Commercialization of Humanoid Robots Accelerates; VLA Models and Data Become Critical

Overview

Goldman Sachs published its mid-year review following participation in the Asia Tech Conference in Hong Kong (May 18–22, 2026) and on-site visits to 14 Chinese humanoid robotics and automation companies in Shenzhen and Beijing. The report notes tangible progress in the integration of VLA/VTLA models with world models, scalable data collection infrastructure, and commercialization efforts, indicating a bright long-term investment outlook. However, it also cautions that current commercialization remains largely at the proof-of-concept (POC) and small-batch trial stage, requiring patience until large-scale deployment materializes between 2027–2029—contingent on continued improvements in model quality and cost reduction.

Core views

**Model Level: Evolution Toward Action-Oriented Multimodal Architectures** Discussions have moved beyond standalone VLA frameworks toward rapid integration of VLA with world models (e.g., VTLA incorporating tactile modalities). VLA/VTLA handles policy and action generation, while world models enhance planning robustness under real-world uncertainty through next-state prediction and pre-action validation. Model scales have grown substantially, with multiple companies referencing larger stacks in the 40B–80B parameter range, though consistent deployment-grade quality still requires several more iterations. **Data Level: High-Quality Real-World Data Remains the Core Bottleneck** Discussions have shifted from generic 'data recipes' to scalable architectures capable of generating high-quality data. Human-centric collection methods—via wearables, VR, and deployed device telemetry—are now preferred. Investment trends include government-backed, large-scale data factories (e.g., PaXini operating five nationwide) and companies building distributed collection loops via deployed systems or VR. Multiple firms expect data-related revenue contributions to rise notably in 2026, with sustained strong demand for government data factories. **Commercialization Level: Scope Broadens but Remains POC-Dominated** Commercial applications are expanding into standardized or semi-structured industrial scenarios such as material handling and logistics sorting. Industrial deployment follows a multi-step path: 'Proof of Concept (POC, 3–6 months) → Small-Batch Testing (<50 units) → Validation (~12 months) → Pilot Deployment (50–100 units).' Large-scale deployment is projected for 2027–2029, contingent on achieving deployment-grade model quality and accumulating tens of millions of hours of high-quality data. Near-term opportunities include sorting, material handling, pick-and-place, and inspection/testing. **Cost Level: Scale Drives Cost Reduction; Wheeled + 2/3-Finger Grippers Are Current Rational Choices** Cost declines stem mainly from economies of scale and full-stack in-house control. Multiple companies note that wheeled bases with two- or three-finger grippers can address 70%–90% of industrial applications, making them the rational near-term choice; bipedal platforms with five-finger dexterous hands remain a long-term aspiration. UBTECH’s BOM cost has fallen from ~RMB 400k in early 2025 to ~RMB 250k by end-2025, currently slightly above RMB 200k, with a long-term target of ~RMB 100k (around 2027).

Analysis framework

Goldman Sachs gathered primary insights through on-site research—attending the Asia Tech Conference and visiting 14 robotics companies in Shenzhen and Beijing. Its analytical approach first outlines key industry-wide trends (model evolution, data bottlenecks, commercialization progress, cost-reduction pathways), then details each company’s specific advancements, data strategies, and cost-control measures. Finally, it provides a comprehensive investment thesis, valuation, and risk analysis for the only covered company (Estun Automation). The overall logic flows from macro industry trends to micro company validation, using company cases to reinforce sector-level conclusions.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply-demand framework

  Supply-demand framework for the humanoid robotics industry

  The report implicitly applies a supply-demand framework: the industry currently faces strong demand (from vast potential substitution opportunities in industrial, logistics, and service scenarios) but constrained supply (due to limitations in high-quality models, massive real-world datasets, and mature low-cost hardware). Commercialization progress hinges on gradually overcoming supply-side bottlenecks in data, models, and costs.

• Cycle and Sentiment FrameworkInflection Point Analysis

  Inflection point analysis for humanoid robotics sentiment

  Based on field research, Goldman Sachs judges the industry to be at an inflection point transitioning from concept validation to early commercialization. Large-scale deployment is expected in 2027–2029, effectively forecasting the onset of an upcycle. The report highlights accelerated investments in data and models across multiple companies as leading indicators of impending industry momentum.

• Competition and Strategy FrameworkProduct life cycle

  Product lifecycle of humanoid robots

  The report segments commercialization into distinct product lifecycle stages: Proof of Concept (POC), small-batch testing, pilot deployment, and large-scale deployment—specifying duration and order volumes for each—to help readers assess the industry’s current lifecycle position.

• Industry/Industrial Analysis FrameworkVolume-Price Breakdown

  Volume-price breakdown for humanoid robots

  The report quantifies market size and revenue potential by breaking down shipment volumes (e.g., Dobot’s 200–300 humanoid units in 2025; UBTech’s ~10,000-unit guidance for 2026) and average selling prices (ASP, e.g., Dobot’s humanoid ASP of RMB 200k–500k per unit).

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Estun Automation (002747.SZ / 2715.HK)

  Domestic industrial robotics leader rapidly expanding into automation and humanoid ecosystems. Goldman Sachs assigns a Sell rating for A-shares and Neutral for H-shares.

  Strengths

  No. 1 domestic market share in industrial robots; in-house production of servos and controllers; benefits from localization and automation trends; credible monetization path in physical AI, with AI projects expected to contribute 10% of company revenue by 2030.

  Weaknesses

  Primary end markets (automotive and automotive electronics) face intense price competition; inconsistent profitability track record.

  Comparison

  As the domestic leader, competes against foreign brands like Fanuc and KUKA while also facing pricing pressure from domestic peers.

  Risks

  Slower-than-expected robotics market share growth; weaker-than-expected margin trends; slower-than-expected humanoid robot development.

• Dobot (2432.HK, Not Covered)

  Collaborative robotics platform extended to humanoid robots; data-related revenue grew 4x in 2025.

  Strengths

  Reusable collaborative robotics platform and upper-limb capabilities; complete product portfolio (robotic arms, quadrupeds, wheeled/bipedal humanoids); shipped 200–300 humanoid units in 2025 at ASP of RMB 200k–500k with 45% gross margin.

  Weaknesses

  Humanoid robot sales represent a small portion (~4%) of total company revenue.

• UBTech (9880.HK, Not Covered)

  Humanoid robotics leader; 2026 delivery guidance of ~10,000 units; strong demand for data factories.

  Strengths

  Leading commercialization progress with high shipment volume (~10,000 units in 2025 split evenly between industrial and data factory segments); continuous BOM cost reduction; high product quality and market recognition.

  Weaknesses

  Data factory model relies heavily on government orders; commercial/service robots currently serve more as capability demonstrations, with unproven profitability.

• Geek+ (2590.HK, Not Covered)

  Logistics robotics company; ~30% YoY growth is sustainable; market remains a blue ocean.

  Strengths

  Deepening relationships with existing clients drive ~80% of growth; market perceived as underserved blue ocean; ~30% growth rate is sustainable.

  Weaknesses

  Growth constrained by internal resource allocation (sales, delivery, resource distribution), not by lack of demand.

• One Robotics (6600.HK, Not Covered)

  General-purpose humanoid robotics company; Onero targets $10k ASP; scaled deployment planned for second half of 2026.

  Strengths

  Clear integration roadmap for VLA/VTLA + world models; dual-track strategy in service scenarios (retail, elderly care) and consumer products (tennis robot Acemate launching on Amazon in 2027); software algorithms compensate for hardware precision limits to enable cost reduction.

  Weaknesses

  Onero still in iteration; BOM of ~$3k remains a barrier for household adoption.

Key data

• Expected timeframe for large-scale humanoid robot deployment2027–2029Industry consensus; requires tens of millions of hours of high-quality data and deployment-grade models.
• Parameter range for VLA/VTLA models40B–80BMentioned by multiple companies; a significant increase from previous single-digit billions.
• Industrial application coverage by wheeled robots + 2/3-finger grippers70%–90%Current rational choice; bipedal + five-finger dexterous hands are long-term direction.
• UBTech BOM cost reduction trajectory~Rmb400k -> ~Rmb250k -> ~Rmb200k -> ~Rmb100kEarly 2025 → End 2025 → Current → Long-term target by 2027.
• Dobot humanoid robot shipments in 2025200–300 units2026 guidance: 300–500 units (wheeled + quadruped + humanoid).
• UBTech 2026 robot delivery guidance~10,000 units5,000 industrial + 5,000 commercial/residential.
• Number of PaXini data factories5Operated nationwide with government support.
• Industrial deployment cycle from POC to pilotApprox. 12–18 monthsPOC 3–6 months → small-batch test → ~12-month validation → pilot deployment.
• Estun A/H share target pricesRmb14.9 / HK$11.8Based on 2030E P/E of 35x/25x, discounted to 2027E at 11.5% cost of equity.

Impact & implications

Goldman Sachs believes multiple companies across the humanoid robotics value chain—spanning models, hardware, data, and integration—are accelerating toward commercialization, which will profoundly impact industrial automation and service robotics markets. In the near term, companies offering standardized solutions for industrial scenarios (e.g., wheeled robots with grippers) are likely to secure early POC and small-batch orders. In the long run, firms with capabilities in high-quality real-world data collection and full-stack in-house hardware development will hold competitive advantages during the large-scale deployment phase (2027–2029). Data-related revenue streams—such as data factories and dataset sales—will emerge as a new growth vector.

Risks

• High-quality real-world data remains the primary bottleneck; data collection speed and scale may fall short of expectations.
• The complex and time-consuming conversion path from POC to large-scale deployment (~12–18 months) carries delay risks.
• Insufficient model capabilities—especially in cross-scenario generalization and post-training efficiency—could hinder commercialization timelines.
• Slower-than-expected cost declines—particularly in core BOM components like actuators/reducers and structural parts—could impede large-scale deployment.
• Lack of industry-wide standards (e.g., technical paths and finger counts for five-finger dexterous hands) introduces strategic uncertainty.
• Intensifying domestic competition, worsening price wars, and increasing participant numbers could pressure margins.
• For Estun Automation, intense price competition in its primary end markets (automotive/automotive electronics) and unstable profitability history pose specific risks.

What to watch

• Progress in VLA/VTLA + world model iterations and when deployment-grade quality is achieved across companies.
• Investment and output from scalable data collection infrastructure (data factories vs. distributed deployment loops).
• Conversion rates and cycles for industrial validation (POC → small-batch → pilot), especially progress with benchmark clients like CATL, JD, and Bosch.
• Pace of cost reduction: whether BOM reaches the RMB 100k–200k threshold, particularly for actuators/reducers and structural components.
• Changes in humanoid robot shipment volumes and ASPs, watching for a shift from conceptual sales to real industrial deployments.
• Estun Automation’s domestic market share growth and margin trends.