Problem Statement: Limited physical-world data + costly robot testing -> slow and expensive robotics development. Impact if unsolved: more robot hardware, human monitoring, GPU orchestration, and longer deployment cycles. Alternatives: physical robot testing, teleoperation fleets, classical simulators, synthetic data, and proprietary robotics world models.

Founded: 2025 LinkedIn; 2026 Y Combinator. Incorporated: Not publicly disclosed. Status: Active. Funding history: Y Combinator seed funding reported in June 2026. Funding Amount: 0.125 USD million, as listed by Dealroom; not independently confirmed by the company or Y Combinator. Acquisition: None publicly disclosed. Founder LinkedIn: Shehryar Saroya — https://linkedin.com/in/shehryar-saroya; Ahmad Saroya — https://www.linkedin.com/in/ahmad-saroya-a05400237/

UNIT METRICS(to be removed later)
----------------------------------
Unit Metric | Number
Physics-IQ Verified | 56.6 points
Physics-IQ Original | 62.6 points
Input Horizon | 3 seconds
Output Horizon | 5 seconds
Benchmark Cases | 66 cases
Rollouts Analyzed | 1142 rollouts
Drop Failures | 231 rollouts
Drop Cost | 9.1 points
Slip Failures | 148 rollouts
Clutter Failures | 87 rollouts

Due Diligence Segment
---------------------
Market & Sector Thesis: Robotics foundation models may reduce the cost and time required to train and validate physical automation systems. World models could become an infrastructure layer between perception, planning, and robot control.
Market Sizing: No company-specific market-size estimate publicly disclosed.
Operating Traction: Atom 1 claims a 56.6 score on DeepMind's Physics-IQ Verified benchmark; the company states it is onboarding a small cohort of design partners.
Direct Competition: NVIDIA Cosmos; World Labs; Physical Intelligence; Enact. These represent overlapping physical-AI, world-model, or robot-training capabilities; comparable company size is not publicly established.
MOAT Durability: Potential moat is a growing dataset and evaluation feedback loop combining internet-scale video, robot actions, failure analysis, and environment-specific fine-tuning. The moat could be disrupted by better open-source world models, proprietary robot-data networks, or incumbent platforms with greater compute and customer access.
Plausible probable AI models used: Atom 1; specific base-model architecture and model providers are not publicly disclosed.
Verified technology: Internet-scale video training; video-conditioned forecasting; action-conditioned rollouts; robot-camera and control inputs; LLM-orchestrated motion; typed rollout verdicts; GPU-parallelized simulation and evaluation.

Sources:
Moving Atoms official website — https://movingatoms.ai/ — accessed 2026-08-28; page displays copyright year 2026.
Y Combinator company profile — https://www.ycombinator.com/companies/moving-atoms — accessed 2026-08-28; lists founders, YC Summer 2026, Active status, San Francisco, and company description.
Moving Atoms LinkedIn company page — https://www.linkedin.com/company/movingatoms/ — accessed 2026-08-28; lists founded year 2025, company type, employee range, and company description.
Moving Atoms Physics-IQ page — https://movingatoms.ai/physics-iq/ — accessed 2026-08-28; reports benchmark scores, methodology, and rollout details.
Dealroom company profile — https://app.dealroom.co/companies/moving_atoms — accessed 2026-08-28; lists June 2026 Y Combinator funding of USD 125,000. Dealroom also shows an enterprise-value range, but this is not treated as a verified valuation because it is not confirmed by the company or an investment announcement.