Goldman Sachs: AI Investment Theme Shifts from Computing Power to Humanoid Robots
Nashnova编辑部
Goldman Sachs surveyed 14 Chinese robotics firms and now sees humanoid robots as the next AI growth driver after computing infrastructure — but the commercialization tipping point won't arrive until 2027 at the earliest, and Asian valuations still trade at a ~21% discount to U.S. peers, with fund rotation barely underway.
Why does Goldman say the AI investment narrative is changing lanes?
Goldman argues the AI story is shifting from "build compute" to "use compute," with humanoid robots as the most significant application-layer destination.
The structural case rests on two slow-moving forces: persistent labor shortages + rigid demand for automation — not a short-term catalyst.
This means → robots aren't replacing compute spending; they're the next "exit ramp" for compute already built. Compute is the road; robots are the vehicles running on it.
How far has the technology actually progressed?
Goldman strategist Jacqueline Du's team visited 14 companies across Hong Kong, Shenzhen, and Beijing in May. They found VLA/VTLA models — AI systems that let robots perceive surroundings and plan actions — are converging rapidly with world models, boosting robots' ability to understand real environments.
Hardware milestones are visible: Tesla's Optimus appeared at the "We, Robot" event; Unitree's H1 performed on China's 2025 Spring Festival Gala, showing clear gains in flexibility and stability.
But Goldman flags three hard gaps — precision, consistency, and cost — and identifies general autonomous AI capability as the binding constraint. Robots can demo in controlled settings; operating autonomously in any factory remains out of reach.
What makes the data bottleneck so hard to solve?
Unlike large language models, robot training requires force, torque, and motion data from the physical world — the kind that can't be scraped from the internet the way text and images can.
In plain terms = you can train ChatGPT by having it "read" the internet, but training a robot means letting it "touch" the real world. The difficulty and cost are on a different level.
The industry is building centralized data factories and scaling human-robot collaborative collection. Goldman believes data collection itself could become a standalone revenue source within the supply chain.
What does the commercialization timeline look like?
Goldman sees the sector in a tech-validation-to-commercialization transition. Industrial and logistics deployments remain at the proof-of-concept stage.
Most players expect that after accumulating tens of millions of hours of high-quality data and building deployment-ready models, humanoid robots can enter scaled commercial use between 2027 and 2029.
Shipment forecast: 76,000 units in 2027 → 502,000 in 2032. Goldman's long-term view frames humanoid robots as a "next-generation mass-adoption platform" after smartphones and cars.
Why are Asian valuations still at a discount?
The Asia-Pacific robotics basket trades at a median P/E of 22× vs. 28× for U.S. peers — a gap of roughly 21%. On PEG, Asia-Pacific sits at 1.5× vs. 2.0× in the U.S.
Mutual funds are beginning to rotate into the robotics supply chain, but holdings concentrate in components, automotive automation, industrial automation, and precision manufacturing — still early-stage positioning.
This means → the robotics sector has not yet entered a crowded-trade phase. The current moment looks more like the starting line of a capital rotation, not the mid-race sprint.
Which parts of the supply chain have the highest barriers?
Goldman expects value to concentrate in high-barrier core components. Harmonic reducers — precision gear assemblies that convert a motor's high-speed rotation into slow, high-torque output — carry the highest technical threshold, with strict requirements on precision, weight, and torque performance.
Actuator assemblies have relatively high adoption certainty in high-spec robots. Planetary roller screws are still in flux, with yield rates and capacity readiness uncertain. Dexterous hands lack a settled technology roadmap.
This reflects a supply-chain landscape where "buy the basket" won't work — whoever commands volume production of harmonic reducers and actuators holds the highest certainty in this rotation. The commercialization test around 2027 will determine whether the capital rotation pays off.
Content is for reference only, not financial advice.