The cost curve is about to cross the line where adoption goes from boardroom experiment to operational default.
Bank of America published its updated Physical AI forecast this month. The headline number: material costs for a humanoid robot will fall from $35,000 to between $13,000 and $17,000 by 2035. Manufacturing costs are declining 40% year on year, roughly double the rate analysts projected 18 months ago. BofA expects annual humanoid shipments to reach 1.2 million units by 2030, up from 90,000 this year.
These aren't just hand-waving speculations. The industrial robot market already ships 542,000 units per year, with 4.66 million operational worldwide. China alone accounts for 54% of global deployments. The IFR expects the market to hit 700,000 annual installations by 2028. Humanoid robots are the next wave on top of this existing base.
The cost curve changes everything
A $13,000 robot reshapes the economics of labour-intensive industries. In Australia, the average warehouse worker earns between $60,000 and $70,000 per year. A robot at a quarter of that annual cost, operating around the clock, improving with each software update, represents a permanent shift in the build-vs-hire equation. This isn't a Silicon Valley abstraction. It applies directly to Australian logistics, food processing, aged care, mining support, and agriculture.
The adoption curve for industrial technology follows a predictable pattern. Costs decline, early adopters demonstrate ROI, and then the mid-market moves in bulk. We're in the early-adopter phase now. BofA's cost projections suggest the mid-market inflection point arrives somewhere between 2030 and 2035.
The chassis is a commodity. The intelligence layer is the margin.
Consensus opinion treats humanoid robots as a hardware market. We think that's the wrong framing.
Robot hardware will follow the same trajectory as smartphones: rapid cost deflation, margin compression, and eventual commoditisation. The manufacturers competing on bill-of-materials cost will face the same squeeze that hit Android handset makers in the 2010s.
The value sits one layer up. Companies building three things will capture disproportionate returns:
First, sim-to-real training pipelines. The software that lets you train a robot in simulation and deploy it in a physical environment with minimal gap. ABB and NVIDIA's RobotStudio HyperReality partnership (announced earlier this month, claiming 99% sim-to-real correlation) is a signal of where this is heading.
Second, vertical skill libraries. A general-purpose humanoid is useful. A humanoid with a pre-trained skill set for pick-and-place in cold-chain logistics, or patient transfer in aged care, or weld inspection in shipbuilding, is worth 10x more. The companies encoding domain expertise into deployable skills own the margin.
Third, fleet orchestration. Managing one robot is a demo. Managing 500 across three warehouses, with real-time task allocation, predictive maintenance, and continuous learning from fleet-wide data, is a business. This is the enterprise software layer that turns cheap hardware into expensive capability.
Why this fits our thesis
Physical AI is one of Unlock Capital's three investment pillars (alongside Sovereign Infrastructure and Bio AI) because it sits at the intersection of AI capability and physical-world impact. Our investment framework prioritises what we call Essential AI: technology that operates at the substrate level, where bits meet atoms. Robots that pick, weld, and carry are essential infrastructure.
The $13,000 price point is a milestone, not a destination. What matters is the rate of decline and the compounding effect of software improvement on top of cheaper hardware. The companies we're watching build the intelligence that makes robots worth deploying.
The question for investors
Who captures the margin when the hardware costs race to the bottom? That's the Physical AI investment question for the next decade. The answer may well sit in the orchestration layer: training, skills, and fleet management.