PSI is a public benefit corporation building an AI-native research institution to accelerate physics discovery. The company combines human-directed research missions with virtual physicists that autonomously run experiments, verification, and engineering campaigns. PSI targets a scaling law where more inference-time computation produces stronger performance on hard research problems, with its first proving ground in AI infrastructure physics.
Funding
Funding not disclosed
Founders
Product
Problem
Modern civilization relies on physics discoveries from the twentieth century, yet the pace of fundamental research has slowed dramatically. Institutional barriers, grant cycles, and human career timelines constrain how quickly new physics can be discovered and translated into technology. As AI collides with physics—through energy limits, heat density, materials constraints, and sensing gaps—there is a growing need to overcome these bottlenecks.
Solution
PSI operates as an AI-native laboratory that fuses human research direction with machine-executed scientific workflows. Human scientists define missions, set verification criteria, and choose which problems matter most, while virtual physicists carry out thousands of simultaneous research threads across hypothesis generation, simulation, experiment, and engineering. The company uses an Autonomous Campaign model where each effort runs continuously from problem definition through deployment, with results gated by independent verification methods such as conservation laws, symmetry, formal proof, and physical measurement. Early evidence points to a scaling law where more inference-time computation yields stronger performance on hard research problems, suggesting discovery can be planned and resourced rather than waited for.
Target Audience
PSI serves advanced research institutions, government agencies, and industrial partners in sectors that depend on frontier physics—including energy, semiconductors, materials science, and AI hardware—who need accelerated discovery cycles beyond what traditional academic or national laboratory timelines allow.
Features
- Autonomous Campaigns: continuous research efforts that integrate hypothesis, verification, experiment, engineering, and deployment into a single workflow
- Verification-gated results: candidate outcomes must pass checks against conservation laws, symmetry, dimensional consistency, formal proof, simulation, digital twins, and physical experiment
- Separation of generation and verification: machine-generated hypotheses are independently judged by systems that cannot be argued with
- Scaling-law-driven research planning: inference-time computation budgets correlate with research performance, enabling milestone-based campaign scheduling
- Human-directed taste and mission selection: scientists define objectives and success criteria while machines execute the work
- First proving ground in physical AI infrastructure, targeting constraints in energy, heat, materials, sensing, and communication