Rescale Deep Dive: An HPC Operating System and Desktop Engineering Workbench in the Cloud, a $115M Series D, and Five Misread Facts

Executive Summary

Rescale, Inc. is a privately held cloud high-performance computing platform founded in 2011 and headquartered in San Francisco. The cleanest way to hold the company is as an HPC operating system for the cloud plus a desktop engineering workbench on top of it. It sells compute orchestration and the carrying of engineering workflows across AWS, Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure, CoreWeave and other capacity partners, on top of a control plane that pre-tunes more than sixty-five hardware profiles, brokers commercial software licences, hosts remote visualisation desktops and exposes a Slurm-compatible job interface. Solvers themselves and in-house physics foundation models both sit outside its product boundary.

The funding record is real and recent. The last officially announced round is a $115M Series D closed on 7 April 2025, with cumulative capital raised stated by the company at more than $260M. The company remains private and unacquired, and the site schema's phase enumeration carries a dedicated Series D value, so this report maps the company to series-d directly.

Five claims circulate widely and each of them misleads.

First, the founders are Joris Poort and Adam McKenzie. Both are former Boeing 787 structural and software engineers. One alternative pair of names attached to this company has no supporting source anywhere in the record, and the single matching individual of that name is a developmental biologist unrelated to computing.

Second, no Stanford background exists for either founder. Poort holds a bachelor's from Michigan, a master's from Washington and an MBA from Harvard Business School; McKenzie holds a bachelor's from Oregon State.

Third, the founding date is 18 November 2011, taken from the company's own government registration. The year 2012 is the year of its Y Combinator W12 batch, which is an accelerator cohort year and not an incorporation year. A separate claim that the company pivoted to software-as-a-service in 2019 also fails: it launched in 2015 already in a platform-as-a-service form, and 2019 brought a government product and a Siemens partnership instead.

Fourth, no acquisition of Rescale has been announced, agreed, closed or completed. NVIDIA is an investor, an ecosystem partner and an upstream supplier, and its public appearances are supportive quotations inside Rescale press releases. The company's own news index carries no merger entry, NVIDIA's published acquisition history for 2023 to 2025 does not include Rescale, and the legal entity, the legal entity identifier and the management team were all unchanged as of October 2026. A single funding aggregator reported a $150M round at a $1.1B valuation led by NVIDIA on 23 June 2026; that report is unverified, is inconsistent with the official cumulative figure, and is excluded from the main tables here.

Fifth, the only official pricing that can be verified is a 2016 price list. On-demand compute ran from $0.10 to $0.30 per core-hour and prepaid from $0.05 to $0.09 per core-hour in that year. The current pricing page is a guided configurator with no numbers, and current list prices, annual recurring revenue, gross margin and top-ten customer concentration were not found in any public source.

The commercial evidence that does exist is solid. Named customers include Arm, General Motors Motorsports, Samsung, SLB, the United States Department of Defense, McLaren Automotive, Honda under a multi-year agreement, Daikin Industries and Bristol-Myers Squibb. Ansys states on its own partner site that Rescale serves seven of the ten largest global automakers, seven of the ten largest aerospace firms and three of the six largest life-science groups. The company states that its customer base spends more than $1B a year on HPC infrastructure.

Our view: this is a well-capitalised, technically credible infrastructure company whose moat sits in engineering-adjacent machinery, namely commercial licence brokering, remote desktop visualisation, mixed-vendor workflow continuity, scheduling and data movement. Its defensibility depends on standing between hyperscalers and engineering budgets, and its main exposure is that the hyperscalers are increasingly selling the same abstraction themselves. The platform is real, the customer list is real, the funding is real, and the performance marketing is not yet independently reproduced.

Company Overview

The legal entity is Rescale, Inc., a Delaware corporation. Its identifiers are unusually well documented and worth recording because so much of the surrounding online material is wrong: Employer Identification Number 45-3837167, SEC Central Index Key 0001674741, Legal Entity Identifier 254900UNWNJ44V69AT48 issued on 17 May 2024 through Bloomberg Finance L.P., SAM.gov Unique Entity ID HPSAC9M3RS44, and CAGE code 6P4V2. The statutory registered address sits with the Delaware registered agent at 3500 South Dupont Highway, Dover, Delaware 19901. The operating headquarters is 981 Mission Street, Unit 21, San Francisco, California 94103, moved there on 17 April 2025 from 33 New Montgomery Street. The fiscal year ends on 31 December.

The company dates itself to 18 November 2011, recorded as the business start date in SAM.gov, and Y Combinator lists it as founded in 2011. Its YC cohort is W12, which ran from January 2012. Joris Poort founded and leads the company as chief executive officer and also appears as an executive officer and director on the Form D filings; Adam McKenzie is co-founder and chief technology officer, likewise listed as executive officer and director. Poort spent four years at Boeing on the structural and software side of the 787 tail and wing optimisation programme, then worked in technology product development consulting at McKinsey in Amsterdam, and holds a bachelor's with high distinction in mechanical engineering and a minor in applied mathematics from the University of Michigan, a master's with high distinction in aeronautics from the University of Washington, and an MBA with distinction from Harvard Business School, class of 2010. McKenzie spent seven years at Boeing managing 787 wing structural and software engineering projects and holds a bachelor's with distinction in mechanical engineering from Oregon State University.

Headcount has eight published readings and each carries a different date: 246 at Crustdata in September 2026, 251 on the company's own LinkedIn page in 2026, roughly 250 in the YC directory in 2026, roughly 244 at LeadIQ in August 2026, 236 at PrivateMarketView in October 2026, roughly 216 quoted by ENGtechnica in January 2025, 244 at Dealroom in 2025 and 180 with no stated date in the English Wikipedia infobox. The usable statement is roughly 245 to 250 people in 2026, with the historical path running about 252 in December 2022, down to 217 in November 2024, back to 246 by September 2026. A Dealroom breakdown places the team in fifteen countries with 77.4% in the United States.

The corporate timeline contains several items that a reader should carry into the rest of this report. ScaleX Enterprise shipped on 18 May 2015 as the first coherent product line. A $6.4M round was announced on 14 July 2015. Pay-per-use Ansys licensing through Ansys Elastic Licensing arrived on 18 January 2018, and the $32M Series B closed on 24 July 2018. ScaleX Government and a FedRAMP Marketplace presence followed in November 2019, alongside a strategic partnership with Siemens PLM Software for Simcenter delivered as software-as-a-service. High-Throughput Computing capability shipped on 22 August 2021 with Arm as the first customer. Rescale Data Intelligence with the Rescale Assistant landed on 23 October 2025. Agentic Digital Engineering with AI Physics OS and Compute Economics was announced on 12 May 2026. An initiative with Lawrence Berkeley, Oak Ridge and Lawrence Livermore national laboratories was announced on 9 June 2026, and on 23 July 2026 the company disclosed Phase I funding under the Department of Energy Genesis Mission from a $293M pool, as the only commercial member of that programme. A CoreWeave integration followed on 25 August 2026, and on 14 October 2026 Rescale and Honda expanded into a multi-year agreement.

Senior leadership beyond the founders is intact and stable: Matt McKee as chief operating officer since January 2022, David Si as chief financial officer, John Moonshower as chief revenue officer, plus heads of engineering, product, finance, legal, global people, revenue operations, APAC and HPC. Ben Verwaayen, former chief executive of BT and of Alcatel-Lucent, joined the board in 2018.

Common Misreading Warning: Five False Premises About This Company

The following five claims circulate in secondary material and in informal briefings. Each has been checked against primary sources, and each is wrong.

Premise one, the founders. Some accounts name Jeff Generic and Michael Shook. More than twenty sources, including the company leadership page, the Y Combinator directory and SEC Form D filings, consistently name Joris Poort and Adam McKenzie. The only individual matching the second name in the wider record is a developmental biologist whose work concerns salamander morphogenesis. The premise is treated as erroneous information.

Premise two, a Stanford pedigree. No source links either founder to Stanford. The documented degrees are Michigan, Washington, Harvard Business School and Oregon State.

Premise three, the founding year. Government registration gives 18 November 2011 as the business start date, and every company release says founded in 2011. The year 2012 is the accelerator batch year. A related premise that the company turned to software-as-a-service in 2019 also fails: contemporaneous coverage from May 2015 describes a platform-as-a-service for high-end computer simulations, and the 2018 Series B release is titled around development of a platform as a service. What actually happened in 2019 was the federal government product and the Siemens Simcenter agreement.

Premise four, an NVIDIA acquisition. No announcement, no agreement, no closing and no integration have been found, and the reverse evidence is intact: the company's news index from October 2023 to October 2026 contains no merger or listing item, NVIDIA's published acquisition record for 2023 to 2025 does not list Rescale, and the legal entity name, the legal entity identifier, the domain and the management roster were all unchanged in October 2026. The 23 June 2026 funding report describes a new round, not an acquisition, and remains unverified.

Premise five, the company name. The legal name has always been Rescale, Inc. ScaleX is a platform product brand first seen in 2015 with ScaleX Enterprise and later with ScaleX Government. One database records the company name as ScaleX, which is an error in that database rather than a fact about the company.

Technology

The architectural characterisation that matters most is that Rescale operates a control plane, and that its products are layered on top of it. Layer one is elastic cloud HPC with pre-tuned hardware images called coretypes, of which the official architecture page lists more than sixty-five, with category counts including 75 general purpose, 65 high-speed interconnect, 57 large memory, 48 GPU, 32 large disk, 26 GPU rendering and 25 high clock-rate. Layer two is advanced modelling and simulation. Layer three is data intelligence and engineering intelligence, a unified data fabric and digital thread. Layer four is AI Physics for surrogate modelling. Layer five is agentic digital engineering. Layer six is compute economics and financial management. Rescale is neither a hyperscale infrastructure owner nor a solver vendor; its compute comes from technology partners, and the company's own wording for 2025 is more than 500 cloud datacenters with technology partners such as AMD, Amazon Web Services, Arm, Intel, Microsoft Azure, NVIDIA and Oracle Cloud Infrastructure.

Interface coverage is where the desktop workbench part of the positioning lives. There is a browser graphical interface, a virtual desktop for remote visualisation so that post-processing does not require downloading large result files, a command line, an API, a Slurm-compatible scheduler interface, Ansys RSM and ARC integration, and designated-of-experiment job types. Containers and Spack allow customers to bring their own applications, and the 2026 CoreWeave integration runs on CoreWeave Kubernetes Service. Licence handling runs on two tracks: pay-per-use licences billed in step with compute, and bring-your-own-licence where the customer's own licence server must be network-reachable and Rescale brokers the licence handshake. Ansys Elastic Licensing has been purchasable on the platform since January 2018, and Siemens Simcenter is offered as a monthly graphical subscription plus pay-per-use solver licences.

The solver catalogue has been verified item by item against the official software list. ANSYS is confirmed with Fluent, CFX, HFSS, Mechanical, Icepak, Maxwell, Polyflow, Savant, SpaceClaim, EnSight, Discovery, Rocky, Autodyn, Q3D Extractor, SIwave, EMC Plus, STK, WindModeller, Motor-CAD, OPTIS HPC, optiSLang, PerceiveEM, Forte, FreeFlow, Zemax OpticStudio, Lumerical, Workbench, RSM, Chemkin-Pro, and ANSYS LS-DYNA as a separate entry. OpenFOAM is confirmed, appearing in the ITAR documentation list of supported software and in the video tutorial index. Dassault SIMULIA and Abaqus are confirmed with several Abaqus CAE variants. COMSOL is confirmed with Multiphysics, Server and graphical variants. Siemens is confirmed with Simcenter STAR-CCM+, Amesim, 3D, Nastran, NX Nastran, STAR-CD and MSC Digimat. MSC Software and Patran are confirmed. Altair is confirmed with Time Zone, Compose and FEMFAT. CONVERGE is confirmed for versions 2 through 4.0 and Studio. AVL is confirmed with BOOST, EXCITE, FIRE and the Simulation Suite. Autodesk Alias Workstation and SOLIDWORKS are confirmed. Blender is confirmed as a rendering and visualisation entry. The EDA family is confirmed with Cadence Fidelity, Innovus, Virtuoso, Spectre, Xcelium, Pegasus, Reality DC, Sigrity and Indigo, while Synopsys appears only partially. NASA FUN3D and CART3D are confirmed as open-source computational fluid dynamics, and the Boom Supersonic case confirms use. Open-source and scientific entries include OpenFOAM, CalculiX, Elmer, FEniFS, FEMZIP-CFD, DualSPHysics, FLAC2D and 3D, GROMACS, AMBER, CP2K, DFTB+, LAMMPS, GAMESS, ParaView, VMD and Quantum ESPRESSO in part. Machine-learning frameworks are confirmed with TensorFlow, Keras, PyTorch, Caffe, Anaconda, cuDNN, NVIDIA CUDA and PhysicsNeMo.

Three mutually exclusive counts circulate for that catalogue and they must be separated. The official catalogue total is 1,000+ in 2021 to 2024, 1,250+ in April 2025 and 1,350+ on the 2026 home page, and this counts application packages and versions rather than independent solvers. A third-party cloud CAE comparison states 50-odd solvers, and another page of the same source states 200-plus solvers, and a 2021 Chinese-language article states more than 800 applications. The defensible reading is that the application-package count is 1,250 to 1,350-plus and that independent commercial solvers run in the 200-plus range, and the third-party figures clearly undercount.

The AI and machine-learning capability is genuine and already productised, but its shape matters. AI Physics OS is a workflow that takes fragmented simulation data, turns it into AI-ready datasets, trains a validated surrogate model and serves it for inference, with no-code connection to NVIDIA PhysicsNeMo, multistream transformers, GeoTransolver for transient finite element analysis, training on the public DrivAerML dataset, and a parallel hyper-parameter search called Auto Tune. Agentic Digital Engineering, released on 12 May 2026, provides simulation-native prebuilt agents that validate input decks, diagnose failed jobs, troubleshoot solver-specific problems, size hardware against benchmarks and generate reports, plus agent nodes embeddable inside workflows and graded autonomy levels running from observe-and-recommend to act within defined guardrails. Rescale Assistant provides natural-language cross-discipline analysis, and Rescale Data Intelligence provides the unified data fabric, digital thread and metadata management. The company has also published an MCP connector for Claude Desktop, the RAFT agent evaluation framework, the open-source Rescale Interlink transfer tool with FIPS 140-3 certification, and a simulation monitor with divergence detection and log-level root-cause analysis. The Compute Recommendation Engine recommends hardware across architectures from workload metadata. The correct characterisation is that these are product features in an agentic engineering and surrogate modelling layer, not an in-house physics foundation model company.

The performance claims must be read as vendor marketing. The 1000x acceleration for design validation, the greater than 99% accuracy and 85% cost reduction quoted in an NVIDIA technical blog authored by the chief executive in April 2026, the up to 10x performance and 30x cost reduction in the European industrial AI cloud announcement of June 2025, the 3600x faster carbon-emission detection and 10x aerodynamics speedups, the near 500,000 instantaneous vCPUs across two cloud providers and twelve regions with more than 50 million core hours over eight weeks and 99.99% job reliability from the 2023 High-Throughput Computing release, the 78% reduction in solve time, 4x workload throughput and 60% single-job cost reduction from the architecture page, the 78% shorter research and development cycle and 4 to 8x design candidates from the home page, and the 99.9% success rate with millions of jobs daily from the demonstration page are all first-party statements. Test conditions, baselines and error bounds are not published, and the largest single job scale was not found in any public source.

Product and Commercial Model

The billing unit is the core-hour. Charges are for use rather than upfront commitment, and the model runs in parallel with platform subscription tiers: Rescale Enterprise, Rescale Foundation, Rescale Government, Rescale Academic Essentials and the historical ScaleX naming of Pro, Enterprise, Universities and Developers.

The one verifiable official price list is historical. The 2016 pricing announcement published on-demand, low-priority and prepaid per-core-hour rates by configuration: low memory at 0.10 on demand and 0.05 prepaid, standard at 0.15 and 0.07, high-performance computing at 0.20 and 0.07, high memory at 0.25 and 0.08, and high-memory high-performance computing at 0.30 and 0.09. Low priority took roughly half the standard rate for spare short-lived capacity, signalled by a three-colour availability light. A 2022 official conference deck added node-level hourly prices for named coretypes. These are 2016 figures and they are explicitly not current pricing. The 2026 pricing page is a guided configurator that collects workload details and returns a tailored package, with no numbers on the page.

Two forms of free access are documented. Academic Essentials grants the platform licence to universities at no charge, with the institution paying only the cloud infrastructure it consumes. A free trial is available, recorded by one database as 14 days, and a historical NVIDIA on Rescale test drive offered up to a week of free GPU-accelerated hardware from AWS, Azure, Google Cloud or Oracle. Whether free core-hour credits are offered today was not found in any public source.

Named customers with official confirmation are Arm, General Motors Motorsports, Samsung, SLB and the United States Department of Defense in the Series D release; Applied Materials, GM Motorsports, Samsung, SLB and the Department of Defense in the Data Intelligence release; McLaren Automotive, Honda under a multi-year agreement, Daikin Industries and Bristol-Myers Squibb in 2026 cases and announcements; and historically Nissan, Boom Supersonic, Benchmark Space Systems, Zeleros, Exponent, Metabiota, AZothBio, Amogy, NOV, Liberty University, Texas Tech University, Specialized, pSemi, Kairos Power, Sumitomo Precision Products, Vertical Aerospace and Airbus in official data tables and case material. The Ansys partner page states that Rescale serves seven of the ten largest global automakers, seven of the ten largest global aerospace organisations and three of the six largest life-science groups. Claims about Airbus, Lockheed Martin, Boeing, Ford and Pfizer appear in aggregators and are not confirmed by the company; Boeing in particular reads as a founder's former employer rather than a verified customer, and should not be cited as one.

Customer-count statements are layered by date and by vagueness. The 2019 release said 125-plus Global 2000 enterprise customers. The 2025 Series D release says hundreds of enterprise customers. Databases quoting company descriptions give more than 300 customers from start-ups to Fortune 50 companies, and one generated market report claims 250-plus enterprise customers with production-grade finite element analysis workloads. Separately, the company states that its customers together spend more than $1B a year on HPC infrastructure. That last number is a statement about customer budgets, not about Rescale revenue, and the distinction is easy to lose.

On the financial side the answer is short. Annual recurring revenue was never disclosed. Gross margin was never disclosed, and for a business that resells cloud compute while brokering commercial software licences the margin is squeezed on both sides. Current list prices are not public. Top-ten customer concentration is not public, and neither are Department of Defense contract values nor service-level terms.

Competitive Landscape

SimScale, in Munich, delivers cloud-native computer-aided engineering in the browser with a freemium tier, built on OpenFOAM, CalculiX and free-fem, aimed at education and small and medium enterprise, and does not serve very large cell counts. It is a genuine competitor for the same engineering budget with a different delivery model: browser-only front-end and back-end, platform-owned solvers, no licence brokering. Rescale is a solver-agnostic control plane where the customer keeps the existing toolchain. Both companies are in the same procurement conversation, and the distinction is browser-native simplicity against heterogeneous enterprise workflow.

Inductiva, founded in 2021 in Porto with roughly 17 people and about $2.45M raised, offered Python-driven simulation and optimisation for researchers with zero configuration, and PitchBook listed only Rescale and SimScale as directly comparable companies in that category. Inductiva has permanently ceased operations as of 31 January 2026, which leaves its corner of the market vacant and is a reminder of how short the runway can be for a well-positioned but undercapitalised entrant. SolveEth and Cirrus Technologies are also named in the competitive set, and comparable public material on either was not found; the name Cirrus Technologies in particular is confused in search results by unrelated companies with similar names.

The relationship with hyperscale-native batch platforms is one of dependency rather than replacement. AWS Batch, AWS ParallelCluster, Google Cloud and Azure CycleCloud provide raw compute and scheduling infrastructure, leaving the customer to manage software licensing, licence servers, Slurm tuning and licence cost. Rescale's own 2021 material lists AWS, Google Cloud, IBM, Azure and Oracle as partners, and the 2025 wording is partners such as AMD, Amazon Web Services, Arm, Intel, Microsoft Azure, NVIDIA and Oracle Cloud Infrastructure. Rescale therefore sells a cross-cloud abstraction layer with pre-tuned software stacks, licence brokering, cost governance and compliance on top of those clouds. Whether Rescale can keep that layer as the hyperscalers push their own simulation-ready instance types and scheduler integrations is the central structural question of the business.

Solver vendors also sell cloud forms of their own. Ansys Cloud, Dassault 3DEXPERIENCE and Siemens Xcelerator Cloud bundle their own solvers with the simplest possible licensing and carry vendor lock-in. Rescale's counter-position is multi-vendor neutrality. Infrastructure and virtual-desktop providers such as V2 Cloud and Penguin Solutions sit alongside rather than against it. Within the site's own coverage, the differentiation is worth stating plainly: PhysicsX, Neural Concept, Luminary Cloud and Vinci build physics engines and learned models; BeyondMath and Neural Concept build neural-operator surrogates; Rescale builds the substrate that runs other people's solvers and other people's models. Its nearest neighbours in the AI-physics cohort are therefore complements as well as competitors.

One framing point deserves emphasis. Engineering simulation is only one workload on the platform. The catalogue includes GPU rendering coretypes and Blender, the company runs molecular dynamics and computational chemistry workloads, and its AI and machine-learning acceleration path has served training jobs since 2017. The market figures the company itself cites in April 2025 are a $50B HPC market, a $30B product lifecycle management market and a $20B simulation software market, with only 20% of high-performance computing workloads on cloud in the 2021 framing and 95% of the market still on-premise in the 2018 framing. A label such as simulation company therefore understates the business, and equally overstates the portion of the business that touches physics.

Funding

The rounds are unusually well documented and largely reconcilable.

• Y Combinator seed, 2012-03, $125,000, recorded by Dealroom.
• Angel seed, 2012-04, participants ff Venture Capital, DCVC, ACE and Company and several individuals, amount not disclosed.
• Seed, 2014-03, Streamlined Ventures, amount not disclosed.
• Seed and pre-Series A, 2015-07-14, $6.4M, with Sam Altman, Jeff Bezos, Richard Branson, Paul Graham, Ron Conway, Chris Dixon and Peter Thiel among the participants.
• Series A, 2016-05, $14M, with TransLink Capital, M12, ITOCHU, UpHonest, Jump Capital and Two Roads Group. The Form D records $13.5M offered and $11.5M actually sold.
• Series B, 2018-07-24, $32M, taking cumulative funding to $52M, with Initialized Capital, Keen Venture Partners and SineWave Ventures leading and more than thirty further investors.
• Form D offering, 2020-01-24, $50M offered and $29.3M actually sold across thirteen investors under Rule 506(b).
• Series C, 2021-02-02, $50M, taking cumulative funding above $100M, with Hitachi Ventures, M12, Nautilus, NVIDIA, Republic Labs, Samsung Catalyst and Fort Ross Ventures among the participants.
• Series C extension, 2021-11-23, $55M, taking the Series C total to $105M and cumulative funding to $155M, with Fort Ross Ventures, Initialized Capital and Yield Capital leading.
• Venture round, 2024-09, $355K, led by Gaingels, recorded by Crustdata.
• Series D, 2025-04-07, $115M, with Applied Ventures, Atika Capital, Foxconn, Hanwha Asset Management Deeptech Venture Fund, Hitachi Ventures, NEC Orchestrating Future Fund, NVIDIA, Prosperity7, SineWave Ventures, Translink Capital, the University of Michigan and Y Combinator participating. No lead investor was disclosed.
• A further Series D reported by FundraisingFox in October 2025 with the amount undisclosed, which has no corroboration.

The company states cumulative capital raised of more than $260M. Databases report $272M, $272.4M and $273M, $234.36M across seven rounds, $220M across three rounds and a $146M snapshot as of February 2021, with a third-party compilation reporting more than $284M. The differences come from whether the small 2012 to 2014 rounds and the Form D issuances are counted.

Valuation for the April 2025 round exists in five mutually exclusive versions and this report does not average them: $672M pre-money with roughly $787M post-money from two placement firms, $715.3M post-money from Preqin, $750M from Multiples.vc, $889.6M from PrivateMarketView marked low confidence, and a $500M to $1B range from Dealroom. Separately, a November 2021 report put the valuation above $1B and called the company the first unicorn in cloud high-performance computing; that report came from a single outlet and the company has never confirmed any valuation. A defensible reading is $672M pre-money and roughly $750M post-money with a range of $715M to $790M, company confirmation absent.

No exit has occurred. Rescale remains privately held, no Form S-1 and no listing plan was found, and the actual event after the Series D was expansion rather than exit: government funding through the Department of Energy Genesis Mission in July 2026, a CoreWeave channel integration in August 2026 and a multi-year Honda agreement in October 2026. The company name, the legal entity identifier and the leadership team were all unchanged in October 2026.

DISCREPANCIES

Thirteen source conflicts were identified and adjudicated item by item.

1. Founding date. The government registration business start date is 18 November 2011, the YC directory says founded in 2011, and every company release says 2011. The competing figure of 2012 refers to the year the W12 batch began. Adjudicated: founded in 2011, joined Y Combinator as a W12 batch company in 2012. Both statements are true and the 2011 date is the founding year.

2. Founders. Joris Poort and Adam McKenzie are named consistently by more than twenty sources including official, YC and SEC records. The alternative names have no supporting source. Adjudicated: Poort and McKenzie. The alternative names are erroneous information.

3. Founders' academic background. Michigan, Washington and Harvard Business School for Poort, Oregon State for McKenzie, all from the official biographies. No source supports Stanford. Adjudicated: the official biographies stand and the Stanford claim is removed.

4. Headcount. Eight readings between 180 and 300-plus, most without a date. Adjudicated: roughly 245 to 250 in 2026, with the historical series quoted with dates at 252 in December 2022, 217 in November 2024 and 246 in September 2026.

5. Cumulative funding. Official more than $260M against database readings of $272M to $273M, $234.36M, $220M and $146M, plus a media compilation above $284M and an unverified report of $243M. Adjudicated: more than $260M as of 7 April 2025 from the official release. The database spread reflects inconsistent treatment of the small early rounds and the Form D issuances. The $243M figure is logically incompatible with the official number and is excluded from the main tables.

6. Valuation for the Series D. $672M pre-money, $715.3M post-money, $750M, $889.6M marked low confidence, and a $500M to $1B range. Adjudicated: $672M pre-money and roughly $750M post-money with a $715M to $790M range, stated as database-derived and never confirmed by the company. The $889.6M mark is not adopted because its publisher labels it low confidence.

7. The 2021 unicorn claim against the 2025 pre-money figure. A November 2021 report put the valuation above $1B and called it the first cloud HPC unicorn, repeated by Wikipedia and secondary media. The April 2025 database reading is $672M pre-money, which would be below the 2021 post-money figure. Adjudicated: both are listed with their source levels and neither is adopted as fact, because the company has confirmed no valuation and the two numbers may differ in pre-money versus post-money basis, in round identity, or may imply a down round. No averaging is performed.

8. Acquisition and takeover claims. No announcement, agreement, closing or completion has been found for any acquirer. The company's news index from October 2023 to October 2026 contains no merger entry, NVIDIA's published acquisition history for 2023 to 2025 does not include Rescale, and the legal entity name, the legal entity identifier, the domain and the management roster were unchanged in October 2026. Adjudicated: no acquisition. The single-source report of 23 June 2026 describing $150M at $1.1B led by NVIDIA is unverified, describes a new round rather than an acquisition, and is excluded from the main tables.

9. Annual revenue. Estimates run from $10M to $50M at Dealroom, about $75M quoted by ENGtechnica in January 2025, and $250M to $500M at LeadIQ in August 2026, a fifty-fold spread across eight database-level sources with no official figure. Adjudicated: revenue is not disclosed. The report uses the span of $10M to $500M with source level and date attached to each figure, and states that these cannot be used for valuation modelling. The 592% three-year growth and the Deloitte Fast 500 ranking are used only as directional corroboration, since both trace back to the same secondary compilation.

10. Application count against solver count. The official catalogue reads 1,000+ in 2021 to 2024, 1,250+ in April 2025 and 1,350+ in 2026; a third-party comparison states 50-odd solvers and 200-plus solvers on different pages, and a 2021 article states more than 800 applications. Adjudicated: the official figure counts application packages and versions, and independent commercial solvers run in the 200-plus range. The third-party figures undercount and are not adopted.

11. Pricing. The only verifiable official price list dates from 2016, giving on-demand rates of $0.10 to $0.30 per core-hour and prepaid rates of $0.05 to $0.09, while the 2026 pricing page is a guided configurator with no numbers. Adjudicated: the 2016 figures are cited with the year attached and explicitly marked as historical, and they are not used in any current cost model.

12. The claimed 2019 pivot to software-as-a-service. Contemporaneous coverage from May 2015 describes a platform-as-a-service for high-end computer simulations, and the 2018 Series B release is titled around platform-as-a-service development. Adjudicated: the company launched in 2015 already in platform form. The 2019 milestones were ScaleX Government with FedRAMP and the Siemens Simcenter software-as-a-service partnership.

13. Company name. The legal name has always been Rescale, Inc., with the legal entity identifier unchanged and the Delaware registration intact. ScaleX is a platform brand from 2015 onwards. One database records the company as ScaleX. Adjudicated: Rescale, Inc. is the company and ScaleX is a product brand; the database entry is a data error.

Risks and Diligence Gaps

Current pricing is not verifiable. The only official price list is a decade old and the pricing page has become a configurator. Diligence ask: a current rate card for compute, GPU and licence components, and the discount schedule for committed capacity.

Revenue, margin and concentration are entirely absent. No ARR, no revenue, no gross margin, no top-ten concentration, no Department of Defense contract values, no service-level terms. For a business reselling cloud compute and brokering commercial licences, margin is squeezed on both sides of the invoice. Diligence ask: revenue by line, gross margin before and after cloud cost, licence revenue share, and the ten largest contracts with terms.

Revenue estimates are unusable. Eight database sources span $10M to $500M, a fifty-fold range across different years and definitions. No valuation model should be built on them.

No valuation has investor backing. Five figures exist for one round and none is company-confirmed, while the 2021 unicorn claim has no primary source. Diligence ask: the Series D term sheet, the preferred share counts and the last-round board composition.

Performance claims lack third-party reproduction. The 1000x acceleration, the 78% solve-time reduction, the 4 to 8x design-candidate figure, the 99.99% and 99.9% reliability numbers and the near 500,000 vCPU concurrency figure are all first-party, with no published test conditions, baselines or error bounds. Diligence ask: benchmark definitions, raw logs and a customer reference willing to be named.

Licence brokering depends on upstream vendor authorisations. Rescale resells pay-per-use Ansys and Siemens entitlements and brokers licence handshakes for bring-your-own-licence setups, which requires agreements with each vendor that are not public. Diligence ask: the terms, the revenue split and the renewal status of each vendor agreement, and whether any exclusivity applies.

AWS concentration is undisclosed. Rescale owns no data centres and does not disclose the compute share taken from AWS, Azure, Google Cloud and Oracle. Hyperscaler bargaining power over pricing and roadmap is therefore unquantified. Diligence ask: compute spend by provider and the contractual terms of the largest one.

NVIDIA occupies three roles at once. It is a shareholder, the principal GPU supplier and the distribution channel for the European industrial AI cloud and DGX and RTX PRO capacity. Whether exclusivity obligations exist was not found in any public source. Diligence ask: any exclusivity, minimum-purchase or rights-of-first-refusal terms.

Cloud-side intellectual property and export control. Engineering IP runs across customer premises and multiple clouds, and the company markets ITAR, EAR and DFARS positions alongside FedRAMP, TISAX, SOC 2 and ISO 27001. Note that the 2019 release described FedRAMP Moderate as in process while also stating it had entered the FedRAMP Marketplace, and the 2025 material refers to new government and defence authorisations, so the final authorisation status is not clear from public sources. Diligence ask: the current FedRAMP authorisation status and the data-residency architecture for export-controlled work.

Academic conversion is unmeasured. The Academic Essentials platform licence is free to universities and the company states that its customers collectively spend over $1B a year on HPC infrastructure, but no academic-to-commercial conversion rate is published. Diligence ask: academic signups, paid conversion and average contract value by segment.

Workload concentration in one family. The named customer base is heavily aerospace, automotive and defence. A cut in automotive or aerospace R&D budgets would hit the same cohort of buyers at once. Diligence ask: revenue by vertical and by end customer.

The near 500,000 vCPU figure has no stated basis. It covers two providers and twelve regions with no definition of instance mix, and the largest single job scale was never published. Diligence ask: the definition of the concurrency metric and the peak single-job size actually served.

Cost competition against hyperscale-native HPC. As hyperscalers package simulation-ready instance types, managed schedulers and licence marketplaces, the abstraction layer becomes easier to compress. Diligence ask: a win-loss record against AWS Batch and ParallelCluster over the last eight quarters.

Narrative disturbance from the 2026 rumour. The unverified $150M report creates confusion about the actual round count and valuation, and its own cumulative figure is lower than the official one. Diligence ask: confirmation from the company or from NVIDIA's investment arm, and the most recent cap table.

Missing Form D filings. SEC Form D records appear for 2016, 2018 and 2020, with the most recent in January 2020, while the 2021 and 2025 rounds together amount to $170M. Whether further filings exist and whether the amounts match the press releases was not found in any public source. Diligence ask: the full financing history including any exempt issuance that was never filed.

Board and Form D officer gaps. Beyond Ben Verwaayen, the full board has never been published, and three names appearing in Form D records have no counterpart on the company site, in the releases or in SEC filings. Diligence ask: the current board roster and the role of every officer listed in the filings.

Post-October 2026 status is unknown. The research cut-off is 22 October 2026, the classic interface was sunset in May 2026, Auto Tune is listed as coming soon and the new interface migration has no acceptance data. Diligence ask: migration completion and stability metrics for the current platform release.

Overall Assessment

• Technology 7.5. A real control-plane architecture with more than sixty-five pre-tuned coretypes, a verified solver catalogue spanning ANSYS, SIMULIA, COMSOL, Simcenter, MSC, Altair, AVL and CONVERGE alongside open-source codes, a genuine two-track licence model, a Slurm-compatible interface with Ansys RSM and ARC integration, and a productised AI layer spanning AI Physics OS, agentic engineering with graded autonomy, surrogate modelling, an assistant and an agent evaluation framework. Deducted for the absence of any third-party reproduction of the headline performance figures, the absence of published error bounds on the surrogate claims, and the fact that the platform's own capability claims have no independent benchmark.
• Commercialisation 6.5. Named customers across automotive, aerospace, defence, chemicals and life sciences including Arm, Samsung, SLB, GM Motorsports, McLaren and Honda, a free academic programme, a trial path, and a placement inside the Ansys partner ecosystem. Deducted for the total absence of current list prices, revenue, ARR, gross margin and customer concentration, for the ten-year gap between the only verifiable price list and today, and for a revenue line that resells third-party compute and third-party licences.
• Funding 8.0. Ten-plus documented rounds from a $125,000 YC seed through $6.4M, $14M, $32M, $50M, $55M and finally $115M, with an investor roster that spans Amazon, NVIDIA, Samsung, Hitachi, Foxconn, NEC, Hanwha, TransLink, M12 and the University of Michigan, and official cumulative funding above $260M. Deducted for five mutually exclusive valuation figures with no company confirmation, for the unresolved relationship between the 2021 unicorn claim and the 2025 pre-money reading, for two rounds totalling $170M with no visible Form D, and for an unverified June 2026 funding report that complicates the record.
• Product maturity 7.0. A coherent six-layer product stack with a browser interface, a remote visualisation desktop, containers and Spack for bring-your-own-application work, containerised orchestration through CoreWeave Kubernetes Service, and a 2026 delivery cadence across Data Intelligence, AI Physics OS and Agentic Digital Engineering. Deducted for the classic interface being sunset in 2026 with no acceptance data on the replacement, for several announced capabilities still marked as coming soon, and for the pricing surface having been withdrawn rather than published.
• Market position 7.0. A credible position as the multi-vendor, solver-agnostic layer between hyperscalers and regulated engineering budgets, with FedRAMP, TISAX, ITAR and DFARS positions that few platform peers hold, a place in the European industrial AI cloud and in the Department of Energy Genesis Mission, and a decade of incumbency in a category where SimScale competes on simplicity and Inductiva has already ceased operations. Deducted for a third-party taxonomy that files the company as infrastructure rather than a physics contributor, for the absence of any published market-share figure for the platform layer, and for the structural risk that the hyperscalers are progressively selling the same abstraction themselves.

References

1. Rescale, "Rescale Secures $115 Million to Accelerate Innovation with AI-Driven Digital Engineering", 7 April 2025, rescale.com/news, for the Series D amount, the investor list and the statement of more than $260M raised.
2. Rescale news index, October 2023 to October 2026, rescale.com/news, for the complete event sequence, the absence of any acquisition item and the 2026 product announcements.
3. Rescale official platform pages, rescale.com/platform, including the architecture and coretype page, the software list, integrations and the pricing configurator.
4. Rescale, "Rescale Announces New Pricing Options", 2016, for the only verifiable official price list.
5. SEC EDGAR, Form D for Rescale, Inc., filed 24 January 2020, CIK 0001674741, for the Delaware incorporation, the Employer Identification Number and the offering and sale amounts.
6. Legal Entity Identifier record for Rescale, Inc., 254900UNWNJ44V69AT48, and SAM.gov entity record HPSAC9M3RS44, for the legal name, the jurisdiction, the business start date of 18 November 2011 and the headquarters address.
7. Y Combinator company directory entry for Rescale, for the W12 cohort, the founder names and the team-size field.
8. Crunchbase News weekly funding round-up, April 2025, and Silverwood Partners placement review for the week ending 13 April 2025, for the Series D and the $672M pre-money reading.
9. VentureBeat, "Rescale raises $55M more to meet demand for high-performance compute", 23 November 2021, for the Series C extension, the above-$1B valuation report and the 2021 network scale figures.
10. HPCwire, "Rescale Introduces High-Throughput Computing", August 2023, and Rescale's High-Throughput Computing release, for the near 500,000 vCPU, twelve-region, 99.99% reliability claims.
11. NVIDIA technical blog post authored by the chief executive, "How to Run AI-Powered CAE Simulations", 10 April 2026, and the NVIDIA European industrial AI cloud announcement of 16 June 2025, for the 1000x, 10x and 30x claims.
12. CoreWeave announcement of 25 August 2026, and Rescale's Department of Energy Genesis Mission release of 23 July 2026, for the Kubernetes Service integration and the $293M Phase I award.
13. Ansys partner ecosystem page for Rescale, for the seven of ten automakers, seven of ten aerospace and three of six life-science statements and the compliance attestations.
14. Dealroom, Preqin, Crustdata, LeadIQ, PrivateMarketView, Multiples.vc and FundedIQ company records, accessed 22 October 2026, for the conflicting funding, valuation, headcount and revenue figures used in the discrepancies section.

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.