Accelerating Optimus ramp-up, China humanoid robot volume growth, and a declining cost curve

Overview

This report, based on Nomura's advanced manufacturing field research and industry interviews on June 15-16, 2026, assesses the mass-production cycle for humanoid robots, the substitution of non-robot data collection, and the trend toward complete-unit contract manufacturing. The core conclusion is that Optimus capacity targets have been revised upward, China's humanoid robot shipments may rise further, and the cost curve is falling rapidly; however, industrial scenarios are still constrained by verification of accuracy, cycle time, and hardware costs.

Core views

First, Tesla Optimus is shifting from start-up production to scaled manufacturing, and 2026 supply chain order signals point to shipments of about 25k ± 10k units. Second, China's 2026 humanoid robot shipments may reach 40k-50k units, with volume growth driven mainly by government procurement and consumer demand rather than purely by data collection demand. Third, non-robot data collection is becoming mainstream and may begin to compress demand for real-robot data collection starting in 2027. Fourth, the share of complete-unit contract manufacturing may rise over the long term, with automotive supply chains and 3C manufacturers benefiting at different stages.

Analysis framework

The report combines company visits, supply chain order checks, capacity planning, downstream application structure, data collection costs, and BOM cost trends to form its judgments. Research targets include Unitree, Deep Robotics, Shenhao Technology, Mech-Mind Robotics, and Tesla supply chain companies.

Methodology notes

• Field researchAdvanced manufacturing field research

  Industry chain visits and management interviews

  By visiting complete-unit manufacturers, vision companies, and supply chain enterprises, the report verifies capacity ramp-up, BOM costs, application deployment, and outsourcing trends.

• Capacity and order verificationSupply chain order check

  Use supply chain signals to validate shipment expectations

  The report uses supply chain order signals to estimate Optimus 2026 shipments at about 25k ± 10k units, and cross-validates this against production line targets and industry capacity pacing.

• Data paradigm analysisMixed-data training

  Combination of non-robot data and real-robot data

  The report believes that teleoperation, universal manipulation interfaces, and first-person-view data collection are lower-cost and faster, and that future training data structure may shift toward about 90% non-robot data and 10% real-robot data.

• Cost curve analysisBOM and price curve

  Impact of scale, in-house production, and outsourcing on complete-unit costs

  The report assesses the downward industry cost curve through BOM costs, outsourcing ratios, and product price ranges, and notes that building an in-house supply chain can strengthen pricing power.

Asset mapping & comparison

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

• Tesla Optimus / TSLA US (Not rated)

  A global benchmark for scaled humanoid robot manufacturing and the core observation target for supply chain orders and upward capacity revisions.

  Strengths

  The Fremont production line target has been raised, the September 2026 weekly production target is about 1,000 units, and the long-term capacity plan offers significant scale potential.

  Weaknesses

  Overall production pacing among overseas non-China players still lags Chinese manufacturers, and Optimus's actual ramp-up still needs validation.

  Comparison

  Compared with overseas companies such as Figure AI and Boston Dynamics, Tesla's capacity target is more aggressive; but compared with leading Chinese manufacturers, its mass-production pace still needs to catch up.

  Risks

  If supply chain order conversion, production yield, software capabilities, and end demand come in below expectations, shipment pacing will be affected.

• China humanoid robot OEMs (Unitree, Deep Robotics, Shenhao Technology, Mech-Mind Robotics, etc.)

  The core industry participants behind the upward revision to China's 2026 shipments and the downward cost curve.

  Strengths

  Domestic manufacturers have advantages in shipment scale, low-priced products, government procurement, and supply chain costs; leading manufacturers are expected to ship about 10k-15k units each.

  Weaknesses

  Model capability, breadth of B-end scenarios, and replicability of industrial applications have not yet been fully validated.

  Comparison

  Chinese manufacturers are generally ahead of overseas non-China players in production pace, while tier-2 and tier-3 players rely more on outsourced manufacturing.

  Risks

  If the inflection in consumer demand, the pace of government procurement, accuracy in industrial scenarios, and hardware cost validation fall short of expectations, growth quality will be affected.

• Complete-unit contract manufacturing and automotive supply chain/3C manufacturers

  Beneficiaries of rising outsourcing ratios among tier-2 and tier-3 OEMs and the reshaping of labor division in complete-unit manufacturing.

  Strengths

  Automotive supply chain companies have end-to-end supply chain management and large-scale assembly experience, while 3C manufacturers have short-term brand and manufacturing recognition advantages.

  Weaknesses

  Capital expenditure for hardware supply chains is high, payback periods are long, and the outsourcing model still needs to solve for quality, delivery, and cost control.

  Comparison

  Over the long term, entrants from the automotive supply chain may be more competitive than pure 3C manufacturers, while in the short term 3C companies may find it easier to gain brand attention.

  Risks

  If OEMs shift toward self-built supply chains or industry demand slows, utilization and pricing power for contract manufacturing capacity may come under pressure.

• Data collection factories and non-robot data solutions

  A key link determining training data cost, speed, and the intensity of real-robot procurement.

  Strengths

  Non-robot data collection costs about 20% of real-robot collection and is faster, helping accelerate software iteration.

  Weaknesses

  Requirements for data quality, consistency, and cross-platform standardization are higher, and real-robot data still cannot be fully replaced.

  Comparison

  Compared with the CNY50-100mn investment required for a 1,000-unit-scale real-robot data factory, non-robot data solutions significantly reduce upfront capital pressure.

  Risks

  If non-robot data cannot be effectively transferred to real-world scenarios, model training results and commercial deployment may be constrained.

• 3D vision and industrial vision supply chain

  An important upstream capability supporting robot perception, customer diversification, and overseas expansion.

  Strengths

  The report mentions that a solid 3D vision business helps serve diversified customers and support overseas reach.

  Weaknesses

  Industrial deployment is still constrained by accuracy, cycle time, and model training costs.

  Comparison

  Compared with the single complete-unit link, vision capabilities can be reused across customers and application scenarios.

  Risks

  If validation progress in industrial scenarios is slower than expected, revenue elasticity for the vision supply chain may be delayed.

Key data

• Optimus Fremont capacity targetAbout 70k units annualized in 2027Retrofitted from the old Model S/X line, above the previous target of about 50k units.
• Optimus Austin second lineAbout 70k units in 2028The report says the long-term combined capacity target is moving toward 1.5mn units.
• Optimus September 2026 ramp-up targetAbout 1,000 units/weekReflects the phased weekly production target after the upward revision to capacity planning.
• Optimus 2026 shipment signalAbout 25k ± 10k unitsEstimated based on industry supply chain order signals.
• China 2026 humanoid robot shipment forecastAbout 40k-50k unitsDriven by government procurement for embodied intelligence bases and an inflection in consumer demand in 2H26.
• China downstream demand structureConsumer about 30%, performance/entertainment about 30%, government procurement about 20%, education about 15%, other commercial/industrial about 3-5%Government procurement mainly corresponds to data collection demand, while industrial applications remain bottlenecked by validation.
• Non-robot data collection costAbout 20% of the cost of real-robot data collectionIncludes methods such as teleoperation, universal manipulation interfaces, and first-person-view data.
• Mixed-data training structureAbout 90% non-robot data + 10% real-robot dataThe report expects that once this structure becomes standard, demand for real-robot data collection will decline further in 2027.
• Data collection factory investmentAbout CNY50-100mn for a large 1,000-unit-scale facilityPayback period is about 3-5 years, and the government remains the main sponsor.
• Humanoid robot price rangeFull-size CNY150k-300k, small-size CNY10k-100kPrices in 2026 have already fallen by more than 50% year over year, and the pace of price cuts is expected to narrow in 2027.

Impact & implications

The drivers of industry volume growth are shifting from simple hardware procurement to a more complex combination of capacity ramp-up, lower-cost data collection, software iteration, and supply chain restructuring. Non-robot data collection reduces early-stage procurement pressure for real-robot hardware, but raises requirements for data quality, consistency, and cross-platform standardization. Complete-unit outsourcing will shift more value to manufacturing partners with supply chain management and scaled assembly capabilities, while OEMs with self-built supply chains may still gain stronger pricing power through BOM cost advantages.

Risks

• Optimus capacity ramp-up and supply chain order conversion may fall short of expectations.
• China's consumer demand inflection and government procurement pace may come in below expectations.
• Non-robot data quality, consistency, or cross-platform transfer effectiveness may be insufficient.
• Industrial scenarios may fail validation in accuracy, cycle time, and hardware cost.
• Prices may decline faster than expected, compressing margins for OEMs and the supply chain.
• Higher outsourcing ratios may bring risks in quality, delivery, cost control, and supply chain traceability.

What to watch

• Whether Optimus reaches the ramp-up target of about 1,000 units/week in September 2026.
• Whether Tesla Optimus actual shipments in 2026 approach the 25k ± 10k unit range.
• Whether leading Chinese manufacturers achieve shipments of 10k-15k units each.
• Whether consumer demand truly inflects upward after the launch of low-priced products in 2H26.
• The ordering pace for government embodied intelligence bases and data collection factories.
• The share of non-robot data in training datasets, quality standards, and transfer effectiveness to real-world scenarios.
• The ratio of complete-unit contract manufacturing, the share of automotive supply chain companies, and the pace of BOM cost reduction.