Physical AI Hits A Data Labeling Wall That Only Cash Can Fix
$10B+ raised in 2025. Physical AI robots still training on under 5,000 hours of real-world data. The bottleneck isn't compute—it's labels.
Why it matters
Physical AI funding is exploding, but the industry is hitting a critical infrastructure wall: real-world training data is scarce and expensive to label. This shifts capital allocation from model development to data collection and annotation—a fundamental constraint that will determine which companies survive the race.
The key facts
4 to know$10B+ raised in Physical AI in 2025
Robots training on under 5,000 hours of real-world data (critical constraint)
Data labeling emerging as capital-intensive bottleneck vs. compute-centric narrative
Funding trend shifting toward annotation infrastructure and synthetic data solutions
Go to the source
Forbes Innovationforbes.com
Publisher excerpt: Physical AI raised $10B+ in 2025, but robots still train on under 5,000 hours of real-world data. Who's funding the race to fix it.