The Physical AI Data Supply Chain: A Redistribution of Profits
The advancement of physical artificial intelligence relies heavily on high-quality first-person video data. However, a recent robotic research report from Nomura highlights a persistent decline in the market price for this data, fundamentally reshaping profit distribution across the industry. A clear trend has emerged: value is rapidly consolidating upstream, while traditional business models dependent on scaling manual collection are under significant strain.
A Squeezed Profit Chain: Diverging Fortunes for Tier-1 and Tier-2
Nomura's industry research indicates that Tier-1 suppliers, which interface directly with major AI model labs, occupy a relatively advantageous position. The report estimates that within a low-cost raw data supply chain in Southeast Asia, Tier-1 suppliers retain approximately $5 per hour of data flow, while frontline collection workers earn only $1 to $2 per hour. This allows Tier-1 suppliers to maintain profit margins of 10% to 20%.
This profitability, however, is not guaranteed. It heavily depends on the ability to resell data to multiple clients, a practice often limited as most customized datasets are sold only once. This places a premium on Tier-1 suppliers' ability to continuously secure new clients.
Tier-2 Operators at the Brink: The Scale Model Falters
The industry's turmoil is most acute for Tier-2 data collection operators. Nomura's findings show that since mid-2026, some Tier-2 operators have reached breakeven points or even slipped into losses. This suggests that in an environment of broadly falling data prices, business models relying solely on scaling up manual collection through more labor are becoming unsustainable.
Typically serving as subcontractors to Tier-1 suppliers rather than end-client AI labs, Tier-2 operators bear the brunt of cost pressures through the price transmission mechanism, yet possess the least bargaining power.
Automated Annotation: A Disruptive Force in Cost Structure
Beyond market prices, technological evolution is profoundly altering the industry's cost foundation. The adoption of automated annotation tools is drastically reducing reliance on human labor in data processing. Nomura's report cites a telling case: after implementing a machine annotation system in August 2026, one Tier-1 supplier reduced its annotation team from over 1,000 personnel to around 50.
This efficiency revolution signifies a fundamental shift in the competitive landscape for physical AI data suppliers. Nomura's analysis suggests future advantage will stem not from labor scale, but from three core pillars: secure partnerships with leading AI labs, highly efficient data processing pipelines, and rigorous quality validation systems.
Looking Ahead: Specialization and Efficiency as Differentiators
In summary, the physical AI data industry is undergoing a consolidation. Profits are aggregating towards Tier-1 suppliers with technological, client, and capital advantages, while margins in traditional labor-intensive collection are being relentlessly compressed. For industry players, adaptation is no longer optional but essential for survival. Investing in automation, enhancing data value-add, and specializing in vertical niches may define the path forward.