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

Executive Summary

SimScale GmbH is a Munich company that sells engineering simulation the way utilities sell compute: you open a browser, upload a geometry, configure a physics setup, and pay for the CPU and GPU core hours your job actually consumes. The company was founded in October or November 2012 and is registered in Munich under HRB 201813, so it has been operating for fourteen years, not the two years that circulate in secondary summaries. Its legal form is a German GmbH in the UG variant, with registered capital of EUR 169,339 and a VAT identifier DE286403296, and its address has never changed. The commercial proposition is a substitution rather than a breakthrough: a customer who would otherwise buy a solver licence, install it on a workstation or cluster, and pay a seat fee, pays instead for elastic cloud capacity with no perpetual licence and no hardware.

The value of that substitution is real but it is also a well-crowded idea, so the interesting question is what SimScale has built that a commodity cloud CAE reseller could not. This report concludes there are three differentiators, and that they carry very different weight.

First, multi-solver orchestration is the load-bearing asset. The platform runs OpenFOAM-derived solvers including interFoam for multiphase, an in-house GPU lattice Boltzmann implementation, Code_Aster for structural mechanics, Simerics-MP for rotating machinery, Hexagon Marc for nonlinear structural analysis, PAMICS for smoothed particle hydrodynamics, and an in-house Multipurpose solver, all behind one job submission and one post-processing surface. That is a genuine engineering and integration asset, it is the hardest of the three to replicate quickly, and it is also the one that carries the most legal and commercial dependency risk, because Code_Aster is GNU GPL and three of the strongest solvers are licensed from third parties.

Second, the in-house GPU LBM accelerator is real engineering with unverifiable marketing. The company states that its GPU lattice Boltzmann path is 20 to 30 times shorter than standard CFD, and separately that the SPH path is 10 to 20 times faster. Neither claim publishes the model, mesh, core count, baseline solver, or error tolerance that would let a reader check it.

Third, the AI layer is genuine but subordinate, and this is the single most mis-weighted part of the story. SimScale is not an AI-native company. It is a cloud delivery and solver orchestration platform that has been layering AI components on top year by year: Physics AI (surrogate and foundation models for parametric study, first shipped September 2023), the Engineering AI agentic layer (RFQ to intent to manufacturability check to mesh and solver configuration to audited report), and a centrifugal pump Foundation AI Model built on NVIDIA PhysicsNeMo, Warp and Omniverse in March 2025, claimed at roughly 2700 times faster than traditional CFD. Each of the three is a layer on a platform that already worked without them. That distinction matters for how an investor compares SimScale with the model-first cohort.

The evidence base is asymmetric and should be read accordingly. Exactly one publicly checkable accuracy result exists: a NAFEMS FV4 cantilever beam frequency case solved with Code_Aster and Frequency Analysis on three meshes of 32,663, 118,696 and 240,757 nodes, with per-mode deviations between -0.7% and +2.6% across six modes. Everything else in the performance space is vendor self-report with no published test matrix, and there is no third-party reproduction of any claim.

The financing history is clean and must be read in two separate columns. Equity: 2013 seed, 2014 angel led by Earlybird, December 2015 Series A led by USV with HTGF joining, January 2017 Series B led by June Fund, then EUR 27 million Series C on 30 January 2020 led by Insight Partners, and EUR 25 million Series C extension on 28 October 2021 co-led by Draper Esprit and Insight Partners, which the company states brings the C round total to EUR 52 million. Debt: EUR 25 million signed with the European Investment Bank under InvestEU on 20 December 2024, project reference 20240117, against a total project cost of EUR 66 million. The two are different instruments with different consequences and are not added together here.

Commercial disclosure is the weakest link. There is no public audited financial statement, no ARR, no paid-customer count, no customer concentration data, and no price list beyond the free tier. The only revenue figures anywhere are database estimates, and the widely circulated ARR and AWS-spend numbers come from AI-generated template sites and are excluded.

SimScale remains private and unacquired. The chief executive said in 2025 that the company remains private and does not share specific financial metrics. No IPO filing, listing venue, timetable, employee option pool, or acquisition has been found.

Company Overview

The legal entity is SimScale GmbH, registered at Ridlerstrasse 31B, 80339 Munich, with the commercial register at the Amtsgericht Muenchen under HRB 201813, legal form GmbH in the UG variant, VAT identifier DE286403296, and registered capital of EUR 169,339. No source records a rename, and no source records a relocation: the founding address and the current address are the same address in Munich. The LEI record 391200DYHWXRDIJ8HK07 shows status LAPSED, which normally reflects an unpaid annual renewal rather than an operating problem, but the 2024 and 2025 German commercial register annual reports could not be retrieved, so the current share register and shareholder list should be confirmed with the register directly rather than inferred.

The founding date is documented to within a month by two independent routes. The company timeline states November 2012 and traces the initial consulting concept to 2011, with the cloud simulation platform going live in August 2013. The database Implisense records a registration date of 17 October 2012. A third date, 1 January 2012, appears only as the placeholder date of a "University Spin-Out" entry in the PitchBook timeline and reflects Technical University of Munich technology transfer, not a registration event.

The operating history since then runs as follows. December 2015 brought the free Community Plan, the USV-led Series A and the arrival of HTGF as an investor. May 2016 brought conjugate heat transfer. January 2020 brought the EUR 27 million Series C and a user count of 150,000. May 2021 brought in-browser CAD import. September 2021 brought the exclusive cloud arrangement with Simerics for rotating-machinery CFD, and October 2021 the EUR 25 million C extension. September 2023 brought AI-based Physics Simulation. October 2024 brought 600,000 cumulative users, the Hexagon Marc cooperation, and the EIB InvestEU loan. March 2025 brought the centrifugal pump Foundation AI Model on NVIDIA PhysicsNeMo, and June 2025 Parasolid and HOOPS Exchange integration plus native nTop implicit geometry import. February 2026 connected Hexagon Marc to agentic AI, March 2026 reached 800,000 users and published the State of Engineering AI 2026 report, May 2026 launched SimScale Workflows, and June 2026 opened the Simerics-MP mesher and solver free to all Community users for non-commercial work.

Headcount has seven published readings and none of them should be averaged. Implisense says 78 people in 2023, Crunchbase gives a 101 to 250 band, Dealroom 141, PitchBook and LeadIQ 143, TipRanks 144 as of 21 September 2026, and two aggregator pages say roughly 130 to 160 and roughly 150. The company's own recruiting blog gives an internal series of 5 people in 2012, 25 in 2015, 60 in 2018, 80 in 2020, more than 120 in 2023 and more than 150 in 2025, without precise dates on each figure. The defensible statement is roughly 140 to 150 people on a 2025 to 2026 snapshot, having been 78 in 2023. The 78 and 144 figures are points in a growth series, and averaging them produces a number that was never true on any date.

The founders page names five founders, all with a Technical University of Munich background: David Heiny, co-founder and chief executive, with a mathematics bachelor's and a mechanical engineering diploma from TUM and a master's in computational science and engineering from Georgia Tech; Vincenz Doelle, co-founder and chief operating officer, with a computer science master's from TUM and exchange visits to Georgia Tech and ETH Zurich; Johannes Probst, a co-founder with a TUM computer science background; Alexander Fischer, co-founder and product manager, from TUM; and Anatol Dammer, a co-founder from TUM. Insight Partners' Joshua Fredberg joined the board at the 2020 Series C.

The current team page lists Mariano Barrios as VP Engineering, Alfred Saad as chief commercial officer, Jon Wilde as VP Product Management, John Webster as VP Marketing, Elena Parangoni, Erwan Donat, Andrew Acuff at the Chattanooga base, Paul Manhas for US enterprise sales, and Dr Steve Laine leading solutions engineering. Notably, no chief technology officer is listed and no source confirms who holds that role, which matters because the technical argument rests on in-house solver work. Preqin lists Rajarshi Mandal as a co-founder; no official source supports this and it is treated as a database artefact. No patent record, journal publication list, or academic appointment outside TUM was found for any founder or executive, so the team's academic claim rests on the degrees named above.

Common Misreadings: Four Errors That Change the Analysis

SimScale was not founded in 2022. It was founded in October or November 2012, which means fourteen years of operating history, a 2015 free tier, a 2020 Series C and a 2021 extension. A reader who believes the company is two years old will read the financing table, the customer list and the product roadmap as early-stage evidence when they are in fact evidence of a long-lived business.

There is no SCALE Digital incubation relationship. SCALE GmbH, also known as SCALE.sdm, is a simulation data management and SPDM company based in Ingolstadt, in Upper Bavaria. The two names collide and nothing else connects them. No source places SimScale inside any Munich incubator.

The founders are not Thomas Rausch, Mathias Kraus, or Stefan Schroerl. No official or reliable source links those three names to SimScale. The official founders page names David Heiny, Vincenz Doelle, Johannes Probst, Alexander Fischer and Anatol Dammer.

SimScale does not put Ansys Fluent in the cloud. No source indicates that Fluent is used on the platform at all. The CFD stack is OpenFOAM-derived interFoam for multiphase plus an in-house GPU LBM implementation, structural analysis is Code_Aster with Frequency Analysis, rotating machinery is Simerics-MP, nonlinear structural analysis is Hexagon Marc, and smoothed particle hydrodynamics is PAMICS. Code_Saturne is likewise not present in any source. Anyone who models SimScale as a thin cloud wrapper around a single commercial code has the wrong competitive set and the wrong margin structure.

Technology

The central technical fact about SimScale is that it is a solver orchestration layer, and the orchestration is the product. A user submits one job to one platform and the platform decides which solver, which mesh, how many cores, how many GPUs, how to run it, and how to render the result. Underneath that sit seven distinct engines that would otherwise be seven separate products with seven separate licences.

The CFD side runs OpenFOAM-derived solvers, with interFoam named explicitly in the company's own CFD documentation for multiphase and free-surface work, alongside an in-house GPU implementation of the lattice Boltzmann method. The structural side runs Code_Aster, the EDF open-source structural solver under GNU GPL, paired with the Frequency Analysis module for modal and frequency response work. Rotating machinery CFD comes from Simerics-MP under an exclusive cloud arrangement announced in September 2021. Nonlinear structural analysis comes from Hexagon Marc, announced December 2024, generally available 9 July 2025, and connected to agentic AI in February 2026. Smoothed particle hydrodynamics comes from PAMICS, the AI Engineering GmbH solver, added March 2026. An in-house Multipurpose solver covers general cases. No source indicates Code_Saturne or Fluent on the platform.

This stack carries three consequences that a reader should price in. First, Code_Aster is GNU GPL while the platform is closed-source commercial software, so the boundary between using the solver as a service and distributing it needs legal verification; the only public related statement on the company's side is the trademark notice in its imprint. Second, three of the strongest engines are licensed in from third parties (Hexagon, Simerics, AI Engineering), so the revenue-sharing, renewal and exclusivity terms of those agreements are undeclared and represent both a margin and a concentration risk. Third, the more capable the menu, the more the platform's value depends on orchestration quality rather than on any single solver, which is actually the company's structural advantage, provided the orchestration layer remains hard to copy.

The AI stack is three separate things with three separate maturity levels, and the temptation to describe them as one platform is the main analytical error available here.

Physics AI is the surrogate and foundation model tier for parametric study and design exploration, first shipped as AI-based Physics Simulation in September 2023. The headline figure is that proxy results sit within 5% of a high-fidelity solver. The geometry family, mesh, solver, sample size and metric definition behind that 5% are not published, so it is a vendor statement relayed by media, not a benchmark.

Engineering AI is the agentic layer and its workflow description is the most concretely specified thing the company has published: natural-language RFQ and intent extraction, then CAD manufacturability checking, then automatic mesh and solver configuration, then parallel execution, then a generated and auditable engineering report. The company says the agent draws on experience from more than one million real simulation projects. It opened an enterprise waiting list on 7 May 2026 and opened the agent to community users on 18 August 2026. Alongside it, the State of Engineering AI 2026 report claims 80% of engineering teams are stuck in an AI pilot purgatory with only 9% reaching scale; this is SimScale's own survey with sample and methodology undisclosed, so it is a marketing-adjacent statistic.

The centrifugal pump Foundation AI Model, released March 2025, is the most specific AI artefact: built on NVIDIA PhysicsNeMo, Blueprint, Warp, Omniverse and Blackwell, presented as the first foundation AI model for centrifugal pump simulation, with a claimed roughly 2700 times speedup against traditional CFD. The comparison object and the equivalence of accuracy are not defined. It also creates a single-vendor dependency on one vendor's physics-ML stack, a point worth weighing against the otherwise open-source posture of the platform.

Against these claims, the verifiable ground is thin but not empty. The single public benchmark is the NAFEMS FV4 case, a cantilever beam with off-centre masses taken from the NAFEMS TNSB Rev.3 benchmark of 1990, solved with Code_Aster plus Frequency Analysis on three meshes of 32,663, 118,696 and 240,757 nodes, with per-mode deviation from the reference values between -0.7% and +2.6% across six modes. That is a grid-refinement study on a classical academic case, published with node counts and per-mode numbers, and it is the strongest thing in the technical file. It is also an in-house test on self-selected meshes in a well-understood structural regime, and it is one example. The remaining performance statements, the GPU LBM 20 to 30 times, the PAMICS SPH 10 to 20 times, the Physics AI within 5%, the pump model at roughly 2700 times, and a Hazleton Pumps customer testimonial of 99.9% quicker with no baseline defined, are all vendor self-report without a published test matrix. One figure that circulates as a SimScale claim, a 10 to 50 times GPU linear algebra speedup, comes from a third-party general GPU engineering explainer and is not a SimScale statement.

The cloud architecture is documented in one authoritative place and almost nowhere else. SimScale appeared in the AWS re:Invent 2020 HPC Workloads session material, with AWS P3 instances and NVIDIA GPUs, arguing that aerodynamic CFD could be compressed from weeks into hours. That is the only cloud-provider evidence in the file. No official statement describes the platform as running on 100% AWS or on any single cloud, and whether it is multi-cloud or uses Azure or Google Cloud was not found in public sources. The execution pattern described in product documentation is browser submission, elastic cloud HPC queue, multi-core and multi-GPU parallel solve, results returned for in-browser visualisation, and from May 2026 the same execution infrastructure opened to third-party solvers and AI methods through SimScale Workflows.

Geometry handling is where the enterprise integrations live. The platform carries the Parasolid kernel through a Siemens PLM Software partnership and HOOPS Exchange from Tech Soft 3D, both announced for 2025, plus native import of nTop implicit geometry for heat-exchanger bodies in June 2025, STL and STEP as standard entry points, and an embedded SimScale workflow inside Onshape announced on 15 September 2026. The imprint carries the OpenFOAM, Siemens PLM and nTop trademarks, which is indirect confirmation of where the toolchain comes from.

Product and Commercial Model

The pricing model is a subscription combined with a core-hour allowance. Hours are metered against CPU and GPU core hours, and each plan also caps maximum compute instance size and user count.

• Community, free: 2,000 core hours, 16 cores maximum, 1 user, 10 simulation runs, positioned as a zero-cost entry point for teaching, exploration and small validation work.
• Mechanical: 3,000 core hours, up to 192 cores, user count custom.
• Professional: 10,000 core hours, up to 192 cores, custom quota supported, user count custom.
• Enterprise: core hours, instance size and user count all custom.
• Beyond the free tier, no price figures are published anywhere. Mechanical and Professional are quote-only.

The pricing page also carries the marketing line "Trusted by 900,000+ engineers", which is the company's cumulative user counter rather than a customer count. A third-party scraped version of the page shows Community at 3,000 core hours, which reflects an older configuration, and the original 2015 Community Plan terms were a one-off 3,000 computation hours plus 500 GB of storage. Those historical values must not be mixed with the current 2,000 core hours.

The free tier is the acquisition mechanism, and the company is investing in it rather than harvesting it. In June 2026 it opened the Simerics-MP mesher and solver free to all Community users for non-commercial use, which is a meaningful commercial decision because it puts a third-party commercial solver in the hands of every free user in a non-paying context, and it signals that the volume strategy runs through the bottom of the funnel rather than through enterprise-only entry.

The product line is now four things: the Cloud Simulation platform, Physics AI, Engineering AI, and SimScale Workflows, which was introduced in May 2026 and which the company describes as opening its execution infrastructure to any solver or AI method, with a workflow being a data-linked sequence of steps. The last of these is strategically the most interesting, because it converts SimScale from a platform with a proprietary solver menu into a neutral execution substrate, which is how the company can keep relevance if model-first competitors take the physics.

Positioning is explicit: "the world's first cloud-based, AI-native simulation platform" and "the Cloud Engineering Simulation Company". The commercial argument is the substitution itself, replacing a high-priced perpetual licence with elastic compute and no seat fee, and the company maintains a comparison page against Ansys built on no installation, elastic capacity and no seat licensing. Target industries it names include automotive and transportation, industrial equipment, energy, mechatronics and medical devices, and the case material covers an electric hypercar, heat exchangers on implicit geometry, data-centre and pedestrian wind comfort, centrifugal pumps, rotating machinery, structural analysis for Harcourt Industrial, and gear-box work through the SimuTech partnership.

The customer list that can be evidenced from official case studies and the official news flow runs to roughly seventeen names: Rimac on the electric hypercar, Siemens Energy, Convion on fuel cells and power generation, Silent-Aire, ITW Automotive, Buhler, Hazleton Pumps, Withings, Catalytic Consulting, Varisco, Dolphin Global, Dexory, SunCubes, Harcourt Industrial, Cloud 9 Surf Foils, QRC Technologies, and Tokyowheel. Five further names, Vale, Emrgy, L&T, Magna and Johnson & Johnson, appear only on secondary Chinese-language sites and are not treated here as customers. The pace of additions is itself informative: Dexory, SunCubes, Dolphin Global and Varisco were all announced within the three months from July to September 2026, so the case-study set looks actively curated for marketing rather than a retention record, and no customer is described as a multi-year, high-usage account in any source.

On revenue the answer is thin, and the thinness is the finding. The only figures in the public record are database estimates: EUR 4,809,499.90 for the most recent financial year and EBITDA of EUR -7,247,220.10 from Preqin via TipRanks, EUR 4,443,388 for 2023 from Implisense, and a USD 25 million to 50 million revenue band from LeadIQ that is a machine-generated estimate off by an order of magnitude. There is no public audited financial statement, no reliable ARR figure, no paid-customer count, no net revenue retention, no gross margin and no customer concentration disclosure. The circulating figures of EUR 28.4 million and EUR 45.2 million ARR, together with USD 6.2 million of annual AWS spend, 600 universities, 250 consulting partners and 2,100 enterprise seats, all originate from AI-generated business-model and marketing-mix template pages and are excluded from this report.

The one external quality signal is G2, which named SimScale a Leader in the Fall 2025 Enterprise Grid for Simulation and CAE: overall satisfaction 96, 94% of reviewers giving four or five stars, 90% willing to recommend, quality of support 95%, ease of setup 93%, ease of use 90%, fluid analysis 89%, FEA 83%, and multibody dynamics 79%, the last being the lowest category score. SimScale also appears in G2's 2025 list of best German software companies. Two caveats apply: the Leader designation is quoted on the company's own release page, and G2 subscores are vendor-selected reviews rather than an audited survey.

Competitive Landscape

SimScale is best understood as a cloud delivery and orchestration business, and that is precisely where it sits between the two comparison sets that matter.

Against cloud HPC platforms, the difference is what runs versus where it runs. Rescale, founded in San Francisco in 2015, is a cloud high-performance computing platform whose product is the scheduler, the marketplace of more than 1,350 applications and enterprise software delivered as a service, including commercial CAE codes such as Abaqus and Ansys. Rescale's customers bring their own licences; SimScale's customers bring their own geometry. Rescale completed a USD 115 million Series D on 7 April 2025 with NVIDIA among the participants and cumulative funding above USD 260 million, has not been acquired, and a rumoured USD 150 million Series D-2 at a USD 1.1 billion post-money that circulates from a single funding-roundup site is not corroborated by the company's own newsroom. The two companies compete for the same elastic compute budget and the same engineering user, and the functional difference is the difference between renting infrastructure and renting an engineering workflow.

Against GPU orchestration for open-source solvers, the market has just emptied out. Inductiva, the Portuguese platform that combined GPU orchestration with hosted open-source solvers such as OpenFOAM and CalculiX, declared on its contact page that it would permanently cease operations from 31 January 2026, and as of the research date its main site, documentation site and console all return 404. The shutdown notice itself was obtainable only through a database citing the official page, with no archived snapshot available, so the announcement rests on a second-hand reference to a first-party statement. The strategic reading is that the pure orchestration-for-open-source-CFD niche cleared out, SimScale inherited part of that demand, and the operational fragility of a single-purpose cloud CAE provider became visible in the same quarter.

Against the incumbents, the trade is licence cost against trust. Ansys Cloud and Discovery give customers the full Ansys stack with fast geometry cleanup, deep industrial penetration and long-established trust in result accuracy, at the cost of heavier licensing and a higher entry threshold. OnScale, an early cloud CAE provider, was acquired by Ansys in 2022 and is regularly confused with SimScale, which has no Ansys partnership of any kind in the sourced record; the official partners are Hexagon, Simerics, Siemens PLM Software, Tech Soft 3D, nTop, Onshape and NVIDIA. COMSOL Multiphysics remains the mature local-first multiphysics option with cloud options attached. Hexagon Marc and Simerics are suppliers to SimScale rather than rivals to it, which is simultaneously a partnership advantage and a supplier-concentration risk.

Against the model-first cohort, the routes are genuinely different. PhysicsX in London raised a USD 300 million Series C in June 2026 after a USD 135 million Series B in 2025, with Siemens and Applied Materials among the investors, and positions itself as an AI-native physics engine. Neural Concept in Lausanne closed a USD 100 million Series C led by Goldman Sachs Alternatives Growth Equity and reports more than 50 enterprise customers. Luminary Cloud in San Mateo raised USD 72 million in a Series B in September 2025 for a GPU-native CFD solver with pretrained models. Emmi AI was acquired by Mistral AI in May 2026, Monolith AI was acquired by CoreWeave in October 2025, and NAVASTO was acquired by Autodesk in October 2024. BeyondMath in London takes the first-principles neural-operator route with an USD 18.5 million seed round. Mistral opened a physics AI centre in Munich on 28 September 2026, which is direct competitive pressure on SimScale's home ground. All of these attack the same design-cycle bottleneck, but with learned models; SimScale attacks it with an operating model, a solver menu and a browser, and adds learned components on top. The fair statement is that SimScale is the cloud-delivery counterweight to the AI-native cohort, not a participant in the model race on equal terms.

On the acquisition question specifically: no source indicates that SimScale has been acquired, and the Crunchbase record shows no acquisition event. The company continues to operate independently, publish official news through 2026 and retain its own domain and brand. Industry media has circulated rumours involving Autodesk, PTC and the idea of synthetic engineers, but no confirmed transaction involving SimScale exists in the record, and the only well-documented cloud CAE acquisition in this period is Ansys buying OnScale. A reader encountering the word "acquired" attached to SimScale is most likely looking at the OnScale confusion.

The partner ecosystem is unusually concrete for a company of this stage: Siemens PLM Software for Parasolid, Tech Soft 3D for HOOPS, nTop, Onshape, NVIDIA for the physics-ML stack, Hexagon, Simerics and AI Engineering for solvers, plus a consulting network including SimuTech and Dataintelo. That breadth is a genuine asset and is also the map of the company's dependencies.

Funding

The equity history is short enough to state round by round, which is the only way to keep it legible.

• 2013: seed, amount undisclosed.
• 2014: angel, amount undisclosed, Earlybird recorded as lead in database records.
• December 2015: Series A, amount undisclosed, led by USV with HTGF joining, coinciding with the launch of the free Community Plan.
• January 2017: Series B, amount undisclosed, led by June Fund, referred back to in the 2020 company release.
• 30 January 2020: Series C of EUR 27 million, led by Insight Partners, with HTGF, Earlybird, USV and June Fund participating. Joshua Fredberg of Insight Partners joined the board.
• 28 October 2021: Series C extension of EUR 25 million, co-led by Draper Esprit and Insight Partners, with HTGF, USV and Earlybird participating. The company's own release states the C round totals EUR 52 million.

Separately, on 20 December 2024, the European Investment Bank signed EUR 25 million under its InvestEU framework, project reference 20240117, project code SIMSCALE, against a total project cost of EUR 66 million. This is a loan, not equity. It does not dilute, it carries fixed repayment obligations, and it comes with its own covenants. Adding it to the EUR 52 million to produce a headline of EUR 77 million of "total funding" is a category error that this report does not make.

The three earlier rounds have never disclosed amounts, so EUR 52 million is the confirmable equity total and not necessarily the complete equity total. Database round counts differ because they count different things: LeadIQ reports 6 rounds and USD 59 million of total funding, TipRanks reports USD 60.2 million, Preqin reports 5 deals, and PitchBook records 10 rounds spanning 2012 to October 2025, the latter including a University Spin-Out entry and a private secondary transaction on 1 October 2025 that are not equity raises in the ordinary sense. None of these counts can be compared to each other without restating what is being counted. The USD 59 million and USD 60.2 million figures are not adopted, and the frequently quoted "around USD 60 million across 9 rounds" comes from an AI-generated template site and is excluded, as is the USD 26.3 million debt figure, which no EIB or company source confirms.

On valuation, Preqin reports EUR 241.168 million with no published methodology. Dealroom is quoted at roughly USD 226 million, but the original page was not retrieved and the figure is a second-hand mention, so it is downgraded. The commonly cited EUR 275 million has no source at all and is not adopted. The only listed-company datapoint is Molten Ventures plc's final results for the financial year ended 31 March 2025, which list SimScale at GBP 11 million invested and GBP 11 million fair value, with a portfolio table amount of GBP 11.3 million, valued on the calibrated price of a recent investment. That is one fund's carrying value in its own portfolio and must not be read as a company valuation.

There is no IPO filing, no selected venue, no timetable, no underwriter indication, no employee option pool disclosure and no shareholder register in the public record. The chief executive stated in 2025 that the company remains a private company and does not share specific financial metrics. The 1 October 2025 secondary transaction has no public disclosure of parties or price. The combination of a EUR 25 million EIB loan with an identified public-sector creditor and a private cap table with no visibility is a structure an investor will want explained before pricing anything.

DISCREPANCIES

Eight source conflicts were identified and adjudicated item by item.

1. Founding date. The company timeline says November 2012; Implisense records 17 October 2012; PitchBook shows 1 January 2012 as the placeholder date of a "University Spin-Out" entry. Adjudicated: October or November 2012. The company statement is first-party, the Implisense date is a register-derived record, and the PitchBook date is a round-date placeholder that reflects Technical University of Munich technology transfer rather than an incorporation event.

2. Headcount. 78 in 2023 from Implisense; 141 from Dealroom; 143 from PitchBook and LeadIQ; 144 as of 21 September 2026 from TipRanks; a 101 to 250 band from Crunchbase; roughly 130 to 160 from grokipedia and roughly 150 from a Chinese repost site. Adjudicated: roughly 140 to 150 people on a 2025 to 2026 snapshot, with 78 noted for 2023. The TipRanks, PitchBook and Dealroom figures are near-dated independent snapshots that agree closely; the Crunchbase band is too coarse to use; the aggregator figures are AI-generated or reposted. The 78 and 144 figures are not averaged, because averaging them produces a number that was never true on any date.

3. User count. 600,000 in October 2024, 800,000 in March and May 2026, 700,000+ on the 2026 careers page, and 900,000+ on the 2026 pricing page, all from official channels. Adjudicated: 900,000+ as the latest official cumulative figure, with the 800,000 and 700,000+ readings quoted alongside their timestamps. The disagreement between official channels at different times indicates a marketing cumulative counter, which very likely includes free Community registrations, and it is therefore labelled as an official self-reported cumulative user count and not used to infer paid conversion.

4. Cumulative funding. EUR 52 million of equity from the 2020 and 2021 C round official releases; USD 59 million across 6 rounds from LeadIQ; USD 60.2 million from TipRanks; around USD 60 million across 9 rounds from an AI template site; and a USD 26.3 million debt claim that no source confirms. Adjudicated: EUR 52 million of confirmable equity, from 2020 EUR 27 million plus 2021 EUR 25 million, with the earlier 2013 to 2017 rounds undisclosed; the EIB EUR 25 million loan of 20 December 2024 recorded as a separate debt line. The database figures mix differing round counts and unclear currency treatment, and the template-site figure has no usable provenance. Equity and debt are not added together.

5. Revenue. EUR 4,809,499.90 for the most recent financial year from Preqin, EUR 4,443,388 for 2023 from Implisense, and a USD 25 million to 50 million band from LeadIQ's automatic estimate. Adjudicated: the Preqin figure as the database reading for the most recent financial year, with the 2023 Implisense figure noted, and EBITDA of EUR -7,247,220.10 stated alongside. The LeadIQ band is a machine estimate off by an order of magnitude and is not adopted. The gap between the Preqin and Implisense readings most likely reflects financial-year cut-off and the treatment of other income lines, and it should be put to the company directly. There is no public audited financial statement, and the report says so.

6. Founders. The official founders page names five people, all with a TUM background: David Heiny, Vincenz Doelle, Johannes Probst, Alexander Fischer and Anatol Dammer. Preqin includes Rajarshi Mandal as a co-founder. Adjudicated: the official five. The founders page is a first-party, individually named document; the Preqin entry is a database transcription with no supporting source, and the three names in the prior draft of this brief are not supported by any source.

7. Community allowance and round counting. The current pricing page gives Community as 2,000 core hours, while a third-party scrape shows 3,000 and the original 2015 Community Plan was 3,000 computation hours plus 500 GB of storage. On round counting, LeadIQ says 6 rounds, PitchBook 10, Preqin 5 deals. Adjudicated: 2,000 core hours for Community, with the historical values identified as older configurations that must not be mixed in; and round counting expressed as at least five verifiable equity rounds (2013 seed, 2014 angel, 2015 Series A, 2017 Series B, 2020 Series C and 2021 C extension) plus two non-equity events, a university spin-out and a secondary transaction. The database counts differ in what they include and cannot be compared without restating the basis.

8. Valuation. Preqin gives EUR 241.168 million, Dealroom is quoted at roughly USD 226 million without the original page having been retrieved, and the prior EUR 275 million figure has no source. Adjudicated: EUR 241.168 million as a database valuation with unpublished methodology, the Dealroom figure downgraded to an unverified second-hand mention, and EUR 275 million not adopted. No first-party or investor disclosure backs any of the three, so the report presents roughly EUR 240 million as a database reading and explicitly not as a fact. The Molten Ventures GBP 11 million carrying value is a fund's portfolio mark, not a company valuation.

One related figure is recorded here because it circulates widely and is unusable: the ARR estimates of EUR 28.4 million and EUR 45.2 million, together with USD 6.2 million of annual AWS spend. All three originate from AI-generated business-model and marketing-mix template pages, and none is used as a fact in this report.

Risks and Diligence Gaps

Revenue and profitability are effectively invisible. The only figures are database readings of EUR 4.81 million of revenue for the most recent financial year and EBITDA of EUR -7.25 million, alongside EUR 4.44 million for 2023, with no audited financial statement, no cash balance and no burn rate. A company with EUR 25 million of EIB debt, a reported EUR 4.8 million revenue and a seven-million-euro EBITDA loss has a specific and finite runway, and none of it is disclosed. The diligence ask is the audited statements, the cash balance, the burn rate and the debt amortisation schedule.

Paid customers, ARR and concentration are all unknown. The official cumulative user counter reads 900,000+, which very plausibly includes free Community registrations, and no source discloses how many of those convert to paying accounts, how many enterprise accounts exist, or what share of revenue comes from the top five customers. Until this is answered, the difference between a large business with thin monetisation and a small business with a few very large accounts cannot be assessed.

Every performance claim lacks published test conditions. The GPU LBM 20 to 30 times, the SPH 10 to 20 times, the roughly 2700 times pump model and the within 5% Physics AI figure disclose no model, mesh, core count, error tolerance or baseline solver. The roughly 2700 times figure additionally does not define what the traditional CFD comparison is or whether accuracy is equivalent. The diligence ask is the full test matrix for each claim and an independent reproduction on at least one case.

The precision evidence base is a single case. The NAFEMS FV4 cantilever beam with Code_Aster and Frequency Analysis on three grids of 32,663, 118,696 and 240,757 nodes, deviating between -0.7% and +2.6% across six modes, is the only publicly checkable accuracy result, and it is a self-run test on self-selected meshes in a well-understood structural regime. There is no third-party verification, no industrial validation case and no blind study. For a platform whose commercial pitch includes replacing licensed incumbents in regulated industries, this is a thin base.

Solver licensing and third-party dependency are the sharpest structural risks. Code_Aster is GNU GPL and the platform is closed-source commercial software, so the boundary between using the solver as a service and distributing it needs legal verification. Three of the strongest engines come from third parties, Hexagon, Simerics and AI Engineering GmbH, whose revenue-share, renewal and exclusivity terms are undisclosed, which creates both supplier concentration and margin exposure. The NVIDIA dependency on the physics-ML stack adds a fourth node. SimScale is simultaneously a reseller of other people's solvers, which is the standard distribution risk of the cloud CAE model.

The AI training-data question is unanswered. The company says its Engineering AI draws on more than one million real simulation projects. Whether customer geometries, meshes and results were used for training, under what consent, de-identification and licensing terms, and whether a customer can object to it, is not public. This is the commercial question that will decide how large accounts in aerospace, automotive and energy sign.

Cloud provider dependency is undisclosed. Only the 2020 AWS re:Invent material is available, and no source states whether the platform is single-cloud, multi-cloud, or carries a committed minimum spend. An unquantified cloud commitment is a material cost item for a business with EUR 4.8 million of reported revenue, and the USD 6.2 million AWS-spend figure that would answer it comes from an unusable source.

Governance and leadership visibility are incomplete. The team page names no chief technology officer, no source confirms who holds the role, and the technical argument depends on in-house solver development. No complete timeline exists for CTO, CCO or VP-level changes, and Preqin's claim of a sixth co-founder has no first-party support. Combined with an unpublished shareholder register, a lapsed LEI record, and no retrievable commercial register annual report for 2024 or 2025, the diligence picture on governance is unusually thin for a company at this funding level.

Valuation has no verifiable basis. EUR 241.168 million from Preqin with no published methodology, a USD 226 million second-hand Dealroom mention, an unsourced EUR 275 million in circulation, and a GBP 11 million fund-level carrying value from Molten Ventures. None of these is anchored to a disclosed round price, an option repurchase price or an audited instrument.

There is no public market signal. No IPO filing, no venue, no timetable, no underwriter, no option pool disclosure. Liquidity cannot be inferred from any public indicator, and the only exit routes visible are a strategic sale or a continued private round.

The competition is moving from two directions at once. Ansys Cloud and Discovery continue to advance and Ansys already owns OnScale, so the incumbent is both improving the cloud product and having already absorbed one independent provider. On the model side, PhysicsX, Neural Concept, Luminary Cloud, Mistral's Munich physics AI centre and the AI-native cohort are attacking the same design-cycle bottleneck with learned models rather than with an operating model. SimScale's moat, which is solver IP plus orchestration plus accumulated simulation data, is verifiable in the solver menu but unproven as a moat, and the company is itself hedging by opening its execution infrastructure to outside solvers and AI methods through Workflows.

The customer list is likely curated rather than representative. Roughly seventeen names are officially evidenced, and four of them were announced within a single quarter. No source describes a long-duration, high-usage enterprise account, and no retention, renewal or expansion data exists. The absence of customer complaints is not evidence either way, because the customer base is small relative to the user counter and no customer has published an independent view of the platform's accuracy or support quality.

Third-party reputation data carries vendor-selection bias. The G2 Leader designation is quoted on the company's own release, and G2 subscores are vendor-selected reviews; the weakest category, multibody dynamics at 79%, also shows where the platform is not strongest.

Structural and reporting opacity compounds the rest. A lapsed LEI, no retrievable register filings for 2024 and 2025, a shareholder register that has never been disclosed, and the presence of a private secondary transaction in October 2025 with no announced parties or price all mean that a diligence team cannot independently reconcile who owns what. This is the cost of a company that has raised from four institutional investors plus a public development bank and has published nothing about its cap table.

Overall Assessment

• Technology 7.5. Seven distinct solver engines including an in-house GPU lattice Boltzmann implementation orchestrated behind one submission and one post-processing surface, Code_Aster with Frequency Analysis, third-party rotating machinery, nonlinear structural and SPH engines, a mesh-refined NAFEMS FV4 benchmark published with node counts and per-mode deviations, in-house GPU and cloud-native execution, and a well-specified agentic workflow for RFQ to auditable report. Deducted for publishing no test matrix for any of the four headline performance claims, having exactly one publicly checkable accuracy result, leaving the CTO seat unnamed, and creating a four-way dependency on GPL and commercial third-party solvers plus one AI infrastructure vendor.
• Commercialisation 7.0. Fourteen years of continuity, a free tier that actually works, a four-tier subscription and core-hour model, roughly seventeen officially evidenced enterprise customers spanning automotive, energy, industrial equipment and medical devices, and enterprise AI products that entered general availability in 2026 with a waiting list opened in May. Deducted for having no published price beyond the free tier, no ARR, no paid-customer count, no customer concentration data, no retention data, and a customer list whose recent additions look curated for marketing.
• Funding 7.5. Four credible institutional equity investors across seed through Series C, EUR 52 million of confirmable equity, EUR 25 million of EIB InvestEU debt signed in December 2024 against a EUR 66 million project cost, and a lead investor in each of the two disclosed rounds with a named board seat. Deducted for three undisclosed early rounds, a fully opaque shareholder register and option pool, no IPO path or timetable, a valuation with no verifiable methodology, and a debt instrument whose covenants and amortisation are undisclosed.
• Product maturity 7.5. A coherent ladder from browser-based Cloud Simulation through geometry integrations with Parasolid, HOOPS Exchange, nTop and Onshape, through Physics AI, Engineering AI and the pump foundation model, to SimScale Workflows opening the execution layer to outside solvers and AI methods, supported by a working free tier and a G2 Leader position. Deducted for the flagship AI components being layered on a platform that predates them, the within 5% figure having no published basis, and the current Community allowance having already changed at least once in public view.
• Market position 7.0. A differentiated position as the cloud-delivery counterweight to the model-first cohort, a genuine multi-solver menu that no thin wrapper offers, enterprise design partnerships across Siemens PLM Software, Tech Soft 3D, nTop, Onshape, NVIDIA, Hexagon and Simerics, and a German home-market base in a category that has already seen Ansys absorb OnScale. Deducted for the absence of any ARR or paid-customer metric to size the business, the risk of being squeezed between improving incumbent clouds and model-first competitors, the Inductiva shutdown demonstrating the fragility of pure cloud CAE economics, and a Munich home ground that Mistral entered directly in September 2026.

References

1. SimScale company timeline and About pages, simscale.com/about-us-2, for the November 2012 founding, the 2013 platform launch, the Community Plan, and the 2020 to 2026 product milestones.
2. SimScale Imprint, simscale.com/imprint, for SimScale GmbH, Ridlerstrasse 31B 80339 Munich, Amtsgericht Muenchen HRB 201813, VAT DE286403296, registered capital EUR 169,339, and the OpenFOAM, Siemens PLM and nTop trademark notices.
3. SimScale founders feature and team pages, simscale.com/about/simscale-team, for the five named founders, the executive team, the absence of a listed CTO, and the user counter of 700,000+ on the careers page.
4. SimScale Series C announcement, 30 January 2020, simscale.com/press/simscale-closes-series-c, for EUR 27 million led by Insight Partners, the Series A and B history, and the board seat held by Insight Partners' Joshua Fredberg.
5. SimScale Series C extension announcement, 28 October 2021, simscale.com, for EUR 25 million co-led by Draper Esprit and Insight Partners and the stated EUR 52 million C round total.
6. European Investment Bank project page 20240117, eib.org, for the EUR 25 million InvestEU loan signed 20 December 2024, project code SIMSCALE, total project cost EUR 66 million.
7. SimScale CFD product documentation, simscale.com/product/cfd, for the OpenFOAM interFoam basis and the GPU lattice Boltzmann 20 to 30 times claim.
8. SimScale validation case, NAFEMS FV4 cantilever beam with off-centre masses, simscale.com, for the three grid sizes and the -0.7% to +2.6% per-mode deviation with Code_Aster and Frequency Analysis.
9. SimScale Foundation AI Model announcement, 18 March 2025, simscale.com, for the NVIDIA PhysicsNeMo, Blueprint, Warp, Omniverse and Blackwell stack and the roughly 2700 times claim.
10. SimScale pricing page and Enterprise AI announcement, 7 May 2026, simscale.com, for the Community 2,000 core hour allowance, the Mechanical, Professional and Enterprise tiers, the absence of published prices, and the 900,000+ cumulative user counter.
11. SimScale and Hexagon announcements of 2024-12, 2025-07-09 and 2026-02-03, and the Simerics partnership and community-opening releases of 2021-09 and 2026-06-19, for the third-party solver relationships.
12. SimScale G2 Fall 2025 Enterprise Grid release and the G2 2025 best German software companies list, for the Leader designation and the category subscores including multibody dynamics at 79%.
13. TipRanks SimScale company record, 21 September 2026, carrying Preqin data, for the EUR 4,809,499.90 revenue, EUR -7,247,220.10 EBITDA, EUR 241.168 million valuation and 144 headcount readings.
14. Molten Ventures plc final results for the financial year ended 31 March 2025, for the SimScale holding of GBP 11 million invested and GBP 11 million fair value valued on the calibrated price of a recent investment.

This report is compiled from public sources. It is for reference only and does not constitute investment advice. Data is current as of October 22, 2026; please refer to official company announcements and filings.