Goldman Sachs has sharply raised its humanoid robot forecasts, projecting global shipments of approximately 890,000 units in 2030 and 6.5 million units in 2035, equivalent to a market opportunity of roughly $138 billion. The estimates, published in an 80-page Physical AI report authored by Eric Sheridan, Goldman Sachs' internet and e-commerce equity analyst and managing director, compare with previous forecasts of 256,000 units for 2030 and 1.4 million for 2035. Shipments are projected to rise from 75,000 units in 2026. Logistics and warehousing are expected to lead scaled adoption, with Amazon and Walmart positioned as early beneficiaries, followed by automotive manufacturing. The report estimates meaningful savings from fulfillment automation and says Tesla plans to begin Optimus production in 2026 for internal use, with external sales starting in 2027 and a long-term target of 1 million units annually around 2030. Each robot could represent $3,000 to $6,000 or more in semiconductor demand, including high-performance computing, analog and mixed-signal chips, and edge storage. Existing assumptions also put memory value at $600 to $800 per system, based on 128GB of DRAM and 1TB of NAND, with some HBM potentially required. Falling bill-of-materials costs and average selling prices, together with the adoption of software-defined virtual PLCs, could accelerate commercialization. A Goldman Sachs survey found that about 40% of executives expect general-purpose robots to automate at least 10% of workflows within three to five years, although current humanoid technology can directly handle only about 2% of non-robotized automotive processes.