COMSOL Deep Dive: Four Decades of Multiphysics Solvers, Entering Physical AI via DNN Surrogate-Model Training and All-Physics GPU Acceleration
COMSOL, a privately held Stockholm multiphysics solver vendor founded in 1986, enters physical AI through 6.3 DNN surrogate-model GPU training and 25x acoustics, and 6.4 all-physics GPU acceleration via cuDSS.
Executive Summary
COMSOL is the least platform-like company on this list. It does not sell compute, design chips, build AI factories, or try to wrap whole plants and buildings in twins. It does one thing, multiphysics simulation, and it has done that one thing for forty years. In 2026 it marked its fortieth anniversary and released version 2027, whose centrepiece is the COMSOL MCP Server, which lets external AI agents create, modify, run and analyse models through the Model Context Protocol.
This report makes five arguments.
First, COMSOL has the lowest cost of becoming agent-native among the four, and the reason is architecture rather than spending. Chief technology officer Ed Fontes explains that the entire modelling environment is controlled through one unified object-oriented API, so an agent can run the whole modelling workflow through a single interface. That sentence carries more weight than it appears to. Most CAE software inherits a mismatch between graphical operations and scripting interfaces, so connecting an agent requires a tool definition for every function. COMSOL structure is natively suited to being called by an agent.
Second, scale is small and profit quality is very high. The company is private and publishes no financial statements. Industry media estimate revenue at about 200 million dollars and global users at about 200,000, and describe nearly half of revenue as pure profit. That margin is exceptional in software, and it comes from a small headcount of about 650, direct sales and no acquisition spending.
Third, technical progress runs on two tracks: more physics and faster solvers. Version 6.4 added explicit structural dynamics for impact, blast and drop events, plus a Granular Flow Module based on the discrete element method, and connected the NVIDIA cuDSS direct sparse solver with an official claim of several-fold speedups in single-physics and multiphysics simulations. Those two directions, more extreme physical processes and faster linear solves, have been the route for forty years.
Fourth, large language models assist rather than replace. Version 6.4 upgraded the Chatbot window to connect to GPT-5, DeepSeek, Gemini and self-hosted models, with direct retrieval from COMSOL documentation and the ability to attach model tree nodes, complete models and images as context. The point is not model capability but context quality: injecting the current model as context brings answers far closer to the user's actual problem than any general model's knowledge.
Fifth, the competitive position trades commercial scale for technical depth. On industry estimates for 2026, the Ansys multiphysics business inside Synopsys is about 3.0 billion dollars, Siemens including Altair about 1.6 billion, and COMSOL about 200 million. COMSOL is not in the same revenue league, but peers treat it as the benchmark for technical purity in multiphysics coupling.
Evidence grading. Version features are primary, taken from official release notes. Revenue, user counts and headcount are estimates without audited statements. The claim that nearly half of revenue is profit is a media report. The claim that the MCP Server lets an agent run the full workflow is a vendor statement without third-party reproduction.
Company Overview
COMSOL AB was founded in Stockholm in 1986 by Svante Littmarck and Farhad Saeidi, and remains private with no external venture funding and no listing. It operates 16 offices plus a distributor network covering major markets, with the United States subsidiary COMSOL, Inc. based in Burlington, Massachusetts.
Headcount appears in three versions reflecting timing and scope. Third-party databases give 600 in July 2026 and about 660 in August 2026, while the company's own fortieth anniversary blog says more than 650. These are not contradictory; they are different months. The reasonable statement is about 650.
There is no official revenue figure. Industry outlet PLM and ERP News estimates about 200 million dollars and states that few product lifecycle vendors match COMSOL on profitability, with nearly half of revenue being pure profit. Database entries quoting a range of one to ten billion dollars are too wide to carry information and should not be used.
The user count is about 200,000 globally, again a media estimate rather than company disclosure. Set against 200 million dollars of revenue, that implies roughly 1,000 dollars of revenue per user. Given COMSOL pricing, where a single module typically runs from several thousand to tens of thousands of dollars a year, 200,000 more likely counts cumulative or academic users rather than paid seats.
Forty Years and a Confirmation of Positioning
2026 was the fortieth anniversary. The company marked it in two ways. It held its first Simulation Summit, a one-day event in Santa Clara, California, and it continued the global COMSOL Conference tour, which ran in multiple cities during 2026.
One line from the anniversary blog deserves recording. The founding team stressed that employees built everything and that every single person is needed. That corresponds to an organisational choice: no large acquisitions, and product expansion through internal research. It stands in sharp contrast to Siemens buying Altair, Synopsys buying Ansys and Cadence buying Hexagon D&E in the same period.
Three Common Misreadings
Misreading one: treating COMSOL as a small Ansys. The product philosophies differ. Ansys is a portfolio of solver brands assembled into a platform, while COMSOL is a single codebase with modular physics. Every COMSOL module shares one core, one graphical interface and one API, which is the root reason its agent conversion is cheap. Ansys modules arrived through different acquisitions, so interfaces and scripting conventions differ.
Misreading two: treating 200 million dollars of revenue as a small company. In CAE, 200 million dollars of revenue at close to 50 percent net margin produces better cash quality than many larger but thinner businesses. The constraint is not money; it is appetite to expand.
Misreading three: treating the Chatbot as the AI strategy. The Chatbot is the surface. The strategic move is the MCP Server, which exposes modelling capability as tools an agent can call. The first serves engineers; the second serves engineers plus their agents.
Technology
A Single Codebase: Why It Matters
The technical foundation is one unified partial differential equation solver framework. Every physics module, whether electromagnetics, structural mechanics, acoustics, fluid flow, heat transfer or chemical engineering, adds equation terms, boundary conditions and material models within that same framework rather than existing as a separate program.
Three consequences follow. First, multiphysics coupling is native, with no data exchange between programs, so a user can solve electromagnetic heating and thermal expansion in the same model tree. Second, the interface and the API are isomorphic, because every setting in the model tree corresponds to an API call, so recording code produces a script exactly equivalent to manual operations. Third, the cost of agent integration is low, because an agent that understands the object model can operate every function.
That third point became a competitive asset in 2026. Most CAE vendors connecting agents must build tool wrappers per product and per version, while COMSOL needs one set.
Version 2027 and the MCP Server: Turning the Modelling Environment Into Tools
The centrepiece of version 2027 is the COMSOL MCP Server, which lets AI assistants and agents communicate with COMSOL Multiphysics through the Model Context Protocol and use large language models to create, modify, run and analyse models.
Ed Fontes argues COMSOL is particularly well suited because the whole modelling environment is controllable through a unified object-oriented API, so an agent can complete the full workflow through a single interface.
That explains the path but leaves three questions unanswered. At what granularity does the agent operate, whether it invokes solve this model or can change meshes and boundary conditions. How are errors rolled back, whether a broken model can be reverted. Who owns accuracy, whether validation responsibility sits with the engineer or the tool. Those three determine whether the MCP Server is a demonstration or a production capability.
Version 2027 also names system-level modelling, agentic AI workflows and a new Electric Motor Module. System-level modelling has long been a weak point, since COMSOL strength is three-dimensional field simulation rather than block-diagram system design, so this addition is worth watching.
Version 6.4: Two New Physics Domains and a Solver Upgrade
Explicit structural dynamics. The release adds explicit time integration for high-speed events such as impact, blast and drop tests, with robust contact handling. These problems share extremely short time scales and strong nonlinearity, including material failure, large deformation and contact separation, which implicit solvers struggle to converge. Explicit dynamics is traditional territory for LS-DYNA, Abaqus/Explicit and Radioss, and entering it means COMSOL now covers transient, strongly nonlinear regimes.
Granular Flow Module. Based on the discrete element method, it simulates grain and powder systems in conveying, mixing and additive manufacturing. This adds particulate physics to a portfolio that previously handled continuous media.
NVIDIA cuDSS. The connection to the CUDA direct sparse solver carries an official claim of several-fold speedups in single-physics and multiphysics simulations. Update 3 raised the CUDA Toolkit to 12.9.1 for full Blackwell performance and added cuDSS support on clusters plus gradient-based optimisation with cuDSS. Those last two matter: optimisation and clusters are two hard requirements of industrial workflows.
Surrogate export. Version 6.4 supports exporting deep neural network surrogate models and generating surrogate data on clusters. This is the technical interface to physical AI: high-fidelity simulation generates training data, a surrogate is trained, and the surrogate is deployed into design flow.
LLM Integration: Context Matters More Than the Model
The Chatbot window gained three checkable upgrades. It can search COMSOL documentation directly and draw on API knowledge, improving accuracy in modelling questions and generated code. It can attach model tree nodes, entire models, text files and images as context. Beyond OpenAI and Azure OpenAI, it supports DeepSeek, Google Gemini and self-hosted providers compatible with the OpenAI API, plus current OpenAI models such as GPT-5.
The second point is the real difference. Injecting the currently open model as context means the language model sees the user's actual case rather than answering generic questions about boundary conditions. That upgrades the interaction from document retrieval to diagnosis of this specific model.
Limits are clear. The Chatbot is available only on Windows. Capability depends on the chosen model provider. There is no guarantee mechanism for correctness.
Layering Credibility
Verifiable: the version 2027 and 6.4 feature lists; the existence and technical description of the MCP Server; the cuDSS integration and version; the models and platform limits supported by the Chatbot; and the positioning of new modules.
Partially verifiable: the several-fold speedup, which has no public benchmark; the claim that an agent can complete the full workflow through a single interface, which has a CTO statement but no third-party reproduction; and the 200,000 user figure, which is an estimate.
Not verifiable: revenue and margin, absent audited statements; production reliability of the MCP Server; and accuracy retention when exporting surrogate models.
Product and Commercial Model
The product structure is one platform, a set of modules, and three deployment tools.
The platform is COMSOL Multiphysics itself, including the Model Builder for modelling, the Application Builder for creating simulation apps, and the Model Manager for model and data management.
Modules divide by physics domain: electromagnetics including AC/DC, RF, optics and plasma; structural mechanics; acoustics; fluid flow; heat transfer; chemical engineering; and electrochemistry, plus specialty modules for batteries, corrosion, microfluidics and semiconductors. Pricing combines a base platform with modules purchased as needed, each charged separately.
Deployment tools are COMSOL Compiler, which compiles apps into standalone executables, and COMSOL Server, which distributes apps inside an enterprise. This is where COMSOL differs from most CAE vendors: it lets customers encapsulate simulation capability into apps for non-experts.
That capability has two commercial meanings. Internally, it turns simulation from an expert tool into an organisational capability. Externally, it lets customers embed simulation in their own products and processes, for example when an equipment maker gives customers a simulation app for configuration.
Sales model. Direct sales dominate, with 16 offices covering major markets and distributors elsewhere. There is no large channel expansion and no aggressive subscription conversion; licensing has long combined perpetual licences and maintenance with subscriptions. That explains the near 50 percent net margin, since customer acquisition and migration costs stay low.
Pricing opacity. As with every CAE vendor, unit prices are not published. What can be confirmed is module-based pricing with significant academic discounts.
Competitive Landscape
Against Ansys inside Synopsys, this is a contest of technical routes. Ansys brings a portfolio of solver brands including Fluent, Mechanical, HFSS and Maxwell, plus industrial validation depth, and now an EDA entry point through Synopsys. COMSOL brings native multiphysics coupling and a unified API. In strongly coupled scenarios such as electro-thermal-mechanical, fluid-structure, or electrochemical-thermal problems, COMSOL user experience is generally considered more coherent.
Against Siemens including Altair, this is a contest of scale and ecosystem. Siemens has the full Simcenter line, automation hardware and factory data, plus Altair optimisation and discrete elements. COMSOL cannot compete on scale, but the consistency of a single codebase has distinct value.
Against Cadence including MSC and BETA CAE, a new entrant arrives. With Hexagon D&E, Cadence gained Nastran and Adams in 2026 and entered structures and multibody for the first time. Its advantage is chip customers and electronics simulation; COMSOL advantage is multiphysics depth. The two will meet directly in thermal-mechanical-electrical coupling of electronic systems.
Against MathWorks, the boundary runs between simulation and control. MATLAB and Simulink are strong in system-level and control work, while COMSOL is strong in three-dimensional fields. The overlap is co-simulation between system and field level, currently handled through standards such as FMI rather than a single platform.
Against NVIDIA, the relationship is toolchain dependency rather than competition. COMSOL uses cuDSS for acceleration and CUDA libraries for surrogate training, but it sells no compute and builds no world models. The relationship is purely that of an upstream supplier.
Against open-source and cloud-native CAE, price and flexibility apply pressure. Open solvers such as OpenFOAM and FEniCS, and cloud platforms such as SimScale and Rescale, compete on price and elasticity, but their multiphysics coupling completeness still trails COMSOL.
Financing and Capital Structure
COMSOL has no financing history. Founded in 1986, it remains private, unlisted and without venture funding. That has been a deliberate choice for four decades, keeping full control with the founders and allowing expansion at their own pace.
Financial characteristics, based on media estimates rather than audited data. Revenue of about 200 million dollars, close to half of it pure profit, about 650 employees, and therefore roughly 3 million dollars of revenue per employee on this report's calculation. That ratio exceeds most CAE vendors and reflects high unit prices with low channel cost.
Capital allocation. No large acquisitions, in contrast with Siemens at 10.6 billion dollars for Altair, Synopsys at 34.9 billion for Ansys and Cadence at 2.7 billion euros for Hexagon D&E. Research is internal. Product line expansion is therefore slower, and integration risk is zero.
Risk profile. The benefit of staying private is freedom from quarterly guidance, which allows long-horizon investment such as the MCP Server, infrastructure that cannot pay back within a year. The cost is the absence of financial transparency, which leaves outsiders unable to assess real profitability and growth.
Valuation note. Without public statements, any COMSOL valuation is speculative. A usable anchor is that mature CAE companies often trade at five to ten times sales, so 200 million dollars of revenue at 40 to 50 percent net margin implies a range of one to two billion dollars. That is this report's rough arithmetic, not a valuation conclusion.
DISCREPANCIES
1. Headcount appears in three versions. (Source grade: OFFICIAL versus DB) The company's fortieth anniversary blog says more than 650. Third-party databases give 600 as of July 2026 and about 660 as of August 2026. The differences reflect timing and whether distributors are counted. This report uses about 650.
2. Revenue exists only as media estimates, and one database range is unusable. (Source grade: MEDIA versus DB) PLM and ERP News estimates about 200 million dollars, while a database entry gives a range of one to ten billion, too wide to carry information. The company discloses nothing. This report uses 200 million and labels it an estimate.
3. The 200,000 user figure lacks official confirmation. (Source grade: MEDIA) The number comes from industry media without clarifying whether it counts cumulative registrations, academic users or paid seats. This report treats it as an estimate of about 200,000 users and notes the relationship to the revenue figure.
4. The version naming system changed in 2026. (Source grade: OFFICIAL) The previous sequence ran through 6.x, including 6.4 and 6.4 Update 3, while 2026 saw the name version 2027. These are not iterations of the same sequence; version 2027 introduces an annual naming convention. Some coverage treats version 2027 as the successor release to 6.4, which invites confusion.
5. The delivery status of version 2027 is described two ways. (Source grade: OFFICIAL versus MEDIA) The company homepage shows a countdown to version 2027 describing it as coming soon while also linking a version 2027 press release. Industry media describe the company as having recently rolled out the 2027 platform, with the CTO presenting at a launch event in Cambridge. This report treats it as released and rolling out, and notes both descriptions.
6. The claim that the MCP Server lets an agent run the full workflow is a vendor statement. (Source grade: OFFICIAL) It comes from a CTO interview with no third-party reproduction and no description of error handling, rollback or accuracy responsibility. This report keeps the claim and lists the three open questions.
7. The claim of technical leadership among the big three is a media characterisation. (Source grade: MEDIA) It appears in an opinion piece in an industry outlet rather than in any evaluation body's conclusion. This report does not adopt it and states only the checkable capability differences.
Risks and Diligence Gaps
Scale constraint comes first. About 200 million dollars of revenue limits absolute research spending. In years when AI infrastructure, meaning compute, models and data, requires sustained large investment, a smaller company must either focus, as COMSOL has for forty years, or be surrounded on ecosystem grounds by better capitalised rivals.
MCP Server production readiness remains unverified. The value of agents calling simulation tools depends on complete tool definitions, error handling and rollback, and verifiable results. COMSOL has disclosed only that a single interface can run the full workflow, without engineering details.
Surrogate accuracy responsibility is unresolved. Version 6.4 supports exporting DNN surrogates, but a surrogate fails silently outside its training range. COMSOL has published no error evaluation method or applicability boundary, which is a hard constraint in signoff scenarios.
Talent and geography are concentrated. About 650 people spread across 16 offices, with core research in Sweden and the United States, create higher key-person dependence than rivals with tens of thousands of staff.
The cost of not acquiring. Internal expansion limits the product line. In adjacent areas such as system-level modelling, optimisation and data management, COMSOL trails rivals who filled those gaps through acquisitions.
Diligence gap list, ordered by importance. First, actual revenue, margin and growth, which can be verified through Swedish parent company filings or credit reports. Second, the ratio of paid seats to academic users. Third, the tool definition granularity, error handling and rollback of the MCP Server. Fourth, the accuracy evaluation method and applicability boundary for exported DNN surrogates. Fifth, the concrete form of system-level modelling in version 2027, including whether FMI or FMU co-simulation is supported. Sixth, customer adoption of the new Electric Motor Module. Seventh, China revenue share and localisation strategy. Eighth, the technical depth of the NVIDIA collaboration, specifically whether it is limited to cuDSS calls.
Composite Assessment
techScore 8.0. Multiphysics coupling depth is the industry benchmark, the isomorphism between interface and API that comes from a single codebase gives a structural advantage in agent conversion, and version 6.4 filled two important gaps in explicit dynamics and granular flow. The deduction reflects long-standing weakness in system-level modelling and an unclear mechanism for surrogate accuracy responsibility.
bizScore 7.5. About 200 million dollars of revenue at close to 50 percent net margin with a stable direct sales system is an exceptionally high-quality business. The deduction reflects small scale, limited growth elasticity and the absence of product line expansion through acquisition.
fundScore 8.0. No external funding, no debt pressure, self-funding cash flow and forty years of private independence. The deduction reflects the lack of financial transparency, which prevents outside verification.
productScore 8.0. The combination of platform, modules, Compiler and Server lets simulation capability be encapsulated and distributed, which most CAE vendors cannot do. The deduction reflects complex module pricing and opaque total cost of ownership.
marketScore 7.0. A technical benchmark in the multiphysics niche with a loyal user base. The deduction reflects revenue roughly one fifteenth of the Ansys multiphysics business and a weak position in system-level modelling and EDA entry points.
References
- COMSOL Multiphysics 6.4 release highlights, official: explicit structural dynamics for impact, blast and drop with robust contact handling; a Granular Flow Module based on the discrete element method; the NVIDIA cuDSS direct sparse solver with several-fold speedups in single-physics and multiphysics simulations; automatic surrounding domain creation; quad-dominant and swept meshing improvements; spatially varying transparency; and array-based plot layouts
- Same source: optional LLM-assisted simulation; the Chatbot window supporting GPT-5, DeepSeek, Google Gemini and OpenAI-compatible self-hosted providers; direct retrieval from COMSOL documentation with additional API knowledge; attachment of model tree nodes, entire models, text files and images as context; and Windows-only availability
- Same source: DNN surrogate model export, surrogate data generation on clusters, batch and cluster studies using models stored in a Model Manager database, new optimisation options for time-dependent and parametric studies, and frequency- and time-dependent uncertainty quantification
- COMSOL version 6.4 Update 3 release notes: CUDA Toolkit raised to 12.9.1 for Blackwell GPU performance, cuDSS support on clusters including Cluster Computing and Cluster Sweep, improved CPU-side matrix factorisation performance under cuDSS, gradient-based optimisation with cuDSS, default row preordering or matching, and internal time-derivative tolerance handling improving fluid-structure interaction stability
- COMSOL homepage: countdown to version 2027 and the version 2027 press release link; version 2027 described as including system-level modelling, agentic AI workflows and a new Electric Motor Module; the product suite of Multiphysics, Compiler and Server; and company facts including founding in 1986, 16 offices and more than 700 employees in the quick facts block
- COMSOL blog, Celebrating 40 Years of Multiphysics Simulation: founded in Stockholm in 1986 by Svante Littmarck and Farhad Saeidi, 16 offices, more than 650 people, the first Simulation Summit in Santa Clara in 2026, and the 2026 COMSOL Conference tour
- PLM and ERP News, COMSOL Takes the Technical Lead Among the Big Three in Multiphysics Simulation: CTO Ed Fontes on the MCP Server and the unified object-oriented API; 2026 estimates of about 3.0 billion dollars for the Ansys multiphysics business, about 1.6 billion for Siemens including Altair consolidation and about 200 million for COMSOL; about 200,000 users globally; and the statement that nearly half of revenue is profit
- Third-party database records: COMSOL headcount of 600 in July 2026 and about 660 in August 2026, plus an unusable revenue range estimate addressed in DISCREPANCY 2
- COMSOL product structure: COMSOL Multiphysics with Model Builder, Application Builder and Model Manager, add-on modules by physics domain, and app deployment through COMSOL Compiler and COMSOL Server
- Industry media coverage of the version 2027 launch event: the CTO presenting in Cambridge and version 2027 described as recently rolled out, addressed in DISCREPANCY 5
- Peer scale references on media estimates: about 3.0 billion dollars for the Ansys business inside Synopsys and about 1.6 billion for Siemens including Altair in 2026, used to frame COMSOL relative size
- COMSOL fortieth anniversary founding team remarks that employees built everything and that every single person is needed, corresponding to the organisational choice of internal research over large acquisitions
This report is compiled from public sources for informational purposes only and does not constitute investment advice. Data is current as of 25 October 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
- 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
- MathWorks Deep Dive: MATLAB/Simulink and Simscape Multiphysics Power Physical AI, with R2026a Copilots and an Agentic Toolkit for Edge Intelligence
- 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.