The Three Billion Humanoid Milestone: Inside Bank of America’s Long-Term Physical AI Forecast

The robotics industry has spent decades in a state of incremental progress, but a new analysis suggests we have reached a definitive turning point. According to a report released this week by the Bank of America Institute, the convergence of generative AI and declining hardware costs is setting the stage for a "robot population boom" that could see three billion humanoid units in operation by 2060.

The report, titled Physical AI, part 2: Humanoid robots, suggests that humanoids will eventually surpass cars on a per-capita basis. While the immediate focus remains on industrial and automotive deployments, the analysis projects that household assistants will ultimately dominate the market, accounting for 62% of the total population by 2060.

---

Defining the "Mind, Body, and Soul"

Bank of America frames the modern humanoid as a synthesis of three core layers that allow a machine to move from a controlled factory floor into the "messy reality" of human environments:

• The Mind (AI System): Powered by breakthroughs in Large Language Models (LLMs), this layer handles high-level cognition, task planning, and human interaction.
• The Body (Hardware): This includes the structural materials, energy systems, and the "senses"—vision and tactile systems.
• The Soul (Motion Control): The internal coordinator responsible for the "physics" of the robot, managing body balancing and the execution of movement.

This architectural shift is what enables "embodied intelligence," where a robot doesn't just follow a script but learns directly from the world. We are, of course, seeing this transition in the wild, such as Physical Intelligence’s use of Multi-Scale Embodied Memory to help robots remember their mistakes and adjust their strategies in real-time.

---

The Economics of Adoption

The report identifies five primary drivers accelerating this transition: AI maturity, hardware economics, industrial overlap (leveraging EV supply chains), demographics (labor shortages), and deployment readiness.

Perhaps most critical is the rapid decline in the Bill of Materials (BOM). While pilot-stage humanoids currently cost between $90,000 and $100,000 to produce, Bank of America projects that standardizing specifications will cut these costs in half by 2030. In China, where manufacturers like Unitree are scaling toward 20,000 units a year, BOM costs are expected to drop below $17,000.

While this $17,000 projection for 2030 might seem high compared to existing entry-level platforms like the $1,445 Noetix Bumi or the $15,000 Asimov v1 kit, the report focuses on "full-spec" industrial machines. For these high-performance units, reaching such a price point would still represent a massive 50% reduction from 2025 levels, driven by the same scale effects that transformed the EV industry. These projections align with recent moves by startups like Sunday Robotics, which recently raised $165 million specifically to move away from "viral demos" and toward a $10,000 consumer appliance.

The BOM Breakdown (Projected 2030)

---

The "Phase One" Reality Check

Despite the bullish long-term forecast, the report acknowledges that the industry is still in a grueling "Phase One" of proving reliability. Current humanoids remain too fragile for 24/7 industrial environments, and the "utility gap" between a lab demo and a productive worker remains significant.

We are seeing various strategies to bridge this gap:

• Morphological Flexibility: Boston Dynamics’ production Atlas uses an "alien" aesthetic with 360-degree joints to prioritize utility over human mimicry.
• Specialized "Internships": Companies like Xiaomi are reporting 90% success rates in highly specific factory tasks, such as nut installation, rather than trying to solve general-purpose labor all at once.
• Tactile Focus: Sharpa Robotics is pushing the boundaries of bimanual dexterity, recently demonstrating an apple-peeling milestone using vision-based tactile hands.

As BofA notes, the ultimate ceiling on capability is the "data bottleneck"—the difficulty of collecting enough high-quality manipulation data to replicate human-level improvisation. Whether through Skill Capture Gloves or shared-autonomy copilots, the race to feed the "Physical AI" brain is now the industry's primary mission.