Legal name: Silurian AI, Inc.
Silurian Deep Dive: A Generative Forecasting Transformer Earth-System Foundation Model for Weather Risk on Energy and Infrastructure Assets
Silurian (US) builds an Earth-system foundation model with a 1.5B-parameter Generative Forecasting Transformer (GFT), serving energy and infrastructure weather risk, with NOAA and Hydro-Quebec partnerships.
Executive Summary
Silurian is an American weather AI company founded in 2024 and headquartered in Kirkland, Washington, a participant in Y Combinator's Summer 2024 batch. The founding team came from Microsoft AI and worked on Aurora, Microsoft's atmospheric foundation model: Cristian Bodnar as chief scientist, Jayesh K. Gupta as CEO and Nikhil Shankar as chief engineering officer, with Mark Baum part of the launch team who according to GeekWire later left and should not be confused with current leadership.
The product is the GFT family, the Generative Forecasting Transformer. The global model has 1.5 billion parameters and, per 2024 reporting, roughly 11 kilometre resolution with forecasts out to two weeks. The regional model GFT-US launched in April 2025 for the contiguous United States, which the company describes as roughly 3 kilometre resolution, hourly updates, median delivery about one minute after the hour and about 20 minutes earlier than NOAA's HRRR. The product form is the Earth API, offering global forecasts over land and sea, hourly data, Python and TypeScript SDKs and a browser playground, with variables including 100 metre wind, downward surface solar radiation, snowfall and precipitation type.
Funding information comes in three versions: one source cites about 11 million dollars raised in a single seed round with Lux Capital and Tola Capital; another cites a 6 million dollar seed closed in May 2026; a third lists 11 million raised across one round. Headcount is 11 to 50 in one source and 13 in another. These differences appear in the discrepancies section.
In physical AI terms Silurian's claim is to apply the foundation model approach that transformed language and vision to physical simulation of the Earth system. It does not solve the equations of numerical weather prediction but learns atmospheric evolution from vast observational and reanalysis data, generating future states from the current state. The company calls its approach a physics foundation model, though public material does not disclose architecture details or the specific form of physical constraints, which assessment must flag.
One framing point matters. This is a company with the strongest technical pedigree and the thinnest commercial evidence in this series. Every accuracy conclusion comes from the company's own evaluation with no independent third-party reproduction, there are no published quantified customer results, and the endpoints listed in API documentation are not all confirmed generally available. Compared with Jua in the same field it is earlier, with a dozen employees and funding in the low tens of millions, more focused on the United States and more oriented to APIs and developer experience, but lacking Jua's customer roster of TotalEnergies, Shell and EDF and its agent product. The key question is not whether the model can beat ECMWF but whether anyone will pay, and how much.
Company Overview
Team and Origins
The core asset is the team. All three publicly listed founders worked on Aurora, Microsoft's atmospheric foundation model, which gave them three things: engineering experience with large atmospheric datasets, methodology for training machine learning forecast models, and awareness of turning model output into actionable information.
A common misreading needs correcting: the Microsoft background does not make Silurian a subsidiary, spinout or officially endorsed product. Public sources establish no such relationship. The background is evidence of technical capability, not a guarantee of product quality.
The company was founded in 2024 and joined YC's summer batch, headquartered in Kirkland in the Seattle metropolitan area. The location has practical rationale, since Microsoft and the University of Washington have accumulated strength in atmospheric science and AI, and Seattle is dense with cloud computing and earth science talent.
Product Lines
The product has three layers. The model layer is the GFT global model and GFT-US regional model. The interface layer is the Earth API with global forecast data, hourly and daily endpoints, past forecasts, portfolio endpoints, cyclone endpoints and experimental US regional endpoints. The tooling layer is a browser playground and SDKs.
There are two commercial propositions. The first is custom foundation models built on a client's proprietary data and infrastructure, tailored to its specific weather and climate challenges. The second is rapidly refreshing regional forecasts serving risk management in energy, agriculture, insurance, logistics, aviation and defense.
The two differ in nature. Custom models are project-based with high value per account and few customers, while regional forecast APIs are standardized with usage-based pricing and many customers at low value each. Pursuing both suggests the company is still seeking product market fit, since the more common path is to prove one and then replicate.
Funding and Scale
Cumulative funding is about 11 million dollars including Lux Capital and Tola Capital, with headcount between 11 and 50. That is early for the weather AI field: Jua has raised about 27 million with 40 to 200 people, WindBorne more including hardware, and Tomorrow.io over 200 million including satellites.
Eleven million dollars supporting a team of about a dozen through foundation model training and API productization demands high capital efficiency. The public availability of reanalysis data such as ERA5 and the modest scale required because atmospheric degrees of freedom are far below language make it feasible, while resilience remains limited.
Technology
What Generative Forecasting Means
Generative in the weather context is easily misread as inventing weather. The actual meaning is that the model generates future atmospheric states from a learned distribution, constrained by training data statistics, initial conditions and physical consistency. The mathematical form resembles conditional generation in language models, but the source of constraint differs: language models are constrained by corpora and weather models by physics and observations.
The technical route has three steps. Encode the current atmospheric state into a high-dimensional representation, typically with a Transformer or graph neural network over a spherical or icosahedral grid. Integrate forward in latent or state space through rollout, generating the next state at each step. Decode back to physical variables including temperature, wind, humidity, pressure and radiation.
The advantage over numerical prediction is speed: once trained, a global forecast completes in minutes on one or a few GPUs while a numerical model needs a supercomputer for hours. The disadvantage is extrapolation, since models may fail outside the training distribution such as extreme events or a climate-shifted new normal, and error accumulates along the rollout, a known weakness of these models.
Specifications of GFT and GFT-US
The global GFT has 1.5 billion parameters and per 2024 reporting roughly 11 kilometre resolution with a two week horizon. Against international peers at about 0.25 degrees or 28 kilometres such as GraphCast, Aurora and Pangu-Weather, 11 kilometres is a finer tier, though nominal and effective resolution differ, since a finer output grid does not create information absent from training data.
GFT-US targets the contiguous United States with roughly 3 kilometre resolution, hourly updates, median delivery about one minute after the hour and about 20 minutes earlier than NOAA HRRR. These three metrics have different natures: resolution is grid spacing, delivery time is when a forecast becomes available, and accuracy is closeness to observations. They cannot substitute for each other, and a 3 kilometre grid does not imply hyperlocal accuracy, especially in mountainous terrain, near coastlines and around urban heat islands.
The claim of being 20 minutes earlier than HRRR has clear commercial meaning. HRRR is the United States high-resolution rapid refresh model running hourly and widely used in aviation, energy and emergency response. If a commercial company can deliver an equal or better forecast 20 minutes earlier, time-sensitive operations such as aviation scheduling and power trading gain value. Being earlier is not the same as being more accurate, and the two dimensions require separate validation.
Degree of API Productization
The Earth API documentation lists a fairly complete set of endpoints and variables: hourly and daily forecasts, past forecasts, portfolio endpoints, cyclone forecasts and an experimental US regional endpoint, with variables including temperature, feels-like temperature, precipitation accumulation and probability, snowfall, cloud cover, humidity, wind speed and direction, pressure, dew point, downward solar radiation and 100 metre wind.
That list shows a higher degree of productization than a pure research team, with SDKs, a playground and developer documentation. Listing endpoints does not mean they are generally available, and authentication, quotas, commercial licensing and service levels all require confirmation with the company. Pricing and paying customer counts are undisclosed.
The choice of 100 metre wind and surface solar radiation signals the target market in wind and solar output forecasting. This matches Jua's entry point and shows that weather AI commercialization is concentrating in energy, because energy markets settle daily and forecast accuracy converts to money immediately.
The Evidence Gap on Physics Foundation Models
The company calls its work a physics foundation model, which implies the model encodes physical constraints such as conservation laws, boundary conditions and radiative transfer rather than merely fitting data.
Public material does not specify the form of those constraints: whether physical loss terms constrain training, whether physical quantities are hard-coded into network structure, or whether post-processing enforces physical consistency. The three approaches differ greatly in technical substance and effect, and without disclosure the label cannot be distinguished from marketing.
This is not pedantry. A central debate in weather AI is whether purely data-driven models extrapolate beyond the training distribution, and physical constraints are the main means of improving extrapolation. A company claiming a physics foundation model without disclosing constraint form can only be assessed as data-driven from the outside.
Technical Shortfalls
First, independent validation is absent. All accuracy conclusions come from company-run evaluation with no third-party reproduction by meteorological services, academic groups or independent benchmarks. This is common for early companies but means the claim of surpassing United States and European government agencies rests on company endorsement alone.
Second, extreme event capability is undisclosed. What energy and insurance care most about is extreme weather such as storms, cold snaps, heat waves and prolonged calm, where training samples are sparse and AI extrapolation is a recognized weakness. No dedicated extreme event evaluation has been published.
Third, error accumulation is undisclosed. Rollout error grows with lead time, and whether two week errors remain usable requires error curves by variable and region. Public information lacks those data.
Fourth, regional model generality. GFT-US is trained for the contiguous United States, and transfer to Europe or Asia requires retraining or fine-tuning with unknown cost and effect. The current product focus on the United States is both focus and limitation.
Product and Commercial Model
Logic of the Two Lines
The custom foundation model line reasons that large energy companies, insurers and logistics firms hold their own historical data in the form of private weather stations, asset operating records and claims data, which is noise to a general model and gold to a custom one. Fine-tuning a foundation model on customer data yields forecasts better fitted to their assets.
The commercial advantage is high price and strong stickiness, since a model trained on customer data is costly to replace, while the disadvantage is project-based delivery that does not replicate and long sales cycles. For a company of a dozen people, a few custom projects saturate engineering resources.
The regional forecast API line reasons toward standardization, with developers calling on demand and paying by usage at low marginal cost. It scales, but competition is intense because NOAA forecasts are free, ECMWF data is cheap and Jua and Tomorrow.io already operate there. The only differentiation is more accurate, faster or easier to use, and more accurate requires independent validation.
Relationship With Government Forecasts
NOAA's GFS and HRRR and ECMWF's IFS and AIFS are public goods with free or low-cost data. Commercial weather companies survive by providing what governments do not: higher resolution, faster updates, industry-specific post-processing, better API experience or custom services.
Silurian's positioning is exactly these: 3 kilometres, hourly, one minute delivery, energy and agriculture oriented variables, SDKs and a playground. These are product-level differentiation rather than scientific breakthroughs. Developer friendliness and industry fit have value, but pricing power requires customers to agree that the improvement over free data is worth paying for.
Unknown Customers and Revenue
The company has published no customer list, customer count or revenue. That is normal at seed stage but means product market fit cannot be judged. Observable indirect signals are three: API documentation completeness showing productization, the YC background showing financing and mentoring resources, and team size around 13 showing early stage.
Compared with Jua, which disclosed more than ten leading energy customers, 100 gigawatts of assets, a Belgian case with a 20 percent balancing cost reduction, and an open StationBench methodology to raise credibility, Silurian is blank on all three. That does not mean it is worse, only earlier.
Competitive Landscape
The First Tier of International Weather AI
Google DeepMind's GraphCast and GenCast, Microsoft Aurora, NVIDIA FourCastNet and Huawei Pangu-Weather form the first tier. Most are open, publish papers and have abundant compute, with the common weakness of not being productized: a model in a paper comes without service levels, APIs or customer support.
Silurian's position is to come out of that first tier and build product. The founders worked on Aurora and understand the engineering details, and the company builds APIs and regional models to fill the productization gap. The path resembles Jua's, except Jua entered earlier, has more customers and an agent layer, while Silurian focuses more on the United States and developer experience.
Comparison With Jua
The two overlap heavily, both using AI models for global and regional forecasts, both targeting energy, both offering APIs. They differ on four dimensions.
Market focus: Jua concentrates on Europe with customers including Statkraft, RWE, EDF, Enel and ESB, while Silurian concentrates on the United States with GFT-US and comparisons to United States government models. Product form: Jua has the Athena agent with natural language, position monitoring and backtesting, while Silurian is pure API and SDK. Evidence strength: Jua has a customer roster and a single-case effect figure with open StationBench evaluation, while Silurian has only self-evaluation. Capital and scale: Jua about 27 million dollars and 40 to 200 people, Silurian about 11 million and around 13 people.
The comparison suggests Silurian is earlier, lighter and more infrastructure-oriented while Jua is more mature, heavier and more application-oriented. Both paths can work with different risks: infrastructure faces substitution by free government data, and applications face high customization cost.
Competition With Commercial Weather Services
Traditional commercial weather providers including DTN, Meteomatics, AccuWeather and Tomorrow.io have customer relationships and historical data with weaker model technology. Tomorrow.io takes the heavy asset route of satellites plus models plus platform, the opposite of Silurian's light approach.
The light approach offers capital efficiency and fast expansion but no data moat, since customers can switch if model accuracy is matched. Defense must come from sustaining accuracy leadership or embedding deeply in customer workflows, the former through research and the latter through product.
The Subtle Relationship With Government Agencies
NOAA and ECMWF are both the objects to surpass and the suppliers of data. AI weather models typically initialize from numerical analysis fields and take observations from public meteorological infrastructure. Commercial companies therefore depend on public institutions along the data chain while advertising that they exceed them.
The relationship is stable in the near term because open data is policy, but carries long-term policy risk if agencies tighten access or charge for commercial use, which would change the cost structure.
Financing and Capital Structure
Three Versions of the Funding
Public information is inconsistent. One source cites about 11 million dollars raised in a single seed round with Lux Capital and Tola Capital, another cites a 6 million dollar seed in May 2026, and a third lists 11 million raised across one round, implying only one.
The plausible reading is a YC standard investment in 2024, typically around 500,000 dollars, a 6 million dollar seed in 2026 and possibly earlier SAFE or bridge instruments totaling about 11 million. That is inference and unconfirmed.
Capital Efficiency and Needs
Eleven million dollars across 13 people is about 850,000 dollars per head, lean for a foundation model company. Weather model training costs far less than language model training because data is public and model scale is smaller, which is why a first-tier model is achievable on small capital.
Funding needs depend on progress across both lines. Custom models require engineering for data integration and fine-tuning per customer, while the API requires infrastructure for GPU inference, data storage and global content delivery. Pursuing both, 11 million covers roughly 18 to 24 months.
Valuation and Exit Paths
Valuation is undisclosed. At a 6 million dollar seed with typical 15 to 20 percent dilution, post-money valuation would fall between 30 and 40 million dollars, but that is inference.
Exit paths are acquisition by a weather services provider, energy data company or cloud vendor, or absorption into a larger company. An independent listing is unrealistic for a weather AI company of a dozen people. That framing matters for investment logic: the purchase is an option on a technical team and model assets rather than a discounted long-term cash flow.
DISCREPANCIES
1. Three versions of the funding amount. One source cites about 11 million dollars in a single seed round with Lux Capital and Tola Capital, another a 6 million dollar seed in May 2026, and a third lists 11 million across one round. The plausible inference is a 2024 YC standard investment plus a 2026 seed plus possible SAFE instruments totaling about 11 million, but the company has not confirmed. This report lists all three and labels the inference. Source tier: third-party databases (MEDIA), with no OFFICIAL disclosure.
2. Two headcount figures. One records 11 to 50 employees and another 13. The range is a classification and 13 is a specific figure, so they are consistent with 13 falling inside, but precision differs. This report uses around 13. Source tier: third-party databases (MEDIA).
3. Changes in the founder list. Public sources list four founding members including Cristian Bodnar, Jayesh K. Gupta, Mark Baum and Nikhil Shankar, while the current YC page lists three. GeekWire reported that Baum left. This report uses three current leaders with one launch member having departed. Source tier: OFFICIAL (YC page) crossed with MEDIA (GeekWire).
4. Timing of the resolution and horizon figures. The 11 kilometre and two week figures come from 2024 reporting and describe the model public at that time, while current API configuration may differ. This report labels the timing and does not equate them with all current products. Source tier: MEDIA (2024 reporting).
5. Three GFT-US metrics are conflated. The company cites about 3 kilometre resolution, hourly updates, median delivery about one minute after the hour and about 20 minutes earlier than NOAA HRRR. Resolution, delivery time and accuracy are different dimensions, and coverage often conflates them by calling 3 kilometres hyperlocal accuracy. This report separates the three in the technology section. Source tier: OFFICIAL (company statements) contrasted with MEDIA.
6. Strength of evidence for surpassing United States and European agencies. The company says GFT forecasts can exceed those of United States and European government agencies. The conclusion comes from company-run evaluation with no independent reproduction, and the variables, regions, lead times and metric weights are not fully disclosed. This report retains the claim labeled as company framing. Source tier: company self-evaluation, OFFICIAL but self-assessed, below independent third-party assessment.
7. Physics foundation model without architecture disclosure. The company calls its models physics foundation models, but public material does not specify whether physical constraints enter as loss terms, structural constraints or post-processing correction. The three differ substantially, and without disclosure external assessment can only treat the models as data-driven. This report states the gap in the technology section. Source tier: company statement (OFFICIAL) with architecture information missing.
8. Gap between API endpoints and product availability. The Earth API documentation lists global forecasts, past forecasts, portfolio, cyclone and experimental US regional endpoints. Listing does not mean general availability, and authentication, quotas, commercial licensing and service levels require confirmation. This report notes the distinction in the commercial section. Source tier: OFFICIAL (documentation) with product status unconfirmed.
Risks and Diligence Gaps
First, missing independent validation is the largest uncertainty. The claim of surpassing government agencies underpins pricing and the funding narrative yet rests on self-evaluation. If third-party evaluation reaches different conclusions, the commercial story takes a material hit. Diligence should request evaluation scripts, test set construction, raw scores by model and variable weights.
Second, customers and revenue are entirely blank. No customer list, count or paid scale is public. That is normal at seed stage but prevents judging product market fit, leaving only API documentation completeness and team size as verifiable indirect signals.
Third, the substitution relationship with free government data. NOAA and ECMWF data are free or cheap, so commercial companies must offer an increment worth paying for. Three kilometre resolution and one minute delivery are product differentiation, but whether customers pay depends on the economic value of those increments in their decisions.
Fourth, extreme event capability is unvalidated. The core risk in energy and insurance comes from extreme weather where AI models extrapolate weakly on sparse samples, so the absence of extreme event evaluation leaves applicability in high-value scenarios unknown.
Fifth, scalability of the custom model route. Each custom customer requires data integration and fine-tuning with heavy engineering consumption. A team of a dozen saturates after a few projects, and scaling requires standardizing the customization process, which takes time.
Sixth, limits of regional focus. GFT-US targets the contiguous United States, and expansion to Europe or Asia requires retraining or fine-tuning. Jua already has a European customer base, so entering Europe means facing an incumbent.
Seventh, team stability. Launch member Mark Baum has left, indicating early churn. In a company of a dozen, loss of core researchers has outsized impact.
Composite Assessment
Technology scores 7.5. The founding team came from Aurora with hands-on experience training atmospheric foundation models, GFT specifications at 1.5 billion parameters, 11 kilometres and two weeks are comparable with the international first tier, GFT-US at 3 kilometres and one minute delivery is engineering-demanding, and the Earth API is more productized than a pure research team. Deductions for undisclosed physical constraints, missing independent validation and absent data on extreme events and error accumulation.
Business scores 6.0. Two lines covering high-value custom work and scalable regional forecasts, variable selection aligned with energy needs and APIs with SDKs lowering barriers. Deductions for completely blank customers and revenue, competition with free government data and hard-to-scale customization.
Finance scores 6.0. About 11 million dollars across 13 people is high capital efficiency, since weather model training costs less than language model training. Deductions for small absolute capital, limited resilience and undisclosed valuation and revenue.
Product scores 7.0. The Earth API endpoint and variable coverage is complete with Python and TypeScript SDKs and a playground, and GFT-US update frequency and delivery time reach business-usable levels. Deductions for unconfirmed endpoint availability, undisclosed pricing and the absence of an industry application layer such as Jua's Athena.
Market scores 7.0. Weather AI is moving from papers into products, energy transition raises weather sensitivity, and United States power and aviation have clear demand for high-resolution rapid forecasts. Deductions for intense competition from Jua, Tomorrow.io and free government data, a finite market and the time required to build customer trust.
Overall 6.7. This is the company with the best technical pedigree and the least commercial evidence in this series: a team out of Microsoft Aurora built a model comparable with the international first tier on a dozen people and tens of millions of dollars, and turned it into a developer-usable API. Yet every accuracy conclusion is self-assessed, customers and revenue are blank and physical constraints are undisclosed. The judgment on physical AI is that the technical route of learning atmospheric physics with foundation models is the mainstream direction for earth system AI and that the team's execution is evidenced by model specifications, while the distance from building a model to selling a product still requires independent validation, customer cases and a sustainable business model, none of which exists yet.
References
- Third-party company profile, neuronfeed: Silurian founded 2024 with 11 to 50 employees and about 11 million dollars raised in a single seed round with Lux Capital and Tola Capital and a YC background; positioned as building foundation models to simulate Earth's systems starting with weather and climate; founding team including CEO Jayesh K. Gupta and co-founders Cristian Bodnar, Mark Baum and Nikhil Shankar with backgrounds at Microsoft and Google DeepMind; GFT producing long-range forecasts the company says can outperform United States and European government agencies; two value lines of custom foundation models on client data and rapidly refreshing regional forecasts; target sectors including energy, agriculture, insurance, logistics, aviation and defense
- Third-party company profile, startuphub.ai updated September 26, 2026: headquartered in Kirkland, founded 2024, 13 employees, 6 million dollars of seed funding in May 2026, B2B model; describing GFT as a long-range forecast model integrating atmospheric, land-surface and infrastructure data
- Technical and product analysis, iTechFAQs: founders Bodnar, Gupta and Shankar as former Microsoft AI researchers who worked on Aurora; the company is not a Microsoft subsidiary, spinout or officially endorsed product; GFT as a 1.5 billion parameter global model with 2024 GeekWire reporting of roughly 11 kilometre resolution and two week horizon; Earth API covering land and sea with hourly data, Python and TypeScript SDKs and a browser playground, with variables including 100 metre wind, surface solar radiation, snowfall and precipitation type; endpoints including hourly and daily forecasts, past forecasts, portfolio, cyclone and experimental US regional; GFT-US announced April 2025 for the contiguous United States with roughly 3 kilometre resolution, hourly updates, median delivery about one minute after the hour and about 20 minutes earlier than NOAA HRRR; Mark Baum part of the launch team who later left
- Key judgments in the same analysis: company benchmarks are self-run and do not independently prove superiority by region, variable or decision; resolution, delivery time and accuracy are three different dimensions and no single one proves business value; a 3 kilometre grid does not equal hyperlocal accuracy, especially in mountains, near coastlines and around urban heat islands; generative means producing forecast states from learned patterns rather than inventing weather; the company calls its models physics foundation models while public material does not fully specify architecture or physical constraints
- Technical analysis, MacMyths: confirming Bodnar, Gupta and Shankar as founders, founding in 2024, the YC summer batch and the Kirkland headquarters; confirming GFT at 1.5 billion parameters with 2024 reporting of about 11 kilometres and two weeks; emphasizing that AI models depend on initial conditions, training data, calibration and evaluation quality
- Y Combinator company page as cited by third parties: Silurian in the Summer 2024 batch, headquartered in Kirkland, with Bodnar, Gupta and Shankar currently listed
- GeekWire reporting as cited by third parties: GFT global model at roughly 11 kilometre resolution and two week horizon under 2024 model definitions; Mark Baum having left the company
- Industry background from public sources: NOAA GFS and HRRR and ECMWF IFS and AIFS are public forecast products with free or low-cost data; HRRR is the United States high-resolution rapid refresh model running hourly; the first tier of AI weather models includes Google DeepMind GraphCast and GenCast, Microsoft Aurora, NVIDIA FourCastNet and Huawei Pangu-Weather, most at about 0.25 degrees
- Competitor information from public sources: Jua has raised about 27 million dollars with 40 to 200 people, customers including TotalEnergies, Shell, Enel, Statkraft, RWE and EDF, an Athena agent product and evaluation through the open StationBench method; Tomorrow.io takes a hardware-heavy satellite plus model plus platform route with over 200 million dollars raised; DTN, Meteomatics and AccuWeather are traditional commercial weather providers
- Technical background from public sources: AI weather models typically encode the current atmospheric state into a high-dimensional representation, integrate forward in latent or state space and decode to physical variables; once trained a global forecast completes in minutes on a few GPUs; main weaknesses are extrapolation outside the training distribution such as extreme events and rollout error growing with lead time
- Commercial background from public sources: weather AI commercialization concentrates in energy including wind and solar output forecasting and power trading, because markets settle daily and forecast accuracy converts immediately into money; 100 metre wind and surface solar radiation are the variables energy customers care most about
- Capital background from public sources: YC standard investment is typically around 500,000 dollars; seed rounds typically dilute 15 to 20 percent, which supports the post-money valuation range inferred in this report
This report is compiled from public sources for informational purposes only and is not investment advice. The company is private and funding and team data come from third-party databases that conflict; accuracy conclusions are company self-assessments without independent third-party reproduction, and related judgments are labeled as uncertain.
Related tools
Related cases
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
- 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
- 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.