A new AI-native physics lab wants machines to do the experimenting
Physical Superintelligence, a Cambridge, Massachusetts startup backed by a $58 million seed round, is building AI systems meant to generate and test scientific hypotheses directly, part of a broader shift from AI-as-analysis-tool to AI-as-discovery-tool.
A new startup called Physical Superintelligence has launched in Cambridge, Massachusetts with a $58 million seed round led by Breakthrough Energy Ventures, built around a specific bet: that AI can move beyond analysing scientific data to actively generating and testing physical hypotheses itself.
Most AI applied to science so far has focused on pattern recognition — sorting through existing datasets faster than a human researcher could, or predicting how a molecule might behave before it’s synthesised. The pitch behind Physical Superintelligence is different: building models that propose their own experiments and help physically validate them, effectively acting as a research collaborator rather than just a faster search tool.
The launch fits a broader pattern this year of AI labs and investors moving “up the stack” from consumer and enterprise products toward fundamental scientific discovery, a space long seen as one of the higher-value but harder-to-automate applications of the technology.
