Global physical AI and humanoid robotics, with a focus on logistics Report Interpretation

Overview

Goldman Sachs presents a global physical-AI and humanoid-robotics deep dive centered on logistics and warehouse use cases. Its core conclusion is that improving AI models, growing real-world data, larger-scale production and falling hardware costs support substantially faster humanoid adoption, initially in structured commercial environments rather than broad consumer deployment.

Core views

Goldman Sachs raises its global humanoid forecast to approximately 75,000 units in 2026, 890,000 in 2030 and 6.5 million in 2035, compared with its prior forecasts of 51,000, 256,000 and 1.4 million units. The firm estimates a $138 billion market by 2035. It attributes the revision to progress in physical AI, open-source tools, venture funding, government support and company plans moving from prototypes toward commercial production. The report expects deployment to create a data flywheel: operating robots generate real-world training data, improving models and enabling broader task capability over time. The report argues that logistics and warehouses are the most practical early deployment environments because tasks are varied enough to challenge fixed automation but remain within structured, manageable settings. Automation penetration in warehousing remains low relative to manufacturing: the report cites nearly 1,800 robots per 10,000 employees in automotive versus 250 across non-automotive manufacturing. Rising labor costs, more SKUs, faster fulfillment requirements and greater inventory-handling complexity make productivity investment structural rather than cyclical. Its survey of more than 80 companies found that roughly 30-35% were seriously evaluating or piloting general-purpose robots or humanoids, another roughly 35% had limited deployments, and about 40% expected at least 10% of workflows to be automated or materially enhanced over the next three to five years. Respondents identified technology reliability and integration complexity as the principal barriers. For e-commerce, the report sees automation as both a cost and competitive lever. Amazon had deployed more than 1 million robots across over 300 facilities as of June 2025, while its 12th-generation fulfillment-center design reportedly accelerated fulfillment by 25% and increased same- or next-day item availability by 25%. Goldman Sachs' upside scenario assumes 5.6% cost-to-serve leverage from 2026E to 2030E, equivalent to an approximately 11% per-unit reduction by 2030. It estimates about $72 billion in cumulative savings, with roughly 80% in operating expenses and 20% in shipping COGS, producing a roughly 240 bp consolidated EBIT tailwind and a roughly 400 bp retail EBIT-margin tailwind. The report frames these gains as a way to improve unit economics, delivery speed and the gap versus smaller e-commerce competitors. Walmart is presented as another large-scale proof point. By August 2026, 3,100 US stores were served by some automated freight and more than 50% of e-commerce fulfillment volume passed through automated facilities. Goldman Sachs notes that automated fulfillment centers can manage twice the capacity and throughput of an existing center with the same footprint, while Walmart regards itself as only halfway through its automation progress. The report estimates potential markdown-related savings of $4.0 billion to $23.7 billion under its sensitivity analysis and a further 2.4%-12.9% improvement in delivery cost per order. It argues that lower costs, better inventory visibility and excess fulfillment capacity can reinforce Walmart's marketplace and higher-margin service businesses. Automotive is another early commercial setting, especially for parts sorting, transportation, inspection and final assembly tasks that remain too variable for traditional automation. Goldman Sachs expects the economic crossover to improve as robot costs fall. It forecasts average humanoid prices declining from above $40,000 in 2025 to about $21,000 in 2035, with high-spec units falling from about $150,000 to around $50,000. For factory-use humanoids, it estimates costs equivalent to 2.8 years of labor in 2026 and 1.9 years in 2027 where the technology is ready for the task. In an illustrative automotive scenario using robot prices of $20,000-$60,000 and adoption rates of 10%-50%, the report estimates gross operating-margin expansion of roughly 1%-6%; the net benefit could be smaller if savings are passed through in lower vehicle prices. It stresses that only about 2% of auto processes are currently suitable for humanoids, leaving complex assembly, wiring and inspection dependent on further advances in robot intelligence and hands. The report expects production scale and design optimization to lower average BOM costs by about 7% annually and ASPs by about 6% annually. It identifies dexterous hands, hardware durability, thermal management, flexible wiring, control latency and the sim-to-real gap as important unresolved constraints. The staged adoption path starts in fenced auto factories and large logistics hubs, progresses to general manufacturing and warehouses as unit costs reach roughly $20,000-$30,000 and safety standards mature, and only later extends to services, care, infrastructure and homes. Humanoids also affect industrial-automation architecture. Goldman Sachs expects virtual PLC adoption to grow roughly 20%-30% annually from an approximately $0.5 billion base in 2025, compared with an estimated $13 billion physical-PLC market growing at about 4%. Virtual PLCs can centrally coordinate mobile robot fleets on standard servers and reduce dependence on proprietary controllers, racks and I/O modules, potentially weakening incumbent hardware-software lock-in. However, the report does not expect physical PLCs to disappear quickly: deterministic latency, safety, reliability, legacy-plant migration costs and IT/security skills requirements remain material obstacles. The business model may shift from upfront hardware sales toward subscriptions, usage fees and managed services, although vendors also face risks from consolidation of vPLC instances and more outcome-based pricing. Semiconductors and precision components are described as major enablers. Goldman Sachs estimates total semiconductor content of about $2,950-$6,000+ per humanoid: $750-$1,050+ for analog and mixed-signal devices, $600-$800+ for memory, and $1,600-$4,150+ for non-memory digital content. Compute modules alone account for about $1,500-$4,000+, while analog opportunities include $350-$500 in power semiconductors, $150-$200 in communications ICs and FPGA, $150-$200 in optical components and $50-$100 in other sensors. The report also identifies a growing opportunity for actuators, motors, reducers, bearings, batteries, passive components, connectors and harnesses as robot dexterity and motion complexity rise. On software and infrastructure, Goldman Sachs argues that world models and vision-language-action systems are becoming central to physical AI because text-trained LLMs alone cannot solve real-world manipulation. The report sees proprietary deployment data as a long-term moat, particularly for tasks involving deformable objects, tools and contact-rich interaction that are difficult to simulate. It cites an estimate that one robot can generate 64 TB of camera data in one day—about 200 times a 120 TB frontier-LLM training corpus—and argues that world-model development could create an additional long-term source of token, storage, bandwidth and GPU demand. Microsoft, Oracle and CoreWeave are identified as positioned to serve these workloads. The report highlights a broad group of beneficiaries rather than a single company. It views Amazon and Walmart as users whose automation could improve margins and customer experience; Nvidia as a full-stack AI-compute, simulation and robotics-platform enabler; and suppliers such as Jabil, Flex, TE Connectivity, Amphenol, Hyundai Mobis, HL Mando, Robotis, XPeng, Hesai, Xiaomi, Inovance, Sanhua, Shuanghuan, Daifuku, MinebeaMitsumi, Renesas and Regal Rexnord as potential participants across manufacturing, actuators, sensing, control, warehouse systems and components. Individual investment views and targets remain company-specific rather than a single report-wide rating.

Analysis framework

Goldman Sachs combines a revised top-down market forecast with company commentary, a survey of more than 80 companies, technology and regional case studies, and bottom-up operating scenarios. It tests adoption economics through warehouse cost-to-serve, retail automation, automotive labor-cost and robot-price sensitivities, then maps implications across PLCs, semiconductors, components, software and named public companies.

Methodology notes

• Industry AnalysisSupply-demand framework

  Humanoid market sizing based on projected shipments, specifications, ASPs and BOM costs through 2035.

  The report links rising commercial deployment and falling unit costs to its revised unit and market-value forecasts.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Physical-AI value-chain mapping from software and compute to robots, components, automation systems and end users.

  Goldman Sachs traces how wider humanoid deployment could affect retailers, industrial-control vendors, semiconductor suppliers and precision-component makers.

• Corporate Fundamentals and FinanceFree cash flow analysis

  Operating-cost and margin sensitivity analysis for automation adoption.

  The report estimates cost savings and EBIT-margin effects for Amazon, Walmart and automotive manufacturing under specified adoption and cost assumptions.

• Valuation methodsSOTP (Sum-of-the-Parts) Valuation

  Company-specific sum-of-the-parts valuation for Amazon.

  Amazon's stated 12-month price target applies EV/EBIT to North America and AWS and EV/Sales to International using NTM+1 estimates.

• Valuation methodsDCF (Discounted Cash Flow)

  Discounted cash-flow scenario analysis for selected companies and emerging humanoid businesses.

  The report uses discount rates, terminal-growth assumptions or discounted long-dated valuation inputs for Tesla, XPeng and several component suppliers.

Asset mapping & comparison

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

• Amazon (AMZN)

  Automation user and logistics beneficiary

  Strengths

  More than 1 million deployed robots, broad fulfillment network and scope for system-level automation.

  Weaknesses

  Low-ASP essentials can have poor unit economics before further cost-to-serve gains.

  Comparison

  The report views Amazon's automation scale as a differentiator versus smaller e-commerce competitors.

  Risks

  Competition, scaling high-margin businesses, investment-driven margin pressure, regulation and macroeconomic volatility.

• Walmart (WMT)

  Automation user and retail beneficiary

  Strengths

  Automated freight serves 3,100 US stores and more than half of e-commerce fulfillment volume is automated.

  Weaknesses

  Automation progress is incomplete and returns may diminish as e-commerce scales.

  Comparison

  Automation can widen the price and service gap versus traditional brick-and-mortar peers.

  Risks

  Economic slowdown, pricing competition, wage and transportation costs, tariffs and macro/FX volatility.

• Nvidia (NVDA)

  Physical-AI compute, simulation and robotics-platform enabler

  Strengths

  Isaac GR00T, Omniverse and Jetson position it across models, simulation and edge computing.

  Comparison

  The report sees Nvidia as a foundational full-stack enabler rather than only a semiconductor supplier.

  Risks

  Slower AI-infrastructure spending, competitive share or margin erosion, and supply constraints.

• Tesla (TSLA)

  Humanoid developer and potential scaled manufacturer

  Strengths

  Hardware, AI-training, manufacturing and internal-factory data advantages could lower Optimus costs by several thousand dollars per robot.

  Weaknesses

  Management expects a long, initially flat S-curve ramp; early units are intended for data collection and training.

  Comparison

  Tesla and Hyundai are identified as among the most actively engaged automakers in humanoids.

  Risks

  Execution and competitive risks in physical AI, vehicle pricing, EV demand, tariffs, product delays and vertical-integration risks.

• Hyundai Mobis (012330.KS)

  Actuator supplier to Boston Dynamics

  Strengths

  Sole actuator supplier for Boston Dynamics with targeted 350,000-unit annual capacity by 2028.

  Weaknesses

  Humanoid execution remains dependent on Boston Dynamics' ramp.

  Comparison

  The report sees South Korea as a critical actuator supply-chain partner for non-Chinese developers.

  Risks

  Cost-inflation margin pressure and slower humanoid or autonomous-driving execution.

• Daifuku (6383.T)

  Warehouse-automation system supplier

  Strengths

  Leading material-handling position and growth that has outpaced factory-automation peers.

  Comparison

  Goldman Sachs identifies Daifuku as a major potential beneficiary because humanoids are likely to operate within underlying warehouse-automation systems.

  Risks

  Slower investment from technology, e-commerce and automobile industries.

• Regal Rexnord (RRX)

  Precision-motion and integrated-actuation supplier

  Strengths

  Can combine actuator systems, frameless motors, brakes and bearings across 30-50 motion axes per robot.

  Weaknesses

  Humanoid commercial traction remains early, with approximately $40 million of orders at FY25-end.

  Comparison

  Its multi-component offering could support higher content per robot than a narrower component supplier.

  Risks

  Longer destocking, weaker discrete trends, price/cost headwinds and dilutive M&A.

Key data

• Global humanoid shipments75,000 in 2026E; 890,000 in 2030E; 6.5 million in 2035ERevised from 51,000, 256,000 and 1.4 million, respectively.
• Global humanoid market$138 billion in 2035EGoldman Sachs' revised market opportunity estimate.
• Average humanoid cost trendBOM down about 7% annually; ASP down about 6% annuallyForecast reflects scale and design optimization.
• Amazon automation scenario~$72 billion cumulative savings and ~240 bp consolidated EBIT tailwind by 2030Based on ~5.6% total cost-to-serve leverage from 2026E to 2030E.
• Humanoid semiconductor content$2,950-$6,000+ per humanoidIncludes analog/mixed signal, memory and non-memory digital content.
• Survey adoption expectations~40% of respondents expect at least 10% of workflows to be automated or materially enhanced in 3-5 yearsSurvey covered more than 80 global companies.
• Virtual PLC market~20%-30% annual growth from a ~$0.5 billion 2025 baseCompared with a ~$13 billion traditional PLC market expected to grow at roughly 4%.

Impact & implications

The report sees humanoids as an emerging productivity platform whose early commercial value is most evident in warehouses and auto logistics. It expects the opportunity to spread across hardware, software, automation and component ecosystems as robots become cheaper and more capable, while noting that practical deployment remains contingent on reliable real-world performance, integration and safety.

Risks

• Humanoid deployment may be delayed by reliability limits, integration complexity, safety requirements and insufficient real-world training data.
• Dexterous hands, sim-to-real generalization, hardware durability, thermal management and control latency remain technical bottlenecks.
• Virtual PLC adoption faces latency, functional-safety, reliability, legacy-plant migration and skills-gap challenges.
• Cost savings from automation may be partly passed to customers through lower prices, reducing net margin benefits.
• Competition and vertical integration could reshape the supply chain and limit individual suppliers' economics.

What to watch

• Whether pilots in warehouses, logistics hubs and auto factories convert into scaled commercial deployments.
• The pace of BOM and ASP declines as humanoid production volumes rise.
• Progress in real-world manipulation, dexterous hands, safety certification and sim-to-real performance.
• Evidence that Amazon and Walmart continue reducing fulfillment and delivery costs through automation.
• The shift from physical PLCs toward vPLCs and the ability of incumbents to preserve monetization and customer relationships.
• Growth in world-model training workloads and associated demand for GPUs, storage and high-performance networking.