Physical AI Research
Research connecting the Physical AI hypothesis to its layers, technologies and companies.
How we investigate Physical AI
Physical AI connects sensing, inference and action. The investment question is which qualified components and operating systems turn a model into dependable work in a real environment.
- Research formats
- 5
- Technology themes
- 138
- Evidence standard
- Verifiable sources
Recent and featured Physical AI research
One evidence library, organized by research type and subject. Featured analysis and technology tiers share the same system map.
Physical AI — Earnings Call Recap (Q3 2026)
Physical AI earnings evidence for 2026-07-01 to 2026-09-30: guidance, surprises and thesis changes.
Featured analysisRead research MapAI Data Center 800V DC Power Architecture — System-Level Supply Chain
4 tiers and 3 companies mapped with confidence-labeled evidence.
Featured analysisRead research Stock screenerBest Physical AI stocks for 2026
A research-led shortlist of the public companies supplying the compute, power, sensing, motion and manufacturing constraints behind intelligent machines.
Featured analysisRead research Stock screenerWhat is Physical AI?
Physical AI combines models, sensors, real-time compute and machines that perceive, decide and act in the physical world.
Featured analysisRead research Stock screenerPerception and sensing stocks
A sourced research note on machine vision, LiDAR, radar, edge compute and the public companies that capture value across the perception stack.
Featured analysisRead research Deep DiveAdvanced Metering Infrastructure (AMI) and Smart Grid Distribution Automation
3 mapped public companies across Grid, Power & Thermal Infrastructure.
3 companies · Grid, Power & Thermal InfrastructureRead research GuideAutonomous Machines & Automotive Semiconductor Deep Dive
0 mapped public companies across Autonomy Software, Fleet Platforms & End Markets.
0 companies · Autonomy Software, Fleet Platforms & End MarketsRead research Deep DiveBulk metallic glass (amorphous metal) components
2 mapped public companies across Materials & Critical Components.
2 companies · Materials & Critical ComponentsRead research GuideHumanoid Robot Actuation Economics
0 mapped public companies across Actuation & Motion Control.
0 companies · Actuation & Motion ControlRead research GuideHumanoid Robotics
0 mapped public companies across Autonomy Software, Fleet Platforms & End Markets.
0 companies · Autonomy Software, Fleet Platforms & End MarketsRead research Deep DiveIron nitride (FeN) rare-earth-free permanent magnets
3 mapped public companies across Materials & Critical Components.
3 companies · Materials & Critical ComponentsRead research GuideLiDAR architectures and competitive landscape
0 mapped public companies across Perception & Sensing.
0 companies · Perception & SensingRead research Deep DiveLiquid cooling / CDU
5 mapped public companies across Grid, Power & Thermal Infrastructure, Manufacturing, EMS & Industrial Automation.
5 companies · Grid, Power & Thermal InfrastructureRead researchThe Physical AI research framework
We begin with an investable question inside this universe, map it to the relevant layer and technology, and then build an evidence ledger from primary documents, regulatory material, company filings, technical standards and attributable reporting. Sources are kept close to the claims they support, dates remain visible, and uncertainty is described rather than hidden.
Company exposure is evaluated only after the system map is clear. We distinguish direct revenue and qualified deployment from optional narrative exposure, compare public and private participants on the same capability map, and retain disconfirming evidence. Research is refreshed when a source changes the thesis—not simply to make a page appear new.
A model running on a server can generate a useful answer without moving a machine. Physical AI must connect information to an action whose timing and consequences matter outside the computer. A robot needs observations of its surroundings, a decision process, power and actuation. An automated laboratory needs traceable samples and coordinated instruments. The common theme is a complete operating loop, but the markets and qualification requirements differ.
PXS Research evaluates that loop as a chain of dependencies. Better inference can be valuable, yet it cannot remove a poor sensor installation, an unreliable power rail or a process that lacks a safe recovery path. The most visible AI processor is therefore not always the binding constraint. Infrastructure and specialized components can create valuable opportunities when they relieve a demonstrated operating bottleneck.
This framework separates technical necessity from commercial capture. A component can be essential while its suppliers compete aggressively and earn modest returns. A supplier can also report strong AI-related demand without having an exclusive position. The company-level question is whether qualified content, customer retention and delivery economics support cash generation over the life of an actual production program.