Goldman Sachs' Major Prediction: In the Second Half of the AI Era, Focus Will Shift from Chips to Humanoid Robots

Goldman Sachs believes that the AI investment narrative is shifting from the infrastructure layer to the application layer, with humanoid robots becoming the next clearest frontier for monetization. The bank recommends that investors position themselves early to seize opportunities in the initial cycle brought about by years of capital rotation within the Asia-Pacific robotics ecosystem.

Goldman Sachs equity strategist Jacqueline Du revealed in the latest research report that the bank attended the Asia Technology Conference in Hong Kong from May 18 to 22 and carried out special research on Chinese AI robots in Shenzhen and Beijing, communicating in depth with 14 robotics companies, both private and publicly listed. The report noted that Goldman Sachs was "deeply encouraged"—the rapid integration of VLA/VTLA models and world models is significantly improving robots' planning abilities and robustness, with model scales migrating to approximately 40 to 80 billion parameters.

Peter Sheren, Goldman's trader based in Singapore, stated that the AI narrative is shifting from infrastructure to applications, with structural tailwinds such as labor shortages and automation needs accelerating industry adoption and driving capital rotation towards robot-related targets in Korea, Japan, and China. From a valuation perspective, the median P/E ratio (22x) of Asia-Pacific robot stocks is about 21% lower than their U.S. counterparts (28x), combined with higher expected profit growth, making Asia-Pacific robot stocks stand out in terms of cost-effectiveness.

Despite the optimistic outlook, Goldman Sachs also clarified that large-scale commercial deployment still requires patience. Most applications are still in the proof-of-concept stage, with industry players generally expecting large-scale deployment to occur between 2027 and 2029; high-quality real-world data remains the key bottleneck restricting industry development.

Two Cities, Five Days: Core Findings from Goldman Sachs Mid-Year Research

This research by Goldman Sachs lasted five days and was conducted in two phases: attendance at the GS Asia Communacopia + Technology Conference in Hong Kong from May 18–19, and focused research on Chinese AI robots in Shenzhen and Beijing from May 20–22. In total, 14 companies were involved, covering a broad cross-section of embodied AI, robotics, and automation ecosystems, including: Damon Robotics, Yuejiang, Estun Automation, Xingtu Power, Gabot, Geek+, Zhujing Power, Linkage Robotics, Mech-Mind, Wanjin Robotics, Paxini, Lingjing AI, UBTECH, Fangyu Robotics.

Jacqueline Du characterized the research as a "mid-year check-up for the industry," and summarized two core findings: First, the integration of VLA/VTLA models and world models is advancing rapidly, with significant expansion in model scale; second, high-quality real-world data is still the primary bottleneck for deployment, but industry discussions have shifted from macro-level "data recipes" to how to build scalable data collection architectures.

Technology Acceleration: Rapid Integration of Multimodal Stacks, Intensifying Data Competition

From a technical perspective, industry discussions have clearly moved beyond the single VLA framework, turning toward execution-oriented multimodal stacks. The specific path is to first quickly integrate VLA and world models, and then add tactile capabilities (VTLA) where high-quality physical interaction is required. The world model is no longer treated as a separate model category but operates alongside functional and motion models—VLA or VTLA handles strategy generation and motion execution, while the world model improves execution quality through next-state prediction, action validation, and planning optimization.

Companies explicitly adopting direction of combining VLA/VTLA and world models include Xingtu Power, Gabot, Lingjing AI, and Wanjin Robotics. Xingtu Power will release Fast-WAM in March 2026, with latency as low as 190 ms; Lingjing AI’s open-source model Spirit v1.5 ranked first on RoboChallenge Table30 with 66.09 points and a 50.33% success rate, becoming the first Chinese open-source embodied model to surpass Pi0.5.

On the data front, human-centered, self-perspective data collection methods are becoming industry consensus. Paxini operates five data factories nationwide; Xingtu Power, Lingjing AI, and Wanjin Robotics build distributed data closed-loops via system deployments, wearables, VR, and client-side collection. Multiple companies expect the proportion of data-related revenue to increase significantly in 2026, while UBTECH anticipates government data factory demand in 2026 to remain at or increase from 2025 levels.

Notably, Goldman Sachs observed that due to model capability constraints and cost concerns, many players currently prefer robot forms with wheeled chassis and two or three-finger grippers, believing these can cover 70% to 90% of industrial scenarios, though future evolution toward bipedal humanoids and dexterous five-fingered hands is not excluded.

Commercialization Path: POC Dominates, Scale Deployment Awaits 2027–2029

In commercialization, application scenarios are extending toward industrial handling, logistics workflows, and structured business scenarios, but the overall process is still mostly proof-of-concept (POC), with large-scale deployment yet to arrive.

For industrial adoption, the typical path consists of four steps: POC stage (usually lasting 3–6 months, with 2–3 rounds on average) → small batch testing (usually less than 50 units per order) → validation period of about 12 months → pilot deployment (orders gradually reaching 50–100 units per customer). The most representative recent applications include sorting, material handling, picking and placing, inspection/testing, and other standardized or semi-structured workflows.

The industry generally expects that after accumulating tens of millions of hours of high-quality data and developing deployment-ready models, large-scale commercial deployment will occur from 2027 to 2029. Goldman Sachs believes that despite current challenges, "the long-term investment prospects for this industry are highly promising," but "this journey still requires patience, as enterprises must overcome the complex transition from proof-of-concept to mass commercialization, with milestones being stable quality and continuous cost reduction."

Supply Chain Layout: Focus on High-Barrier Core Components

For supply chain investment frameworks, Goldman Sachs favors products with high content value, focusing on two main directions: harmonic reducers and actuator assemblies.

Harmonic reducers: Goldman Sachs considers these to have the highest technical barriers, with stringent requirements for precision, light weight, and torque. Among high-spec products, Harmonic Drive Systems and Lead Harmonic are mature players; Goldman Sachs raised Lead Harmonic’s high-spec robot market share forecast for 2025–2030 to 30%, keeps Harmonic Drive at 70%, and expects the two to converge to 50%/50% equilibrium post-2030, mainly because the growth of Chinese humanoid robot players benefits Lead Harmonic.

Actuator assemblies: Goldman Sachs sees technical adoption as more deterministic here. With car companies developing humanoid robots, actuator assembly companies are uniquely valuable due to their ability to manage complex supply chain relationships. Goldman Sachs predicts the market share distribution between Sanhua Holdings and Tuopu Group to shift from 50%/50% to 70%/30%, based on Sanhua's global dominance in Tesla EV thermal management module assemblies.

Planetary roller screws: The market pattern is still changing fast, with yield, production consistency, and capacity readiness uncertain. Schaeffler currently holds about 50% to 60% of the market share, but the long-term winner is not clear yet. For dexterous hands, the technical roadmap is also highly divergent, with mainstream solutions still competing.

Valuation & Trading Strategy: Asia-Pacific Discount Plus High Growth, Early-Cycle Opportunity Prominent

On valuation, the average P/E ratio of Goldman Sachs' Asia-Pacific robot basket is close to the U.S. basket (about 29–30x), but the Asia-Pacific median (22x) is much lower than the U.S. (28x), implying a typical 21% discount on Asian industrial/robot stocks versus U.S. peers. The Asia-Pacific basket’s median PEG ratio is 1.5x, lower than the U.S.'s 2.0x, reflecting that each unit of valuation in Asian robot/automation stocks carries higher profit growth potential, mainly benefiting from higher growth expectations for companies like Ningbo Tuopu, Sanhua Holdings, Hengli Hydraulic, and other Chinese/Korean automation companies.

In terms of capital flows, mutual funds have begun rotating toward robot-related supply chains, but overall positions are still in the early stages, focusing more on components than pure machines. Most flows are into Korean and Chinese auto parts, Chinese industrial automation/precision manufacturing, and some robot component suppliers. Passive fund rebalancing is causing visible rotation in Korea (Hyundai Mobis seeing inflows, Hyundai Motor outflows), and after liquidity adjustments, mid-cap stocks like BizLink and Hengli Hydraulic especially benefit.

On shipment forecasts, Goldman Sachs is more cautious than the market: it predicts global humanoid robot shipments of 76,000 units in 2027 and 502,000 units in 2032, lower than most market expectations. Goldman Sachs notes that the market may have already priced in 500,000 units by 2027, corresponding to a 40x exit P/E ratio. Upside valuation comes from core business plus humanoid robot option value; downside is if robot option value falls to zero, leaving only the core business for valuation support.

Goldman Sachs concludes that with structural demand accelerating and Asia still holding a clear valuation discount compared to Europe and the U.S. but stronger growth momentum, "this is an early-cycle opportunity to position before years of capital rotation fully enter the robotics ecosystem."

Risk Warning & DisclaimerThe market has risks, investment requires caution. This article does not constitute personal investment advice and does not consider individual users' particular investment goals, financial situations, or needs. Users should consider whether opinions, views, or conclusions in this article fit their specific circumstances. Invest accordingly at your own risk.