Analog optical computers demonstrate promise on real-world mortgage data but face architectural limitations rather than hardware constraints

Researchers tested an analog optical computer (AOC) on 5.84 million U.S. mortgage records from HMDA, achieving 94.6% balanced accuracy with only 1,024 optical parameters versus 97.9% for XGBoost

The 3.3 percentage-point accuracy gap persists even when expanding the optical core from 16 to 48 channels, indicating fundamental architectural limitations rather than hardware inadequacy

A shared 127-bit binary encoding constraint reduced all model performance to 89.4-89.6%, with the AOC losing only 5 percentage points compared to 8 percentage points for digital models

Seven calibrated hardware non-idealities had no measurable negative impact on performance, suggesting the AOC's optical components operate reliably at scale