The Physical AI & World-Model Landscape — August 2026

A source-verifiable map of selected companies building physical AI, world models, simulation, and autonomous systems — and the user-governance capabilities their public product documentation still does not combine.

Physical world models have moved from research systems into vehicles, robots, drones, industrial sites, defense platforms, and interactive simulation. This landscape is a selected, source-verifiable sample of active companies whose products substantially model, predict, simulate, or control the physical world. It is not an exhaustive census, and its financing figures are not a valuation ranking.

*Reading the headline. The working aggregation is approximately $91B using the disclosed amounts and labels in the table. It includes specifically identified signed, milestone-based, and up to financing; it excludes valuations, secondary-share prices, acquisitions, contracts, grants, parent-company R&D, and unclosed talks. The rounded $90B+ headline is directional, not an audited total or a claim that the capital remains deployed.

I. Scope & Methodology

Unit of analysis. The company, not a model release or parent-company research lab. Each company receives one primary segment so counts and segment totals remain additive. Adjacent capabilities may overlap.

Inclusion rule. An active company must publish evidence of a product, platform, or research-to-product program that models physical state, generates navigable environments, learns embodied policies, simulates autonomous systems, or controls machines in the physical world. Cruise is removed because GM completed its acquisition and integration of Cruise, ending it as a standalone robotaxi company. Incumbent platforms and research programs such as NVIDIA Cosmos/Omniverse, Google DeepMind Genie, Meta V-JEPA, Tesla, and Amazon Robotics are relevant context but are not counted as companies in this venture-focused sample.

Financing rule. The table retains the announcement's native currency where material. The working USD aggregation uses a rounded contemporaneous conversion. It counts disclosed primary equity and separately disclosed debt or preferred financing. Signed, milestone-based, and up to amounts are labeled rather than presented as closed cash. Secondary transactions, tender offers, contracts, grants, acquisition prices, parent-company capitalization, and reported fundraising talks are excluded.

Status and evidence. Public companies are identified by ticker instead of a dated market capitalization. Acquired businesses are identified as subsidiaries. Proposed SPAC transactions are marked pending and future proceeds are excluded. Company announcements, regulatory filings, and investor-relations materials are marked Primary; credible reporting used to aggregate historical rounds is marked Secondary.

Governance review. ODEI evaluated publicly available product documentation through 12 August 2026 against a six-capability rubric: identity and authority, permissions and scopes, data boundaries, audit trails, revocation controls, and escalation policies. This is a documentation review, not an audit of private product capabilities.

II. Financing by Primary Segment

III. Selected Company Map

The 46 companies below are sorted by the working financing figure used for aggregation. The number is a research-normalized financing indicator, not a comparable accounting measure. Read every qualifier and status label.

Coverage boundary. This version favors a reproducible selected sample over an unsupported claim to be a complete market map. Scope-dependent candidates for a future annex include Momenta, DeepRoute.ai, Unitree, Sanctuary AI, Foretellix, Viam, Covariant/Amazon, and a broader Chinese humanoid cohort.

IV. Governance Finding

A bounded public-documentation result

Based on public product documentation reviewed through 12 August 2026, none of the sampled companies documents all six ODEI governance capabilities as a single user-controlled, cross-vendor layer. This is an evidence-of-documentation finding, not proof that a capability is absent from every internal system.

The ODEI research rubric asks whether one user-controlled layer combines:

• Identity and authority
• Permissions and scopes
• Data boundaries
• Audit trails
• Revocation controls
• Escalation policies

Many companies document subsets such as safety controls, operator supervision, fleet permissions, telemetry, simulation evidence, or incident procedures. The gap is the combination: durable authority and policy that belongs to a person or organization, travels across model and vendor boundaries, and produces auditable receipts for delegated action.