MathWorks Deep Dive: MATLAB/Simulink and Simscape Multiphysics Power Physical AI, with R2026a Copilots and an Agentic Toolkit for Edge Intelligence
MathWorks (MATLAB/Simulink) powers physical AI via Simscape multiphysics and the Reinforcement Learning Toolbox; R2026a adds MATLAB/Simulink Copilots and an Agentic Toolkit, with EDGE AI Foundation membership for edge AI.
Executive Summary
MathWorks states its position on physical AI in a single slide: generative AI is the fastest way to create, and MathWorks tools are the fastest way to verify. That sentence locates the company precisely. It does not compete in the model race, it does not sell compute, and what it sells is the step that turns AI output into a trustworthy engineering result.
In 2026 it turned that route into two concrete releases. R2026a in April placed Simulink Copilot and Polyspace Copilot inside the tools engineers already use, and opened the MATLAB MCP Core Server plus the MATLAB Agentic Toolkit so external AI coding agents such as Claude Code and OpenAI Codex can drive MATLAB. R2026b in September thickened the stack: a new Visual Inspection Toolbox for AI-based defect detection, Simulink Copilot inside the Profiler to explain why a simulation is slow, and PyTorch co-execution alongside MATLAB and Simulink.
This report makes five arguments.
First, MathWorks physical AI is an extension of model-based design rather than a new narrative. For four decades the company has done one thing: let engineers design, simulate, verify and deploy with models instead of physical prototypes. With AI in the picture the model becomes more important, not less, because an agent needs an inspectable, traceable object to work against.
Second, the technical substance of the Copilots lies in context rather than in the model. The official position is that the Copilots read model structure and MathWorks documentation rather than acting as generic chatbots. This matches COMSOL's reasoning but the implementation differs: MathWorks exposes tool capability in a standard way through an MCP server and lets customers bring their own model with their own key.
Third, the Agentic Toolkit is the most underrated product of this cycle. It is an open repository under matlab/matlab-agentic-toolkit that packages MATLAB and Simulink capability as skills an agent can call. By October 2026 the installer had reached the point where one command installs the MATLAB Agentic Toolkit, the Simulink Agentic Toolkit and the MATLAB MCP Server together. Agent integration moves from project work to out-of-the-box setup.
Fourth, financials and scale exist only as estimates, while the structure is clear. The company was founded in 1984, is headquartered in Natick, Massachusetts, remains private and unlisted, and operates in 16 countries. Third-party estimates put revenue at about 1.3 billion dollars and headcount between 6,000 and 8,000, with wide dispersion. No public financial statements exist, so every financial judgment here is labelled an estimate.
Fifth, competitive pressure comes from free alternatives and AI-native tools rather than from EDA or CAE giants. The Python ecosystem, meaning NumPy, SciPy, PyTorch and JAX, keeps eroding the algorithm side of MATLAB, and AI-native simulation tools gain entry with younger teams. The MathWorks response is not price cutting but making verification and compliance the hard-to-replace layer.
Evidence grading. Version features and the existence of the MCP server and Agentic Toolkit are primary, drawn from official release notes and the GitHub repository. The statement that Copilots read model structure is an official claim. Revenue and headcount are third-party estimates. The fastest-to-create and fastest-to-verify framing is marketing positioning.
Company Overview
The MathWorks, Inc. was founded in 1984 and is headquartered at 1 Apple Hill Drive, Natick, Massachusetts. It is private, unlisted and has taken no external venture funding. Jack Little, a central figure in the founding team, remains president and chief executive, unchanged for four decades. That is the most important governance fact about the company: without external capital pressure, it can build products on a ten-year horizon.
The product structure has two cores. MATLAB is the numerical computing and programming environment. Simulink is the graphical environment for multidomain dynamic systems and for model-based design of embedded systems. Around them sit more than 130 toolboxes and blocksets covering control, signal processing, communications, image processing and computer vision, deep learning, Simscape multiphysics, Stateflow state machines, Embedded Coder for code generation, and Polyspace for static analysis and formal verification.
Scale figures must be labelled as estimates. Third-party databases give revenue of about 1.3 billion dollars, with one source offering a range of one to five billion, and headcount readings from 6,000 to 8,033. The company's own public statement is more than 6,000 employees in 16 countries. The reasonable reading is that headcount approached 8,000 in 2026, with one database tracking 8,033 at plus 3.6 percent year on year, while the 6,000-plus figure is an earlier official position.
Two Release Cycles in 2026
MathWorks ships two releases a year, a and b, and that cadence has been stable for years. The two 2026 releases carry different emphases.
R2026a in April was the agent-access release. It put Simulink Copilot and Polyspace Copilot into the interfaces engineers already use, and opened the MATLAB MCP Core Server, which lets external agents call MATLAB through the Model Context Protocol, plus the MATLAB Agentic Toolkit, which packages MATLAB capability as agent skills. Roadmap language included bring your own key for OpenAI, agentic mode, sandboxed execution, and support for other products including Polyspace and RoadRunner.
R2026b in September was the release that put AI into specific workflows. Three new products arrived: Simulink Variant Manager for product variants, Polyspace Test Server for running unit and integration tests plus structural coverage analysis on an automation server, and the Visual Inspection Toolbox for building defect detection applications with AI and image processing. Simulink Copilot also entered the Profiler, where it explains why a simulation is slow and summarises solver behaviour.
Three Common Misreadings
Misreading one: treating the Copilots as a chatbot bolted into MATLAB. The official emphasis is reading model structure and the company's own documentation. That difference decides whether the answer is specific to the current model or generic. The first is useful to engineering; the second is dangerous to engineering, because it sounds right while possibly being wrong.
Misreading two: treating the Agentic Toolkit as a demonstration. It is a GitHub repository with continuous releases, two versions in September 2026 and another in October, and it now ships an installer that sets up three components with one command. That level of engineering shows a product line rather than a proof of concept.
Misreading three: treating MathWorks as a legacy tool being replaced by Python. Python is indeed displacing MATLAB in research and prototyping, but across the chain from model through simulation, code generation, verification and certification, MATLAB and Simulink remain the most complete option. In automotive ISO 26262 and aerospace DO-178C settings, where tool qualification matters, switching costs are very high.
Technology
Model-Based Design: Why It Suits Agents
Model-based design uses an executable model as the single source of truth, with design, simulation, testing, code generation and verification all organised around that model.
The structure attracts agents because the model is an inspectable object. If an agent changes a parameter, it can rerun the simulation and see the result. If an agent miswires a block, the simulation reports an error. That is a different reliability class from letting an agent edit code and hoping it runs.
The MathWorks stack divides into four stages along that chain: modelling through Simulink, Simscape and Stateflow; simulation through solvers, accelerator modes and parallel computing; verification through Simulink Test, Design Verifier and Polyspace; and deployment through Embedded Coder, HDL Coder and GPU Coder. With AI, all four stages gained Copilot or agentic interfaces.
Three Copilot Capabilities and Their Boundaries
Simulink Copilot has four published capabilities: understanding models, semantic search, troubleshooting errors and automating predefined tasks. R2026b connected it to the Profiler so it explains performance problems, which is a shrewd placement, because performance tuning is the most time-consuming and experience-dependent part of an engineer's week.
Polyspace Copilot addresses static analysis and formal verification, dealing with undefined behaviour, out-of-bounds access and deadlocks. These problems carry high false positive rates and high investigation cost, so AI assistance has clear value.
MATLAB Copilot covers programming, data analysis, plotting and documentation. It first appeared in R2025b and was extended in R2026a and R2026b.
All three share one boundary. Copilots propose and engineers decide. None is authorised to modify a model and deploy unsupervised. That position is what makes the products viable in regulated industries.
MCP Server and Agentic Toolkit: Turning Tools Into Agent Capability
This is the most substantive technical move of 2026.
The MATLAB MCP Server exposes MATLAB capability through the Model Context Protocol. The R2026a roadmap shows later support for Polyspace and RoadRunner as well as improved extensibility and sandboxed execution.
The MATLAB and Simulink Agentic Toolkits are collections of skills. In the October 2026 version, skills cover MATLAB software development, including creating and editing matlab.toml project files and managing dependencies through the package manager; data analysis guided by R2026b features; plotting; image processing; machine vision inspection with the Visual Inspection Toolbox; signal processing including acoustic metric computation; and wireless communications including USRP signal capture.
The design is clever in one specific way. It does not present MATLAB as one giant tool for the agent to call; it decomposes MATLAB into small skills. An agent loads only what a task needs, which lowers context cost and reduces the surface for error.
The October 2026 installer lowered the barrier further. From Claude Code or OpenAI Codex, one command installs the MATLAB Agentic Toolkit, the Simulink Agentic Toolkit and the MATLAB MCP Server.
The October 22, 2026 build, MATK-2026.10.b, added an installer for GitHub Copilot CLI and refreshed skills including matlab-choose-big-data-solution and matlab-design-radar, extending the set of coding agents that can drive MATLAB beyond Claude Code and OpenAI Codex.
On October 14, 2026 MathWorks announced that TÜV SÜD certification had been extended to a general AI tool category guideline conforming to ISO 26262, covering model-based design use cases built on the Simulink Agentic Toolkit. This is a statement about the certification category that applies to AI-assisted tooling, not a certification of the output of any particular model or workflow.
R2026b Engineering Depth: Three Products and Two Integrations
The Visual Inspection Toolbox is the most notable product of this release. It turns AI defect detection, covering scratches, voids and missing parts, into a toolbox, so control and test engineers can build inspection applications without moving to a deep learning framework. That has real manufacturing significance, because the work previously required either a dedicated vision system or someone fluent in PyTorch.
Polyspace Test Server moves unit testing, integration testing and structural coverage analysis onto an automation server, serving the compliance requirement directly.
Simulink Variant Manager handles product variants, managing multiple configurations from one model set, which is a hard requirement in automotive and aerospace derivatives.
Two integrations follow. PyTorch models co-execute with MATLAB and Simulink, so existing models do not have to be rewritten. The Point Cloud Toolbox supports lidar, stereo and RGB-D cameras and millimetre-wave radar, serving perception development for robotics and autonomous driving.
Layering Credibility
Verifiable: the R2026a and R2026b feature lists from official release notes; the MATLAB MCP Server and Agentic Toolkit, with version history and skill lists on GitHub; the three new products from official announcements; PyTorch co-execution; the MATK-2026.10.b release note; the TÜV SÜD certification extension to an ISO 26262 general AI tool category guideline; and the MATLAB EXPO Online 2026 programme.
Partially verifiable: the claim that Copilots read model structure, which is official without third-party evaluation; the practical effect of the Agentic Toolkit, which has no public benchmark; and the fastest-to-verify positioning, which is marketing.
Not verifiable: revenue and headcount, absent public statements; Copilot accuracy on real engineering tasks; and the reliability of agentic workflows in production.
Product and Commercial Model
The commercial model mixes licences, maintenance and subscription, and has stayed stable for years.
Licensing structure. Products are licensed individually, with the MATLAB base plus toolboxes, under three price tiers for commercial, academic and student use. Academic and student prices sit far below commercial, a long-running strategy of cultivating the education market: students form habits at university and later push procurement at employers.
Maintenance and subscription. Maintenance entitlement delivers new releases. The company has moved gradually toward subscription, particularly for cloud and concurrent licensing, without forcing the transition.
Pricing opacity. As with all engineering software, commercial list prices are not published. Observable characteristics suggest a single toolbox licence runs in the hundreds to low thousands of dollars per year, while full enterprise licensing is substantially higher.
The education moat. Penetration across universities worldwide is the deepest barrier MathWorks holds. Most engineering curricula teach with MATLAB, which means every graduating class is a pool of future users. The spending produces no short-term revenue and determines long-term customer acquisition cost.
Whether Copilots are separately charged. Public information is insufficient. The bring-your-own-key design in R2026a suggests model inference cost sits with the customer's own API key while MathWorks charges for tool licences. That shifts variable AI cost to the customer and keeps pricing simple.
MATLAB EXPO Online 2026, announced on October 28, 2026, runs online on November 4 and 5 with more than forty sessions and nine hands-on workshops.
Competitive Landscape
Against the Python ecosystem, the erosion is on the algorithm side. Python is the default in deep learning, data processing and rapid prototyping. MathWorks does not fight head-on; it holds the segment from model to product, meaning code generation, hardware deployment, verification and certification. R2026b support for PyTorch co-execution is effectively an acknowledgment that Python leads at the front end while MathWorks keeps the back end.
Against COMSOL, overlap sits in multiphysics. Simscape and COMSOL Multiphysics overlap in physical modelling but differ in positioning: Simscape targets system-level and electromechanical work tightly integrated with Simulink, while COMSOL targets fine-grained three-dimensional field solving. The typical relationship is complementary, with COMSOL used for field-level validation and Simscape for system integration.
Against Ansys inside Synopsys, Siemens and Dassault, the battlefields differ. Those companies sell solvers and platforms while MathWorks sells a modelling language and a verification flow. Real competition appears in budget allocation: a team that can buy only one suite chooses based on whether it builds systems or components.
Against AI-native simulation startups, the pressure is narrative. Those companies argue for AI-native, meshless, instant results. The MathWorks rebuttal is that engineering needs more than speed; it needs verification, traceability and certifiability. That argument holds in regulated industries and weakens in fast-moving consumer electronics.
Against NVIDIA, the relationship is toolchain collaboration rather than competition. GPU Coder and Deep Learning Toolbox support CUDA acceleration, and there is integration with the Isaac and Omniverse ecosystems for robot simulation. MathWorks sells no compute, so there is no direct conflict.
Financing and Capital Structure
MathWorks has no financing history. Founded in 1984, it remains private and unlisted with no venture funding, and the founding group around Jack Little retains control.
Financial characteristics, estimated. Revenue of about 1.3 billion dollars, headcount approaching 8,000, and therefore roughly 1.6 million dollars of revenue per employee on this report's calculation. That is below COMSOL at about 3 million, reflecting a larger organisation and a broader product line.
Capital allocation. No large acquisitions, with research funded internally. One 2025 item is worth recording: a 10 million dollar gift to the Appalachian Mountain Club for permanent protection of Barnard Forest in Maine. That is corporate social responsibility spending and does not change the business structure.
Why stay private. Private ownership allows long-horizon investment in education and product without quarterly justification. The cost is transparency, since outsiders cannot verify real growth or margin.
Valuation note. Without public statements, any valuation is speculative. A usable anchor is that engineering software companies often trade at six to twelve times sales, so 1.3 billion dollars of revenue implies a range of 8 to 15 billion. That is this report's rough arithmetic, not a valuation conclusion.
DISCREPANCIES
1. Headcount differs materially across sources. (Source grade: OFFICIAL versus DB) The company states more than 6,000 employees. Third-party databases give 7,900 in July 2026, 8,033 in September 2026 at plus 3.6 percent year on year, and 6,018 and 8,051 in two fields of the same database. Differences reflect timing and whether global subsidiaries are counted. This report uses approaching 8,000 and labels it an estimate.
2. Revenue exists only as third-party estimates, with wide ranges. (Source grade: DB) Two databases give about 1.3 billion dollars while another gives a range of one to five billion. The company discloses nothing. This report uses 1.3 billion and labels it an estimate.
3. The first release containing MATLAB Copilot is described two ways. (Source grade: MEDIA) One technical blog states clearly that R2025b introduced MATLAB Copilot, R2026a added Simulink Copilot and Polyspace Copilot, and R2026b added AI inspection and the Profiler Copilot. Other material attributes all three to R2026a. This report follows the former, which matches a finer version sequence and the official roadmap.
4. The MCP server name differs between the product line and the documentation. (Source grade: OFFICIAL) R2026a calls it the MATLAB MCP Core Server, while the Agentic Toolkit installer and later documentation call it the MATLAB MCP Server. This report notes on first use that both refer to the same component.
5. The R2026b release date comes only from non-official sources. (Source grade: MEDIA) The official release reviewed here carries no specific date, while a technical blog gives 24 September 2026. This report says September and flags the date as second-hand.
6. No official summary of Agentic Toolkit skills exists. (Source grade: OFFICIAL) The GitHub repository lists additions and improvements per version without a consolidated table. This report names representative skills and gives no total.
7. The fastest-to-create and fastest-to-verify framing is marketing positioning. (Source grade: OFFICIAL) It comes from the company's own technical presentation, and no third-party evaluation supports the comparative claim about verification speed. This report quotes it as positioning.
8. Certification scope: a general AI tool category guideline versus certification of any model output. (Source grade: OFFICIAL) The October 14, 2026 announcement describes TÜV SÜD certification extended to a general AI tool category guideline conforming to ISO 26262, covering model-based design use cases built on the Simulink Agentic Toolkit. The statement concerns the certification category that applies to AI-assisted tooling, and it does not certify the output of any specific model or workflow. This report reads it as a qualification-category extension. Source grade: OFFICIAL (announcement).
Risks and Diligence Gaps
Talent migration on the algorithm side. Younger engineers encounter Python more often at university and while self-teaching. Whether MathWorks education investment offsets that trend is the most important long-term variable.
A narrowing differentiation window in AI. Copilot features are easy to copy, since competitors can connect large language models and read their own documentation. The real barrier is depth in understanding model structure plus toolchain completeness, and that requires time to prove.
Adoption of the Agentic Toolkit is unknown. The repository has version history but no public download counts, active user numbers or customer cases. That prevents any judgment on whether it is in use or merely available.
Private ownership creates a transparency gap. Without statements, growth, margin and retention cannot be verified. For external parties making investment or procurement decisions, this is a material obstacle.
Pricing versus AI cost mismatch. Having customers supply their own API key shifts model cost to them, which helps MathWorks now and may later invite the question of why a Copilot licence is still required.
Diligence gap list, ordered by importance. First, actual revenue, growth and margin, which can be inferred indirectly through Massachusetts corporate registration or credit reports. Second, the revenue split between commercial and academic. Third, how Copilot features are priced, whether bundled or separately charged. Fourth, download counts, active users and production cases for the Agentic Toolkit. Fifth, Copilot accuracy on real engineering tasks from any internal evaluation. Sixth, China revenue share and localisation strategy. Seventh, the depth of collaboration with NVIDIA on Isaac and Omniverse robot simulation. Eighth, early customer counts for the three R2026b products.
Composite Assessment
techScore 8.0. The model-based design toolchain is the most complete in the industry, the Copilot route of reading model structure and company documentation is correct, and the Agentic Toolkit engineers agent access properly. The deduction reflects a weaker position on the algorithm side against Python and AI differentiation that no third party has validated.
bizScore 8.0. About 1.3 billion dollars of revenue, a stable licence and maintenance model and deep education penetration make a high-quality business. The deduction reflects slow growth and limited transparency from staying private.
fundScore 8.5. No external funding, no debt pressure and forty years of independent operation with self-funding cash flow. The deduction reflects the absence of public financial data.
productScore 8.0. MATLAB, Simulink and more than 130 toolboxes form a complete chain, and R2026b added two hard requirements in AI inspection and compliance testing. The deduction reflects a sprawling product line and a steep learning curve.
marketScore 7.5. Near-indispensability in automotive, aerospace, industrial automation and communications, with education as a long-term moat. The deduction reflects migration of younger developers to Python and entry by AI-native tools in fast-moving sectors.
References
- MathWorks press release introducing release 2026b of MATLAB and Simulink: three new products in Simulink Variant Manager, Polyspace Test Server and Visual Inspection Toolbox; Simulink Copilot generating Profiler insights; PyTorch co-execution with MATLAB and Simulink; the Point Cloud Toolbox; design of experiments and Gage R and R apps; TOML project files and File Exchange package management in MATLAB; JSON data dictionaries in Simulink; updates to Aerospace Toolbox and Blockset; Simscape integration and ray-based acoustics in Audio Toolbox; distributed analysis in Simulink Design Verifier; and the AMD SoC Blockset
- Same release: comments from Avinash Nehemiah, head of product management and marketing for design automation, on R2026b connecting design, simulation, test and deployment workflows
- MathWorks technical presentation, Accelerating Innovation with Model-Based Design and Generative AI, at MATLAB Expo India 2026: the R2026a and R2026b roadmap including bring your own key for OpenAI, agentic mode, the MATLAB MCP Server, sandboxed execution and support for Polyspace and RoadRunner, plus the skill mechanism in the MATLAB and Simulink Agentic Toolkits
- Same presentation, positioning statement: generative AI is the fastest way to create while MathWorks tools are the fastest way to verify, and trust comes from the process around authorship
- MATLAB Agentic Toolkit GitHub releases, MATK-2026.09.a, MATK-2026.09.b, MATK-2026.10.a and MATK-2026.10.b: the Agentic Toolkit Installer supporting one-command installation of the MATLAB Agentic Toolkit, the Simulink Agentic Toolkit and the MATLAB MCP Server from Claude Code or OpenAI Codex, with skills including matlab-configure-toml-project, matlab-use-package-manager, matlab-process-images, matlab-use-visual-inspection, matlab-evaluate-acoustic-metrics and matlab-detect-capture-usrp
- Technical blog review of R2026b: release date of 24 September 2026; the three new products; Simulink Copilot explaining Profiler results; PyTorch co-execution; and a recap that R2025b introduced MATLAB Copilot while R2026a added Simulink Copilot, Polyspace Copilot, the MATLAB MCP Core Server and the MATLAB Agentic Toolkit
- MathWorks website: R2026b as the latest release; company overview; the three application tracks of model-based design, robotics and wireless communications; and the November 4 to 5, 2026 event on MATLAB and Simulink in engineering systems
- Company and scale information: founded 1984, headquarters at 1 Apple Hill Drive in Natick, Massachusetts, privately held, operations in 16 countries, and Jack Little as president and chief executive, together with third-party headcount and revenue estimates addressed in DISCREPANCIES 1 and 2
- Product line: MATLAB, Simulink, Simscape, Stateflow, Polyspace, Embedded Coder, HDL Coder, GPU Coder, Deep Learning Toolbox and Simulink Design Verifier among more than 130 products
- Copilot capability boundaries: the four published Simulink Copilot capabilities of understanding models, semantic search, troubleshooting errors and automating predefined tasks; Polyspace Copilot for static analysis and formal verification; and MATLAB Copilot covering programming, data analysis, plotting and documentation
- Corporate social responsibility record: a 10 million dollar gift in 2025 to the Appalachian Mountain Club for permanent protection of Barnard Forest in Maine
- Industry media and database characterisation of MathWorks, including a competitor list of SAS, Autodesk, PTC, National Instruments, Ansys, COMSOL, Wolfram and Maplesoft, used for competitive framing
- MATLAB Agentic Toolkit GitHub release MATK-2026.10.b, October 22, 2026: an installer for GitHub Copilot CLI and refreshed skills including matlab-choose-big-data-solution and matlab-design-radar
- MathWorks newsroom, October 14, 2026: TÜV SÜD certification extended to a general AI tool category guideline conforming to ISO 26262, covering model-based design use cases with the Simulink Agentic Toolkit
- BusinessWire announcement, October 28, 2026: MATLAB EXPO Online 2026, held online on November 4 and 5 with more than forty sessions and nine hands-on workshops
This report is compiled from public sources for informational purposes only and does not constitute investment advice. Data is current as of 2 November 2026; please refer to company announcements and periodic reports.
Related papers
Related tools
Related cases
- DrivAerML High-Fidelity Road-Car Aerodynamics Dataset and Surrogates
- AirfRANS High-Fidelity Airfoil Surrogate Benchmark Case
- Physics-Informed CNN for Battery Pack Temperature Fields
- IASC-ASCE Structural Health Monitoring Benchmark and Physics-Informed Damage Identification
- OpenFOAM Industrial Multiphysics Pipeline: From Meshing to Data Supply
Related reports66
- 51World Deep Dive: A Listed Digital Twin Company Where 51Sim Becomes the First Growth Engine While 51Earth Nearly Vanishes, and Two Irreconcilable Financing Naming Conventions Must Both Be Kept
- AMI Labs: A USD 1.03B Seed Bet on World Models, LeCun's Non-Generative Path and Its Blank Scorecard
- Anshi Pacific (PERA) Deep Dive: A Beijing Domestic-CAE Leader Founded in 2003 Whose PERA SIM General Simulation, AIROM Reduced-Order Models, AgentX Multi-Agent and Jingzhi iGPT/Whale Span from Industrial Simulation to Physics AI
- AspenTech Deep Dive: The Process-Simulation Incumbent Acquired by Emerson and Its Hybrid Industrial-AI Route
- Autodesk Deep Dive: Neural CAD Foundation Models and a $200M World Labs Bet on Physical-World AI, with NAVASTO Adding CFD Surrogate Capability
- Bentley Systems Deep Dive: iTwin Digital Twin and Generative AI Empower Full-Lifecycle Infrastructure Engineering
- BeyondMath Deep Dive: A Foundation Physics AI on the Neural-Operator Route, a $19M Consortium Budget, and Three Misread Facts
- Cadence Deep Dive: System-Level Multiphysics EDA and the Millennium M2000 Supercomputer Anchor Physical AI, with AuraStack Agentic AI and Fidelity GPU CFD as Key Levers
- COMSOL Deep Dive: Four Decades of Multiphysics Solvers, Entering Physical AI via DNN Surrogate-Model Training and All-Physics GPU Acceleration
- Dassault Systèmes Deep Dive: 3DEXPERIENCE and SIMULIA Virtual Twin Physics Behavior Anchor Industrial AI, with LEO Virtual Companion and GPU Explicit Dynamics as Physical-AI Levers
- DP Technology: From the Gordon Bell Prize to National-Level Unicorn Status and the Door of a Hong Kong Listing, the AI-for-Science Leader's Technical Delivery and Commercial Blank
- Emmi AI: From Linz Spin-out to a EUR 330M Mistral Acquisition in 17 Months
- Fainite Deep Dive: A Zurich Physics-AI Simulation Startup Whose Caltech, ETH Zurich, Cambridge and Google Team Uses Physics-Aware Deep Learning to Cut Simulation from Tens of Hours to Seconds, With Only Venture Kick Grants and No Priced Round
- Fysics AI: The PhysX Founder's Second Act, a Domestic Differentiable Physics Engine and Its Ecosystem Play
- Frontop (301313) Deep Dive: From Digital Exhibition Halls to a Physical-AI Data Foundation - FTE Engine, FunSim Simulation and the Gap Between Positioning and Revenue
- Godela Deep Dive: A YC X25 Physics-AI Simulation Startup Whose Ex-Apple, Google and Intel Team Uses Physics-Constrained Surrogate Models to Cut Simulation from Days to Seconds, at Seed Stage with Pilot-Only Access
- HollyWave (688682) Deep Dive: RDSim AI Electromagnetic Simulation and Engineering Agents Close the Simulation-Measurement-Product Loop
- IAT (Alt) Deep Dive: China's First Listed Independent Automotive Design Company Pivots to AI-native R&D with the Yuanfang, Yufeng and Taiyi 3.0 Product System and an NVIDIA Omniverse Partnership, Reporting RMB 975M Revenue and a Widened RMB 250M Net Loss in 2025
- Inductiva Deep Dive: A Closed Case in Open-Solver Orchestration, Eight Premises That Must Be Corrected, and Where Co-Founder and CTO Hugo Penedones Went Next
- Intesim (INTESIM) Deep Dive: A Home-grown Chinese CAE Leader Founded in 2009 with Self-developed Solvers for Structure, Fluid, Electromagnetics, Thermal and Acoustics and Multiphysics Coupling, a ~RMB 100M B Round from CNII and Shenzhen Capital, and 2025 Revenue and Bookings Both above RMB 100M
- iSoftStone (301236) Deep Dive: A ChiNext-Listed IT Services Giant Whose TianShu Platform Builds a Three-Layer Physical-AI Stack with Self-Developed TwinHub Digital Twin, FSim Industrial Simulation, CAE Multiphysics and Phyxis Embodied Control
- Jua Deep Dive: The Earth Physics Transformer as a Large Physics Model for Energy Trading
- Keysight (with ESI Group) Deep Dive: Virtual Prototyping and Hybrid AI Bridge Measurement and Simulation, with Natural-Language Assistants in EDA
- Lightwheel Deep Dive: A Physical AI Data and Evaluation Infrastructure Startup Founded by a Former NVIDIA Simulation Lead, Seven Rounds Totalling Roughly CNY 2.65 Billion at a Valuation Above USD 2 Billion, Seats on Both the Newton and EgoVerse International Standards Committees, With Two Incompatible Accounts of Q1 Orders
- Luminary Cloud: The Physics AI Model Factory Backed by NVentures
- Maniformer (Mifeng) Deep Dive: An Agibot-Spun Physical-AI Data Infrastructure Firm Whose Bodyless MEgo Capture Hardware Is in Mass Production and Which Closed Three Rounds in Half a Year Led by China Telecom
- Mistral AI: How Physics AI Moved From Language Models to EUV Lithography and Nuclear Power
- Monolith AI: The British Engineering-AI Firm That Cut Physical Test Counts Instead of Accelerating Simulation, Now Inside CoreWeave
- Monumo Deep Dive: A Cambridge Deeptech Company Redesigning Electric Motors with the Anser AI Engine, Tens of Millions of Multiphysics Simulations per Day, GBP 10.5M Raised in Total, and a Series A Being Raised
- NAVASTO: Moving Solver Insight Onto the Designer's Desk, and How Autodesk Bought That Entry Point
- Navier: Handing CFD Workflows to Agents, a Seed-Stage Bet on AI for Hardware Engineering
- Neural Concept: An EPFL Spin-out Betting on Neural Surrogates, with $138M Raised
- NP Co. (NP Company) Deep Dive: A 2025-founded Paris Inria-born AI-native simulation startup that raised a EUR 6M pre-seed led by Partech with Mistral AI co-founders as angels, claiming a pre-trained physics foundation model at 1,000x speedup — most claims unverified
- NVIDIA Deep Dive: Building the Full-Stack Physical-AI Infrastructure with PhysicsNeMo, Warp, Cosmos World Foundation Models and Omniverse
- Physical Intelligence Deep Dive: A $2.1B Bet on the GPT Moment for Robotics, With pi0.7 Self-Reporting 60-80 Percent Success on Unseen Tasks and Neither Pricing nor Revenue Disclosed for the Pi API
- Physical Superintelligence Deep Dive: A $58M Seed Bet on Virtual Physicists, With Emmy Targeting AI Data Center Multiphysics Optimization as Its First Commercial Proof Point
- PhysicsX: Pretrained Geometry and Physics for Industrial Engineering at a $2.4B Mark
- Prometheus: A $41B Mark, $18.2B Raised, and Not Yet a Product
- PTC Deep Dive: Creo/Onshape and the ThingWorx Digital Thread Power Physical AI, with NVIDIA Isaac Sim Bridging Robotics Design to Simulation
- Qianjue Technology Deep Dive: How a Body-Agnostic Embodied Brain Pairing the Predictive World Model Polibrain with Brain-Inspired Partitioning and Polynomial Representation Reached a Hundred Thousand Cleaning and Service Robots
- Quanscient Deep Dive: Cloud-Native and Quantum-Powered Multiphysics Simulation as the Data Engine for AI-Native Hardware Engineering
- Rescale Deep Dive: An HPC Operating System and Desktop Engineering Workbench in the Cloud, a $115M Series D, and Five Misread Facts
- Schrödinger Deep Dive: Physics-Based FEP+ Simulation and Machine Learning Power Molecular Discovery, with the 2026 Bunsen Agentic Co-Scientist
- Secondmind Deep Dive: A Cambridge Engineering-AI Firm Whose Bayesian Active Learning Cuts CAE Simulation and Test-Bench Runs by up to 90%, with Roughly USD 73M Raised and Mazda Both Customer and Investor
- Shichuang Technology Deep Dive: SupreCAST Cloud-Native Material-Processing Simulation and Industrial Agents Advance the CAE+AI Strategy
- ShonCloud (Shunyun) Deep Dive: Meshless Particle-Method CFD for EV and Nuclear Fluid Multiphysics
- Shuzhi Maijin Deep Dive: A Shenzhen Fully Self-Developed SNN-plus-LBM CFD Startup Claiming Second-Level Results on Hundreds-of-Millions of Grids on Consumer Hardware, With a Tens-of-Millions-RMB Angel Round From a Ministry-of-Finance-Affiliated Fund and an FPGA Migration Plan
- Siemens (incl. Altair) Deep Dive: The 2025 Altair Acquisition Unites Simcenter and Altair, with PhysicsAI Geometric Deep Learning and PhysicsAI Generate as the Physical-AI Core
- Silurian Deep Dive: A Generative Forecasting Transformer Earth-System Foundation Model for Weather Risk on Energy and Infrastructure Assets
- SimScale Deep Dive: A Cloud CAE Platform Built on Core-Hour Subscriptions, EUR 52M of Equity Kept Separate from EUR 25M of Debt, and Five Misread Facts
- Skild AI Deep Dive: A General Robot Foundation Model Incubated at CMU's Robotics Institute, With a USD 1.4B Series C at a USD 14B Valuation, a Conflicting Series B Size, and an ARR Figure That Is a Run-Rate Rather Than Revenue
- Songying Deep Dive: A Huawei-Cloud-Alumni Startup Building a Domestic Physical AI Operating System, With ORCA OS Positioning Against NVIDIA Omniverse, Three Incompatible Financing Narratives, and a Verifiable Third-Party Anchor in a CNY 20 Million Minority Investment
- Sudo Tech Deep Dive: Zero-Real-Data Simulation Training, the Fully In-House Sudo R1 Stack, and an Atomic-Skill Commercialisation Strategy from a Startup Valued Above CNY 20 Billion in 17 Months
- Suochen Technology: The Pain of Becoming a Physics-AI Platform from a CAE Vendor
- Synopsys (Ansys) Deep Dive: The 2025 Ansys Acquisition Turns the EDA Leader into a Silicon-to-Systems Engineering Powerhouse, with Ansys SimAI and the NVIDIA Physical-AI Partnership as the Core Physics-AI Levers
- TenFong Deep Dive: An Academician-Led Chinese Industrial Simulation Software Vendor Whose Official Timeline and Database Records Use Incompatible Round Names, With an Official Financing PDF and an Investment Database Both on File
- Tianfu (Nanjing Tianfu Software) Deep Dive: A 2011-founded Chinese Intelligent-CAE Veteran That Closed a Several-Hundred-Million-RMB Series C, Pivoted to an Industrial World Model in 2024, With Unreconcilable Financing and Valuation Records and No Disclosed Revenue
- Tongyuan Soft Control Deep Dive: MWORKS (Syslab/Sysplorer) as a Domestic MATLAB/Simulink Alternative, with MWORKS AI and the MOGICK Engineering Agent Advancing Home-Grown Physical AI
- Vinci AI: Turning Physics Simulation from a Checkpoint into a Continuous Constraint
- Westlake Robotics Deep Dive: Westlake University's First Robotics Spin-out Builds a Dual-Pretrain Brain-and-Cerebellum Architecture with WR1 Whole-Body Foundation Model and GAE Behaviour Expert
- World Labs: Fei-Fei Li's Spatial-Intelligence Unicorn, Acquired by AMD for $8.2B
- XVERSE Deep Dive: A Simulation-World Route to Physical AI Built on the XVERSE Model Family, an Edge-Cloud 3D Engine and the Physics-Based MotionGen Model
- SimWE: From In-House Solvers to a Physical-AI Engine, the Domestic CAE Leader and What Comes Next
- Yunjing Zhifang Deep Dive: A Peking-University-Incubated Cloud-Native GPU CAE Company with the Simversus Platform and the YJ-DORGP Small-Sample Intelligent Solver, Reporting 275 Percent Order Growth, 160 Percent Revenue Growth and Breakeven in 2025
- Zhongke Diwiji (Five Ages) Deep Dive: A CAS-Institute-of-Automation and Tsinghua Spin-out Whose FAM Embodied Foundation Model Uses Action Optical Flow for Cross-Embodiment Generalisation, Reaching 97% Success from 3 to 5 Demos, with EraBrain and Jiling Robots Serving Sinopec, Leapmotor and Midea
- ZWSOFT (688083) Deep Dive: Overdrive Kernel and ZW3D Wukong 2027 Underpin an Integrated CAD/CAE/CAM Stack, with a CAx Agent Platform as the Practical Route to Engineering AI
This report is compiled from public sources. It is for reference only and does not constitute investment advice. Data is current as of the publication date; please refer to the company’s official filings.