Google's SensorFM Uses Trillion Minutes of Data to Interpret Health

Google has released SensorFM, a wearable foundation model trained on a trillion minutes of sensor data collected from five million people. The model represents a fundamental shift in how health wearables deliver value: not by producing more accurate raw measurements, but by interpreting those measurements through AI.

The company tested SensorFM by integrating its predictions into a Personal Health Agent, then asking physicians to blindly evaluate the summaries it generated. According to details first reported by Forbes contributor Josipa Majic Predin, the AI-generated summaries outperformed a no-data baseline across every dimension physicians assessed. More significantly, the summaries were statistically indistinguishable from those built using actual ground-truth clinical measurements.

That outcome points to a specific product direction: the interpretation layer matters more than the underlying sensor precision.

Why it matters

Wearable companies have spent years competing on sensor accuracy and data volume. SensorFM's physician evaluation suggests the next competitive frontier is not better hardware, but better AI that tells users what their data means and what to do about it. This shift has immediate implications for how investors value wearable companies and how healthcare systems might integrate consumer devices into clinical workflows.

The industry is already moving this direction

Google is not alone in pursuing AI-driven interpretation. Whoop launched Whoop Coach in 2023, a GPT-4 powered conversational layer that sits on top of its biometric data. By 2026, four of the five most common user questions were about self-improvement actions rather than data lookup. Oura introduced Oura Advisor with a similar goal, transforming what the company's product lead described as a one-way stream of insights into a two-way coaching dialogue.

Both products address the same problem SensorFM's evaluation quantifies: users struggle to act on hundreds of daily data points. They need a system that interprets and advises, not just measures.

Integration with clinical systems comes next

The pattern extends beyond consumer coaching. Whoop has begun offering clinician access to patient data and integrating with electronic health record systems. This suggests wearables are positioning themselves as continuous clinical intake systems rather than standalone wellness tools.

SensorFM's architecture supports that direction. By replacing single-purpose health algorithms with one generalist model, Google has built infrastructure that could scale across clinical use cases without requiring new algorithm development for each condition or metric.

Investors are responding to this shift by directing capital toward companies that prioritize diagnostic AI and agentic interpretation over hardware innovation alone. The wearable that wins may not be the one with the most sensors, but the one whose AI layer delivers the most clinically actionable interpretation.

The details on SensorFM were reported by Josipa Majic Predin at Forbes.