warehouse/logistics exposure through SYM and AMZN, and incumbents able to bundle software, service and safety certification, rather than a broad rerating of all "embodied AI" equities.

A reversal of the constructive automation thesis would come from Chinese wage deflation, weak manufacturing capex, or evidence that deployment requires persistent human supervision that erases labor savings. Watch ABB, ROK and TER order growth and margin guidance; sustained order deceleration or rising warranty/service provisions would indicate that robotics hype is not converting to economically viable installations.

Key Decisions for Investors

• Prefer a 6-12 month long basket of ABB, ROK and FANUY over a broad speculative humanoid-equity exposure. These firms monetize automation budgets today; reassess if organic automation orders fall for two consecutive quarters or if price competition drives segment margin down more than 150 bps.
• Use any Hong Kong humanoid-robot IPO as an event-driven watch item rather than an initial core long. Require disclosed customer concentration, paid deployment count, maintenance revenue and independently verified utilization before underwriting revenue; avoid offerings priced on AI-platform multiples without those data.
• Consider a 3-6 month pair trade long ABB / short TER only if semiconductor test capex weakens while industrial automation orders stabilize. ABB has broader factory-automation exposure, whereas TER remains more sensitive to volatile semiconductor utilization; close if semiconductor equipment bookings inflect upward.
• Monitor AMZN and SYM for logistics-automation read-through, but do not chase on robot demonstrations. A material catalyst requires disclosed fulfillment-center labor-productivity gains or accelerated automation capex, not prototype announcements.