Cadence Design Systems, Inc.¶

Company Overview¶

Business Overview¶

Cadence Design Systems was formed in 1988 (from the merger of SDA Systems and ECAD) and is headquartered in San Jose, California, listed on NASDAQ (CDNS, an S&P 500 constituent). Anirudh Devgan — a computational-software researcher by background — has been President since 2017 and CEO since December 2021. Cadence is one of the "big three" of EDA alongside Synopsys and Siemens EDA. [Synthesis]

Key financials:

FY2025 (reported 17 February 2026): revenue $5.297 billion, up over 14%; non-GAAP operating margin ~45%; year-end backlog $7.8 billion. Core EDA grew 13%; the IP business grew ~25%. Initial 2026 guidance: $5.9–6.0 billion. [Evidence] ¹
Q2 2026 (reported 27 July 2026): revenue $1.585 billion, up 24% year-on-year; System Design & Analysis (SDA) — the division that now houses all the physics simulation — grew 37%, helped by the Hexagon D&E integration, which management said is "progressing as expected" and contributed about four points of recurring-revenue growth. Full-year guidance raised to ~$6.3 billion (19% growth); backlog $8.1 billion. [Evidence] ²

The acquisition trail into physical simulation (the reason this report exists):

Year	Acquisition	What it brought
2021	NUMECA (completed Feb/Mar 2021)	CFD solvers; users incl. NASA, Ariane Group, Honda, Ford¹⁷¹⁸
2021	Pointwise (Apr 2021)	CFD meshing¹⁸
2022	Future Facilities (Jul 2022)	Data-centre digital twins / electronics cooling¹⁹
2024	BETA CAE ($1.24bn; announced Mar, completed Jun 2024)	ANSA/META — automotive CAE pre/post-processing standard⁶⁷
2025	ChipStack (Nov 2025)	Agentic-AI startup for chip design/verification¹³
2026	Hexagon D&E (€2.7bn; announced 2 Sep 2025, completed 23 Feb 2026)	MSC Nastran, Adams, Marc, Apex, Cradle CFD, Simufact, Digimat, ODYSSEE³⁴

[Evidence]

Automotive Business¶

Cadence does not report an automotive segment. Its automotive exposure is three-fold. First, automotive chips: EDA tools and Tensilica IP used by semiconductor and increasingly by system companies designing vehicle electronics. Second, BETA CAE: at acquisition Cadence stated BETA CAE serves "the top 10 global automobile manufacturers" and most Formula One teams — a primary-source but promotional claim; no OEM is individually named in the release. [Evidence + Marketing] ⁶ Third, the inherited MSC base: MSC Nastran/Adams are long-standing automotive NVH, durability and vehicle-dynamics tools, and the Digimat/ODYSSEE line carried Hexagon-era automotive users — webinars named General Motors (manufacturing-process effects on structures, with a Moldflow-certified GM engineer), Schaeffler, DOMO Chemicals and Radici Group (uncertainty quantification with recycled materials). [Evidence — Hexagon-era] ²⁴ A partner-published example shows PSA Group using ODYSSEE reduced-order models to predict MSC Nastran vibro-acoustic results. [Evidence — partner-reported] ²⁵

Engineering Business¶

Cadence's core is chip design: digital implementation and signoff, verification (including Palladium/Protium emulation hardware), custom/analog, PCB (Allegro), and semiconductor IP. Since 2020 it has deliberately expanded into System Design & Analysis — physics simulation for the systems that chips go into — via the acquisition trail above, under the strategy banner "Intelligent System Design," extended in 2025–26 to "Physical AI." [Evidence] ³ [Marketing]

Role in Automotive Moulded Parts Workflow¶

Workflow Stage	Relevant?	Notes (product)
Requirements	No	—
Industrial Design	No	—
Concept Design	No	—
CAD Modelling	No	No mechanical part-design CAD (ANSA is CAE pre-processing, not design)
Engineering Review	Marginal	META post-processing, simulation reports
Simulation	Strong (new)	MSC Nastran/Adams/Marc + BETA CAE + Fidelity/Cradle CFD
DFM	No	No moulding DFM tooling of any kind
Tool Design	No	(VISI stayed at Hexagon)
Mould Flow	No solver	Digimat sits downstream, consuming Moldflow/Moldex3D results
Prototype	Marginal	Virtual prototyping via simulation only
Validation	Partial	Structural/durability/NVH validation of moulded+overmoulded parts
Manufacturing Engineering	Marginal	Simufact is metal-centric
Production Release	No	—

What makes Cadence structurally different: it arrives from outside mechanical engineering entirely. Every other platform vendor in this study grew out of CAD or CAE; Cadence grew out of chip design, where it built and shipped production machine learning for years, and then bought a mechanical simulation stack (BETA CAE 2024, Hexagon D&E 2026) that happens to include the moulding-adjacent Digimat/ODYSSEE layer. The moulding solver itself it neither has nor bought. [Synthesis]

AI Strategy¶

Public AI Vision¶

Cadence frames its whole portfolio around AI at three levels. [Synthesis]

AI inside the tools — ML and agents that design chips: Cerebrus (reinforcement learning for physical implementation), Verisium (ML-driven verification), Cerebrus AI Studio (agentic, multi-block), ChipStack (agentic "Super Agent"), all sharing the JedAI data platform. [Evidence] ⁸⁹¹¹¹³
Tools for AI — designing the chips and data centres that run AI (the bulk of its revenue growth), plus the Millennium GPU supercomputers built with NVIDIA for accelerated simulation. [Evidence] ¹⁵
"Physical AI" — the marketing banner over the mechanical expansion: the claim that robots, autonomous vehicles and electromechanical systems will need the same design-automation treatment as chips, and that multibody dynamics (Adams) plus multiphysics will be the substrate. Both Hexagon-deal press releases carry it in the title. [Marketing] ³⁴

Investor Statements¶

CEO Anirudh Devgan, on the Hexagon deal (September 2025): "By adding Hexagon's D&E world-class simulation capabilities, we will expand our vision of Intelligent System Design to encompass the full spectrum of physical behavior — from electromagnetics and fluids to structures and motion." [Marketing] ³
Devgan, at completion (February 2026): the combination will "empower customers to push the boundaries of what's possible — from autonomous systems and advanced robotics to the future of transportation." [Marketing] ⁵
FY2025 call (February 2026): 2025 closed "with over 14% revenue growth and 20% non-GAAP EPS growth"; Hexagon D&E expected to add ~$160m to 2026 revenue, ~28 cents dilutive in 2026, accretive in 2027. [Evidence] ¹⁴
Jensen Huang (NVIDIA CEO), May 2025: "NVIDIA plans to purchase 10 Millennium Supercomputer systems… to accelerate the company's chip design workflows" — the clearest third-party endorsement of the compute line. Note this is a partnership-and-purchase relationship; no NVIDIA equity investment in Cadence was found. [Evidence] ¹⁶

Engineering AI Strategy¶

Where Dassault or Siemens talk about AI they intend to ship, Cadence talks about AI it has already shipped, at scale, for years — in chip design. The pattern of its real ML work: reinforcement learning over tool settings and floorplans (Cerebrus), supervised/big-data ML over verification runs (Verisium), a common data platform (JedAI), and now LLM-based agents orchestrating the classical tools (Cerebrus AI Studio, ChipStack — whose "Mental Model" is described as a grounded source of truth agents must reference to limit hallucination). The open strategic question for this study is whether that playbook — learn from thousands of tool runs, then let agents drive the tools — gets applied to the inherited mechanical portfolio. As of August 2026 there is no public evidence that it has been. [Synthesis] [Evidence of absence]

Timeline of AI Evolution¶

Date	Event
Jul 22, 2021	Cerebrus Intelligent Chip Explorer — RL-based chip-design optimization ships⁸
Jun 8, 2022	Optimality Intelligent System Explorer — "AI-driven" multiphysics system optimization¹⁰
Sep 13, 2022	Verisium AI-driven verification platform + JedAI data platform⁹
Feb 1, 2024	Millennium M1 — "first accelerated digital twin" / CFD supercomputer¹⁴
Mar–Jun 2024	BETA CAE acquired ($1.24bn) — automotive CAE entry⁶⁷
May 7, 2025	Millennium M2000 with NVIDIA Blackwell; NVIDIA to buy 10 systems¹⁵¹⁶
May 7, 2025	Cerebrus AI Studio — agentic, multi-block, multi-user SoC design (see App. C on date)¹¹¹²
Sep 2, 2025	Hexagon D&E acquisition announced (€2.7bn) — "Physical AI" framing³
Nov 2025	ChipStack startup acquired¹³
Feb 10, 2026	ChipStack AI Super Agent launched; early users NVIDIA, Altera, Tenstorrent¹³
Feb 23, 2026	Hexagon D&E completed — Digimat + ODYSSEE become Cadence products⁴
~May 2026	"Fully Autonomous Virtual Engineer" for chip design announced with NVIDIA (headline verified only)²⁶
Jul 27, 2026	Q2 2026: SDA +37%, D&E integration "progressing as expected"²

Products Relevant to Engineering¶

Cerebrus / Cerebrus AI Studio — chip implementation optimization. AI: genuine reinforcement learning (rewards for meeting power/performance/area targets), extended in 2025 to an agentic multi-block platform claiming 5–10× faster SoC delivery. Moulded parts: none. [Evidence] ⁸¹¹

Verisium + JedAI — verification platform on a shared data platform. AI: ML across "multiple runs of multiple engines" of verification data; JedAI is the common substrate under Verisium, Cerebrus and Optimality. Moulded parts: none. [Evidence] ⁹

ChipStack AI Super Agent — agentic front-end design and verification (February 2026), "a symphony of agents" that write code and testbenches, run regressions, debug; up to 10× productivity claimed; grounded by a "Mental Model." Moulded parts: none — but this is the most advanced shipped agentic-AI product of any vendor in this study. [Evidence + Marketing] ¹³

Optimality Intelligent System Explorer — multiphysics optimization of electronic systems (2022). AI: branded "AI-driven optimization"; the public material does not disclose the mechanism. Per the study's concept note, treat optimization-branded-as-AI with care: What is Optimization technology — though given Cadence's record, ML-assisted design-space exploration is plausible here. [Evidence + Marketing] ¹⁰²⁹

Millennium M1 / M2000 — GPU supercomputers for simulation (Feb 2024 / May 2025, with NVIDIA Blackwell in the M2000; up to "80×" over CPU predecessors is NVIDIA/Cadence's own figure). Compute, not AI — the "AI-driven" in the branding refers to what runs on it. [Evidence + Marketing] ¹⁴¹⁵¹⁶

BETA CAE ANSA / META — the automotive-standard CAE pre/post-processor (acquired 2024). No AI capability was verified in ANSA/META during this research. Moulded parts: meshes and post-processes plastic components like any others. [Evidence] ⁶ [Evidence of absence]

MSC Nastran / Adams / Marc / Apex / Cradle CFD / Simufact — the inherited Hexagon D&E stack (February 2026). Classical physics solvers; Apex Generative Design is topology optimization (see concept note²⁹). Moulded parts: structural analysis of moulded components via Digimat coupling. [Evidence] ⁴

Digimat — multi-scale material modelling for fibre-reinforced plastics; the bridge that carries fibre-orientation results from Moldflow/Moldex3D into structural simulation. Continued shipping post-acquisition: Digimat 2026.1 enhanced the injection-gate-position analysis and uncertainty quantification workflows (multiple key results — failure indicators, stresses, strains — and DoE import "from any supported injection molding software"), plus curing-material calibration in Digimat-MX. This is the closest any Cadence product comes to a moulding decision. [Evidence] ²¹²²

ODYSSEE — the machine-learning surrogate platform (reduced-order models trained on small sets of simulations or tests; Quasar engine). Marketed by Cadence as "AI Simulation & Digital Twin Software." Its documented plastics-adjacent deployment remains the Hexagon-era Digimat-RP 2023.2 Quasar UQ plugin. [Evidence] ²⁰²³²⁸

AI Capabilities¶

1. Cerebrus (RL chip optimization). Reinforcement-learning agent that tunes chip physical-implementation flows; shipped July 2021; the earliest production reinforcement-learning product of any vendor in this study. Passes the honest test — it learns from tool-run data. Limitation: chip design only. [Evidence] ⁸

2. Cerebrus AI Studio (agentic implementation). Multi-block, multi-user agentic platform, launched May 7, 2025; "one engineer, many blocks"; 5–10× faster delivery and up-to-20% PPA (power, performance, area — chip-design quality metrics) claims are vendor figures. Limitation: chip design only. [Evidence + Marketing] ¹¹¹²

3. Verisium (ML verification) on JedAI. ML across verification campaigns — bug triage, coverage, regression optimization — on a common data platform. Limitation: chip design only. [Evidence] ⁹

4. ChipStack AI Super Agent (agentic design/verification). Launched February 2026 from the November 2025 ChipStack acquisition; multiple cooperating agents drive Cadence tools end-to-end; early deployment at NVIDIA, Altera, Tenstorrent. Limitation: chip design only; productivity numbers are vendor claims. [Evidence + Marketing] ¹³

5. Optimality (system optimization). "AI-driven" multiphysics optimization for electronic systems (2022). Limitation: mechanism undisclosed; treat the AI label per the concept note²⁹. Electronics, not mechanics. [Evidence + Marketing] ¹⁰

6. ODYSSEE (inherited ML surrogates). Genuine machine learning: train reduced-order models on a handful of expensive simulations or tests, predict new variants in near-real time, including full time histories (crash, NVH); PSA vibro-acoustic ROM example. This is the one Cadence-owned AI capability that sits next to the moulding toolchain (via Digimat-RP). Limitation: no public case of a mould-fill/warp surrogate — the ML stops one step short of the moulding simulation, exactly as it did under Hexagon. [Evidence] ²⁰²⁵²⁸ [Evidence of absence]

7. Digimat UQ / gate-position workflows (inherited). Digimat-MS provides an injection-gate-position analysis and uncertainty-quantification workflows, enhanced in 2026.1 (DoE import from any supported injection-moulding software; multiple key results). Gate location is a real moulding-DFM decision — but the machinery is design-of-experiments plus surrogate evaluation of structural consequences, with the moulding physics still computed by third-party Moldflow/Moldex3D. Tag per the concept note: automated search plus ROMs, not a system that has learned moulding. [Evidence] ²² [Synthesis] ²⁹

8. Millennium M1/M2000 (accelerated compute). GPU-resident solvers (e.g. Fidelity LES CFD) at supercomputer scale; NVIDIA Blackwell inside the M2000; NVIDIA itself buying 10 systems. Limitation: this is compute for physics and for AI training — not itself AI. [Evidence] ¹⁴¹⁵¹⁶

The pattern. Capabilities 1–5 are genuine, shipped, revenue-generating AI — and all of them live in chip design. Capabilities 6–7 are the inherited moulding-adjacent layer — and they predate Cadence. No capability yet combines the two. [Synthesis]

Engineering Workflow Contribution¶

For an automotive moulded part, Cadence's realistic contribution today is the same as Hexagon's was, under new ownership:

part CAD (elsewhere: CATIA/NX) → mould-flow simulation (elsewhere: Moldflow/Moldex3D) → Digimat maps fibre orientation and material behaviour → ANSA pre-processes → MSC Nastran / Marc / Adams compute stiffness, durability, NVH, system dynamics → META post-processes → optionally ODYSSEE builds ML surrogates over those structural runs for fast variant exploration and UQ.

Cadence enters after the part exists and after the moulding simulation has run, and it exits before tooling and production. The chip-side AI (Cerebrus, Verisium, ChipStack) runs in a different universe and touches none of this. [Synthesis]

Public Customer Evidence¶

Case Studies¶

NVIDIA, Altera, Tenstorrent — named early-deployment users of the ChipStack AI Super Agent (February 2026). Strong, current, named — and chip design, not moulding. [Evidence] ¹³
NVIDIA as Millennium customer — Jensen Huang's stated plan to purchase 10 Millennium systems (May 2025). [Evidence] ¹⁶
BETA CAE automotive base — "top 10 global automobile manufacturers" and most F1 teams, per Cadence's own acquisition release; no OEM individually named. [Evidence + Marketing] ⁶
NUMECA users — NASA, Ariane Group, Honda, Ford named in Cadence's 2021 releases (CFD, pre-acquisition heritage). [Evidence] ¹⁷¹⁸
Digimat/ODYSSEE automotive users (Hexagon-era) — General Motors (manufacturing-induced effects on structural performance; the GM speaker's profile references Autodesk Moldflow certification — i.e. the moulding simulation feeding Digimat was a third party's), Schaeffler (AI-enriched creep/ fatigue material data via Nexus Materials), DOMO Chemicals (fatigue of short- fibre-reinforced polyamides), Radici Group (UQ for recycled materials). These demonstrate the acquired layer's real automotive-plastics use — but they are Hexagon-era webinars, not Cadence announcements. [Evidence — Hexagon-era]²⁴
PSA Group — ODYSSEE ROM predicting MSC Nastran vibro-acoustic results; partner-published (SIMTEQ), undated precisely. [Evidence — partner-reported] ²⁵

Conference Demonstrations¶

CadenceLIVE Silicon Valley 2025 (Millennium M2000 launch with NVIDIA). [Evidence] ¹⁵

White Papers / Testimonials¶

Cerebrus AI Studio white paper (2025); ChipStack "Mental Model" description in launch coverage. [Evidence] ¹¹¹³

The gap. A public story combining Cadence AI + moulded plastic parts + a quantified outcome does not exist as of August 2026. The automotive-plastics evidence that does exist belongs to the inherited Hexagon-era products and predates the acquisition. [Evidence of absence]

Technical Architecture (Inferred)¶

What the sources establish¶

JedAI — a unified data platform under the chip-side AI products (Verisium, Cerebrus, Optimality). [Evidence] ⁹
Millennium M1/M2000 — GPU-resident solvers plus NVIDIA HGX B200 / RTX PRO 6000 Blackwell hardware and CUDA-X libraries; positioned for both physics simulation and AI training/inference. [Evidence] ¹⁵¹⁶
ChipStack — agents orchestrating classical EDA tools, grounded by a "Mental Model" as source of truth. [Evidence] ¹³
The inherited stack — MSC solvers, BETA CAE pre/post, Digimat, ODYSSEE — arrived 23 February 2026 as a separate, mature product family. [Evidence] ⁴

Putting it together¶

Cadence now runs two architectures side by side: a chip-design stack with a shared data platform and years of production ML on top, and a freshly bought mechanical-simulation stack whose own ML (ODYSSEE) grew up independently. The company's stated integration story is portfolio-level ("end-to-end multiphysics platform"), not data-platform-level. [Synthesis] ⁴

Reading between the lines¶

Nothing public says the MSC/Digimat/ODYSSEE tools will join JedAI, gain Cerebrus- style RL optimization, or get ChipStack-style agents. Equally, nothing public rules it out — and Cadence's five-year pattern (build data platform → add ML → add agents) is exactly the playbook that could be pointed at simulation workflows. Five months after closing (and with Q2 2026 saying integration is "as expected"), the honest position is: capability proven, intent toward moulding unstated. [Inference]

AI Technologies¶

In one list: reinforcement learning (Cerebrus — shipped, production); agentic AI / LLM agents (Cerebrus AI Studio, ChipStack Super Agent — shipped); big-data ML on engineering runs (Verisium/JedAI — shipped); ML surrogates / reduced-order models (ODYSSEE — inherited, genuine); uncertainty quantification + DoE (Digimat-MS workflows — classical machinery with ROM support²⁹); topology optimization (MSC Apex Generative Design — optimization, not AI²⁹); GPU-accelerated physics (Millennium, Fidelity — compute, not AI); digital twins (Future Facilities lineage, ODYSSEE marketing). [Evidence + Synthesis]

Research Publications¶

Papers / Patents¶

Cadence's computational-software research output (and Devgan's own background) is extensive on the EDA side, and the Cerebrus/Verisium line is documented in vendor white papers rather than peer-reviewed benchmarks. This report did not attempt a patent sweep; no Cadence paper or patent applying ML to injection moulding or reinforced plastics was encountered in any search performed for this report. The Digimat/ODYSSEE academic lineage (e-Xstream/UCL; CADLM) is covered in the Hexagon report. [Evidence of absence — bounded by search effort; see Appendix C]

Standards¶

Nothing moulding-relevant found. [Evidence of absence]

Open Source¶

Near-nil for AI — reported plainly. [Evidence of absence] ²⁷

The Cadence-Design-Systems GitHub organisation exists but has no public repositories at all.
A Hugging Face account named "Cadence" exists but shows no affiliation, no models and no datasets — it cannot be attributed to the company.
Disambiguation warning: the well-known open-source workflow engine called "Cadence" is Uber's project, not Cadence Design Systems' — do not credit it here.
This matches the business model: Cadence monetizes proprietary tools and IP; its AI is a paid differentiator, not a community artifact.

Engineering Service / Platform Mapping¶

Cadence owns no mainstream mechanical CAD or PLM (no CATIA/NX/Creo/ Teamcenter/Windchill equivalent). Its mechanical stack interoperates the CAE way: ANSA reads geometry from the major CAD systems; Digimat couples to third-party moulding solvers (Moldflow, Moldex3D) and third-party as well as in-house FEA. Chip-side platforms (Virtuoso, Innovus, Allegro, Palladium) are out of scope here. [Synthesis]

Engineering Intelligence Stack Mapping¶

Intent — absent for mechanical work; ChipStack captures design intent via natural language, but for chips. [Evidence — chip-side] ¹³
Knowledge — JedAI aggregates chip-design data; no equivalent exists (or is announced) over simulation/materials data on the inherited side. The Digimat material-model lineage is itself a structured knowledge asset. [Synthesis]
Reasoning — strong and genuine on chips (RL, ML, agents); on mechanics, inherited ODYSSEE surrogates and Digimat UQ — one step short of moulding. [Evidence] ²⁰
Execution — agents drive EDA tools end-to-end (ChipStack); nothing comparable drives the mechanical tools. [Evidence — chip-side] ¹³
Feedback — none in the moulded-parts sense (no loop from measured parts or production outcomes; the metrology side stayed at Hexagon). [Evidence of absence]

Strengths¶

Proven production AI, at scale, with revenue attached — RL since 2021, agentic since 2025–26; the deepest genuine-AI bench in this study. [Evidence] ⁸¹³
Owns the ML-next-to-moulding layer — Digimat + ODYSSEE, the study's best-documented case of real ML adjacent to the moulding toolchain. [Evidence] ²³²⁸
Automotive CAE beachhead — BETA CAE's ANSA user base plus the MSC structural franchise. [Evidence] ⁶⁴
Compute muscle and the NVIDIA relationship — Millennium M2000; NVIDIA as both partner and customer. [Evidence] ¹⁵¹⁶
Financial capacity — ~45% operating margins funding serial acquisition. [Evidence] ¹

Weaknesses¶

No injection-moulding solver — and no announced intent to build or buy one; the moulding physics in its workflows remains Autodesk's or CoreTech's. [Evidence of absence]
No stated Digimat/ODYSSEE roadmap — five months post-close, integration statements are portfolio-level generalities. [Evidence of absence] ⁴
Its genuine AI has never shipped in a mechanical product — the entire track record is EDA; transferring the playbook is plausible, not demonstrated. [Synthesis]
No mechanical CAD/PLM anchor — unlike Dassault/Siemens/PTC, it cannot own the moulded part's design-side workflow. [Synthesis]
Integration risk — simultaneous digestion of BETA CAE, ChipStack and a 1,100+-person Hexagon D&E business. [Inference] ³

Current Gaps (largely manual today)¶

Everything that was manual under Hexagon remains manual under Cadence: interpreting mould-flow results, moulding DFM decisions, gate/rib/wall design judgement, correlation of simulation to actual moulding outcomes. Additionally: no geometry retrieval over historical part libraries, no learning loop from production data, and — new gap relative to other vendors — no design-side CAD automation at all for mechanical parts. [Evidence + Inference]

Future Direction¶

"Physical AI" as the stated destination: simulation + AI for robotics, autonomy and electromechanical systems, with Adams multibody dynamics explicitly positioned for robotics. [Evidence + Marketing] ³
Integration first — 2026 guidance and commentary treat D&E as a revenue and portfolio integration, "progressing as expected." [Evidence] ²
My read of the roadmap: the rational sequence is (1) GPU-accelerate the MSC solvers on Millennium, (2) spread ODYSSEE surrogates across the acquired solver portfolio as the "AI layer," (3) eventually apply the agentic playbook to simulation workflows. If step (2) ever reaches Digimat's moulding-coupled workflows at product level — an ML surrogate of fill/warp, or learned gate placement — Cadence would leapfrog every vendor in this study. Nothing public commits it to that. [Inference]

Relevance to Automotive Moulded Parts¶

Capability	Strength	Notes
Plastic Part Design	None	No mechanical design CAD
Surface Design (Class A)	None	—
CAD Automation	None (mechanical)	Its CAD automation is for chips
DFM	None	No moulding DFM product; gate-position analysis in Digimat evaluates structural consequences, not mouldability rules
Tool Design	None	VISI stayed at Hexagon
Mould Flow	No solver	Digimat consumes Moldflow/Moldex3D output — the layer around moulding, inherited intact
Manufacturing Engineering	Marginal	Simufact (metal-centric)
Quality	None	Metrology stayed at Hexagon
Engineering Knowledge Reuse	Latent	Digimat material-model assets; JedAI pattern unproven outside EDA

Critical finding. Cadence is the mirror image of every other platform vendor in this study. The others have moulded-parts products with little or no real AI; Cadence has abundant real AI with no moulded-parts products of its own — only the inherited layer around other vendors' moulding solvers. After deliberate searching: no Cadence statement on a moulding or plastics roadmap, no documented application of Cerebrus-style RL, Optimality or agentic AI to moulding or reinforced plastics, and no moulding solver anywhere in the portfolio. The closest genuine-ML-next-to-moulding artifact remains the ODYSSEE Quasar UQ plugin in Digimat-RP 2023.2 — built under Hexagon — with the Digimat 2026.1 release showing the line is at least still alive and advancing on injection-adjacent workflows under the new owner. [Evidence of absence] [Evidence] ²²²⁸

Net. Today, "Cadence for moulded parts" means exactly what "Hexagon D&E for moulded parts" meant — same products, new logo. What changed is the owner's capability: this owner demonstrably knows how to industrialize AI. Watch the trajectory, not the present. [Synthesis]

Key Takeaways¶

Cadence became a moulded-parts-relevant vendor by acquisition, on 23 February 2026 — €2.7bn for Hexagon D&E, including Digimat and ODYSSEE. [Evidence] ⁴
It has the deepest genuine production-AI record in this study — Cerebrus RL (2021), Verisium/JedAI (2022), Cerebrus AI Studio (2025), ChipStack Super Agent (2026, deployed at NVIDIA/Altera/Tenstorrent). All chip design. [Evidence] ⁸⁹¹³
It has no injection-moulding solver — it inherited the layer around moulding, exactly as Hexagon held it. [Evidence of absence]
No Cadence AI has a documented moulding application; the closest remains the inherited Digimat-RP 2023.2 ODYSSEE/Quasar UQ plugin. [Evidence] ²³ [Evidence of absence]
The inherited line still advances: Digimat 2026.1 enhanced injection-gate- position and UQ workflows (DoE + surrogates — classical machinery per the concept note²⁹). [Evidence] ²²
Automotive presence is real but inherited: BETA CAE's "top 10 OEMs" claim, MSC's franchise, Hexagon-era GM/Schaeffler/DOMO/Radici webinar evidence. [Evidence + Marketing] ⁶²⁴
"Physical AI" is a banner, not yet a product for mechanics — the shipped physical-side offering is classical solvers plus GPU acceleration. [Marketing] ³
NVIDIA relationship is partnership + customer (10 Millennium systems), not a verified equity investment. [Evidence] ¹⁶
Open-source AI footprint: near-nil (GitHub org with zero public repos). [Evidence of absence] ²⁷
Verdict for this study: capability proven elsewhere, intent here unstated — the vendor to re-check first in any future revision. [Synthesis]

References¶

Primary Sources — Cadence¶

Q4/FY2025 results, Feb 17 2026 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr-ir/2026/cadence-reports-fourth-quarter-and-fiscal-year-2025-financial.html (also https://www.businesswire.com/news/home/20260216202086/en/Cadence-Reports-Fourth-Quarter-and-Fiscal-Year-2025-Financial-Results)
Q2 2026 results, Jul 26/27 2026 — https://www.morningstar.com/news/business-wire/20260726697902/cadence-reports-second-quarter-2026-financial-results
Cadence to Acquire Hexagon's D&E Business ("Physical AI" title), Sep 2 2025 — https://www.businesswire.com/news/home/20250902498199/en/Cadence-to-Acquire-Hexagons-Design-Engineering-Business-Accelerating-Expansion-in-Physical-AI-and-System-Design-and-Analysis (also https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2025/cadence-to-acquire-hexagons-design--engineering-business.html)
Cadence Completes Acquisition of Hexagon's D&E Business, Feb 23 2026 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2026/cadence-completes-acquisition-of-hexagons-design-and-engineering.html
Cadence to Acquire BETA CAE, Mar 2024 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2024/cadence-to-acquire-beta-cae-expanding-into-structural-analysis.html
Cadence Completes Acquisition of BETA CAE, Jun 2024 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2024/cadence-completes-acquisition-of-beta-cae.html
Cerebrus product page — https://www.cadence.com/en_US/home/tools/digital-design-and-signoff/soc-implementation-and-floorplanning/cerebrus-intelligent-chip-explorer.html
Cerebrus AI Studio product page + white paper — https://www.cadence.com/en_US/home/tools/digital-design-and-signoff/soc-implementation-and-floorplanning/cadence-cerebrus-ai-studio.html ; https://www.cadence.com/en_US/home/resources/white-papers/cadence-cerebrus-ai-studio-agentic-ai-multi-block-multi-user-soc-wp.html
Verisium launch, Sep 13 2022 — https://www.businesswire.com/news/home/20220913005378/en/Cadence-Revolutionizes-Verification-Productivity-with-the-Verisium-AI-Driven-Verification-Platform
Optimality Explorer launch, Jun 8 2022 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2022/cadence-revolutionizes-system-design-with-optimality-explorer.html
ChipStack AI Super Agent PR, Feb 2026 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2026/cadence-unleashes-chipstack-ai-super-agent-pioneering-a-new.html
Millennium M1 unveiling, Feb 1 2024 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2024/cadence-unveils-millennium-platformindustrys-first-accelerated.html
Millennium M2000 with NVIDIA Blackwell, May 7 2025 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2025/cadence-unveils-millennium-m2000-supercomputer-with-nvidia.html
"Fully Autonomous Virtual Engineer," ~May 2026 (headline only) — https://www.businesswire.com/news/home/20260531072918/en/Cadence-Unveils-Industrys-First-Fully-Autonomous-Virtual-Engineer-for-Chip-Design-powered-by-NVIDIA
NUMECA completion, 2021 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2021/cadence-completes-acquisition-of-numeca.html
Pointwise acquisition, Apr 15 2021 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2021/cadence-acquires-pointwise-to-expand-system-analysis-offerings-a.html
ODYSSEE product page (Cadence) — https://www.cadence.com/en_US/home/tools/msc-software/odyssee.html
Digimat product page (Cadence) — https://www.cadence.com/en_US/home/tools/msc-software/digimat.html
What's New in Digimat 2026.1 (MSC support portal; JS-gated, content via search index) — https://mscsoftware.my.site.com/customers/s/article/What-s-New-in-Digimat-2026-1
Agentic AI for Chip and System Design — https://www.cadence.com/en_US/home/ai/ai-for-design.html
GitHub organisation — https://github.com/Cadence-Design-Systems

Customer / Partner Evidence¶

NVIDIA blog (M2000; Huang purchase quote), May 7 2025 — https://blogs.nvidia.com/blog/cadence-millennium-nvidia-blackwell/
Hexagon Digital Materials webinar series (GM, Schaeffler, DOMO, Radici; Hexagon-era) — https://go.mi.hexagon.com/Hexagon-Multiscale-I-Digital-Materials-Webinar-Series-2025
ODYSSEE brochure (Hexagon MI, © 2023; NAFEMS-hosted) — https://www.nafems.org/downloads/dropbox/nologin/nrc24_dach/hexagon/hexagon-mi_odyssee_brochure.pdf
SIMTEQ Engineering (partner): ML/AI for CAE incl. PSA ROM example — https://simteq.co.za/blog/machine-learning-and-artificial-intelligence-ai-for-cae/

Secondary Sources¶

engineering.com: "Cadence closes on Nastran, Adams, and more from Hexagon," Feb 24 2026 — https://www.engineering.com/cadence-closes-on-nastran-adams-and-more-from-hexagon/
SiliconANGLE: ChipStack Super Agent, Feb 10 2026 — https://siliconangle.com/2026/02/10/cadence-announces-chipstack-super-agent-system-chip-design-verification/
VentureBeat: Cerebrus launch, Jul 22 2021 — https://venturebeat.com/ai/cadence-design-systems-launches-cerebrus-machine-learning-for-chip-design
DCD: Future Facilities acquired, 2022 — https://www.datacenterdynamics.com/en/news/data-center-digital-twin-firm-future-facilities-acquired-by-cadence-design-systems/
StockTitan (10-K note: FY2025 revenue $5.3B) — https://www.stocktitan.net/sec-filings/CDNS/10-k-cadence-design-systems-inc-files-annual-report-6bcef311af10.html
Futurum: Q2 FY2026 earnings analysis — https://futurumgroup.com/insights/cadence-q2-fy-2026-earnings-climb-on-agentic-ai-and-record-backlog/
Zacks: BETA CAE completion, Jun 5 2024 — https://www.zacks.com/stock/news/2283843/cadence-cdns-acquires-switzerland-based-beta-cae-for-124b
EE Times: ChipStack AI agent — https://www.eetimes.com/cadence-unveils-chipstack-ai-agent-for-agentic-chip-design-and-verification/

Cross-references within this study¶

Hexagon report — full history of Digimat, ODYSSEE and the Digimat-RP 2023.2 Quasar UQ plugin under the previous owner.
What is Optimization technology — the honest-AI test applied to optimization-branded products.

Appendix A — Timeline¶

See §3. Key dates: Cerebrus (Jul 2021) → Optimality (Jun 2022) → Verisium + JedAI (Sep 2022) → Millennium M1 (Feb 2024) → BETA CAE (Mar–Jun 2024) → Millennium M2000 + NVIDIA (May 7, 2025) → Cerebrus AI Studio (May 7, 2025) → Hexagon D&E announced, "Physical AI" (Sep 2 2025) → ChipStack acquired (Nov 2025) → ChipStack Super Agent (Feb 10 2026) → Hexagon D&E completed — Digimat/ODYSSEE become Cadence products (Feb 23 2026) → Q2 2026: SDA +37%, guidance ~$6.3B (Jul 2026).

Appendix B — Glossary¶

EDA — electronic design automation: software for designing computer chips; Cadence's core business.
SDA — System Design & Analysis: Cadence's reporting segment for everything beyond chips (PCB, packaging, CFD, and now the acquired CAE stack).
Reinforcement learning (RL) — an AI technique where an agent learns by trial, reward and penalty; the mechanism inside Cerebrus.
Agentic AI — AI agents (usually LLM-based) that plan and execute multi-step tasks by driving other software tools; ChipStack and Cerebrus AI Studio.
JedAI — Cadence's "Joint Enterprise Data and AI" platform: the shared data layer under its chip-side AI products.
Reduced-order model (ROM) — a small, fast model (often ML-trained) that approximates an expensive simulation; ODYSSEE's core technique.
Uncertainty quantification (UQ) — computing how scatter in inputs (material batches, process settings) propagates to scatter in outcomes.
Digimat — material-modelling software bridging injection-moulding simulation (fibre orientation) and structural simulation.
ANSA / META — BETA CAE's pre-/post-processor pair, an automotive CAE de-facto standard.
Mould-flow solver — the physics engine simulating molten plastic filling a mould (fill/pack/warp/cool). Cadence has none.
Physical AI — Cadence's marketing banner for AI + simulation aimed at robotics and electromechanical systems.

Appendix C — Notes (thinnest-evidence areas to revisit)¶

RESOLVED 2026-08-02: Cerebrus AI Studio launched May 7, 2025 — confirmed by multiple dated trade reports of the launch announcement (SemiIP Hub, Digital Engineering 24/7, ELE Times; Cadence's own blog post is 403-blocked to fetchers). The Oct 20, 2025 X post was later promotion, not the launch. §5 updated.
Digimat 2026.1 — PARTIALLY RESOLVED 2026-08-02: released April 2026 (indexed Apr 15, 2026), Cadence-branded ("Cadence announced the availability of Digimat 2026.1"); Hexagon's Nexus Digimat feed stops at 2025.1 (Jul 2025), consistent with the Feb 2026 handover. Day-precision is medium confidence (Salesforce-hosted release note fails to render; date from search-index snippets). Highlights: Digimat-MS key-results/gate-position workflows, Digimat-MX curing calibration, new DoE workflow — classic optimization, no new AI/ML features.
"Fully Autonomous Virtual Engineer" (~May 2026) — verified as a headline in search results only; the release was not fetched.
Employee count and exact founding details — not verified from primary sources; deliberately kept vague in §1.
BETA CAE "top 10 OEMs" — primary-source but promotional; no OEM individually named; do not convert into named-customer claims.
Whether any Digimat/ODYSSEE integration onto JedAI or Millennium is underway — nothing public either way; the report's central open question.
NVIDIA "investment" in Cadence — not verified; the documented relationship is technology partnership plus NVIDIA's stated purchase of 10 Millennium systems.

Executive Summary¶

Who they are. Cadence is one of the big three of EDA — electronic design automation, the software used to design computer chips (NASDAQ: CDNS; CEO Anirudh Devgan). Fiscal-2025 revenue was $5.30 billion, up 14%, at roughly 45% non-GAAP operating margin; after its latest acquisition it raised 2026 guidance to about $6.3 billion. [Evidence] ¹²
Why they are in this study at all. On 23 February 2026 Cadence completed the €2.7 billion purchase of Hexagon's Design & Engineering business — MSC Nastran, Adams, Marc, and with them Digimat (material modelling for fibre-reinforced plastics) and ODYSSEE (a genuine machine-learning surrogate platform). That is precisely the "layer around injection moulding" this study documented in the Hexagon report: Digimat consumes fibre-orientation results from Moldflow or Moldex3D, and ODYSSEE's Quasar engine reached the reinforced-plastics workflow via a Digimat-RP 2023.2 plugin. Cadence is therefore the study's newest platform vendor by acquisition — it bought its relevance rather than building it. [Evidence] ⁴²⁸
The honest AI finding — unusual for this study, mostly positive. Cadence has arguably the deepest genuine-AI engineering bench of any vendor covered here: reinforcement-learning chip optimization shipped in production since July 2021 (Cerebrus), ML-driven verification since 2022 (Verisium, on the JedAI data platform), an agentic multi-block design platform in 2025 (Cerebrus AI Studio) and an agentic "Super Agent" in February 2026 (ChipStack), already in early deployment at NVIDIA, Altera and Tenstorrent. This is AI that passes the study's honest test — it learns from data — and it has paying customers. All of it is chip design. None of it touches moulding. [Evidence] ⁸⁹¹³
The moulding reality. Cadence has no injection-moulding solver — no fill, pack, warp or cooling simulation, and none was acquired. Like Hexagon before it, it owns the layer around moulding: Digimat (materials), ODYSSEE (ML surrogates), plus BETA CAE's ANSA (the automotive CAE pre-processor, acquired 2024 for $1.24 billion) and the MSC structural solvers. No public statement by Cadence about a moulding or plastics roadmap was found, and no documented application of Cadence's own AI to moulding exists. The nearest thing remains the inherited Digimat line, which continued shipping under the new owner (Digimat 2026.1: enhanced injection-gate-position and uncertainty-quantification workflows). [Evidence of absence] [Evidence] ²²
Net. For moulded parts, Cadence today is potential, not product: a company with proven production AI, a GPU supercomputer line built with NVIDIA, a "Physical AI" banner over the whole expansion — and a moulding-adjacent inheritance it has not yet said anything specific about. Its trajectory matters more than its present. [Synthesis]

Cadence Q4/FY2025 results, Feb 17 2026 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr-ir/2026/cadence-reports-fourth-quarter-and-fiscal-year-2025-financial.html; FY revenue also per 10-K note — https://www.stocktitan.net/sec-filings/CDNS/10-k-cadence-design-systems-inc-files-annual-report-6bcef311af10.html ↩↩↩↩
Cadence Q2 2026 results, Jul 26/27 2026 — https://www.morningstar.com/news/business-wire/20260726697902/cadence-reports-second-quarter-2026-financial-results; analysis — https://futurumgroup.com/insights/cadence-q2-fy-2026-earnings-climb-on-agentic-ai-and-record-backlog/ ↩↩↩↩
Cadence to Acquire Hexagon's D&E Business ("…Accelerating Expansion in Physical AI…"), Sep 2 2025 — https://www.businesswire.com/news/home/20250902498199/en/Cadence-to-Acquire-Hexagons-Design-Engineering-Business-Accelerating-Expansion-in-Physical-AI-and-System-Design-and-Analysis ↩↩↩↩↩↩↩↩
Cadence Completes Acquisition of Hexagon's D&E Business ("…Advancing Leadership in Physical AI and Multiphysics"), Feb 23 2026 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2026/cadence-completes-acquisition-of-hexagons-design-and-engineering.html ↩↩↩↩↩↩↩↩↩↩↩
engineering.com on the completed deal (Devgan quote), Feb 24 2026 — https://www.engineering.com/cadence-closes-on-nastran-adams-and-more-from-hexagon/ ↩
Cadence to Acquire BETA CAE ($1.24bn; "top 10 global automobile manufacturers"), Mar 2024 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2024/cadence-to-acquire-beta-cae-expanding-into-structural-analysis.html ↩↩↩↩↩↩↩
Cadence Completes Acquisition of BETA CAE, Jun 2024 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2024/cadence-completes-acquisition-of-beta-cae.html; date corroboration — https://www.zacks.com/stock/news/2283843/cadence-cdns-acquires-switzerland-based-beta-cae-for-124b ↩↩
Cerebrus RL-based chip optimization ships, Jul 22 2021 — https://venturebeat.com/ai/cadence-design-systems-launches-cerebrus-machine-learning-for-chip-design; product page — https://www.cadence.com/en_US/home/tools/digital-design-and-signoff/soc-implementation-and-floorplanning/cerebrus-intelligent-chip-explorer.html ↩↩↩↩↩↩↩
Verisium AI-driven verification + JedAI platform, Sep 13 2022 — https://www.businesswire.com/news/home/20220913005378/en/Cadence-Revolutionizes-Verification-Productivity-with-the-Verisium-AI-Driven-Verification-Platform ↩↩↩↩↩↩↩
Optimality Intelligent System Explorer, Jun 8 2022 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2022/cadence-revolutionizes-system-design-with-optimality-explorer.html ↩↩↩
Cerebrus AI Studio (agentic, multi-block) — https://www.cadence.com/en_US/home/tools/digital-design-and-signoff/soc-implementation-and-floorplanning/cadence-cerebrus-ai-studio.html; Cadence X post, Oct 20 2025 — https://x.com/Cadence/status/1980302415669744099 ↩↩↩↩↩
Launch date May 7, 2025 per dated trade coverage (verified 2026-08-02) — SemiIP Hub — https://semiiphub.com/news/cadence-cerebrus-ai-studio; Digital Engineering 24/7 — https://www.digitalengineering247.com/article/introducing-cadence-cerebrus-ai-studio ↩↩
ChipStack AI Super Agent (Nov 2025 acquisition; launch Feb 10 2026; early users NVIDIA/Altera/Tenstorrent) — https://siliconangle.com/2026/02/10/cadence-announces-chipstack-super-agent-system-chip-design-verification/; Cadence PR — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2026/cadence-unleashes-chipstack-ai-super-agent-pioneering-a-new.html ↩↩↩↩↩↩↩↩↩↩↩↩↩↩
Millennium M1 unveiling, Feb 1 2024 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2024/cadence-unveils-millennium-platformindustrys-first-accelerated.html ↩↩↩
Millennium M2000 with NVIDIA Blackwell, May 7 2025 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2025/cadence-unveils-millennium-m2000-supercomputer-with-nvidia.html ↩↩↩↩↩↩↩
NVIDIA blog (Huang: "NVIDIA plans to purchase 10 Millennium Supercomputer systems…"), May 7 2025 — https://blogs.nvidia.com/blog/cadence-millennium-nvidia-blackwell/ ↩↩↩↩↩↩↩↩
Cadence completes NUMECA acquisition, 2021 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2021/cadence-completes-acquisition-of-numeca.html ↩↩
Cadence acquires Pointwise (NUMECA users NASA/Ariane/Honda/Ford noted in release context), Apr 15 2021 — https://www.cadence.com/en_US/home/company/newsroom/press-releases/pr/2021/cadence-acquires-pointwise-to-expand-system-analysis-offerings-a.html ↩↩↩
Future Facilities acquired by Cadence, 2022 — https://www.datacenterdynamics.com/en/news/data-center-digital-twin-firm-future-facilities-acquired-by-cadence-design-systems/ ↩
ODYSSEE product page ("AI Simulation & Digital Twin Software"), Cadence — https://www.cadence.com/en_US/home/tools/msc-software/odyssee.html ↩↩↩
Digimat product page, Cadence — https://www.cadence.com/en_US/home/tools/msc-software/digimat.html ↩
What's New in Digimat 2026.1 (released April 2026, Cadence-branded; gate-position & UQ workflow enhancements, curing calibration, DoE import from any supported injection-moulding software; JS-gated portal, date from search-index snippets — see App. C) — https://mscsoftware.my.site.com/customers/s/article/What-s-New-in-Digimat-2026-1 ↩↩↩↩↩
ODYSSEE brochure (Hexagon MI, © 2023; NAFEMS-hosted; documents Quasar and the Digimat-RP 2023.2 UQ plugin lineage) — https://www.nafems.org/downloads/dropbox/nologin/nrc24_dach/hexagon/hexagon-mi_odyssee_brochure.pdf ↩↩↩
Hexagon Digital Materials webinar series (GM, Schaeffler, DOMO Chemicals, Radici Group; Hexagon-era) — https://go.mi.hexagon.com/Hexagon-Multiscale-I-Digital-Materials-Webinar-Series-2025 ↩↩↩
SIMTEQ Engineering (partner) — ML/AI for CAE, incl. PSA Group ODYSSEE ROM example — https://simteq.co.za/blog/machine-learning-and-artificial-intelligence-ai-for-cae/ ↩↩↩
"Cadence Unveils Industry's First Fully Autonomous Virtual Engineer for Chip Design, powered by NVIDIA," ~May 2026 (headline verified only) — https://www.businesswire.com/news/home/20260531072918/en/Cadence-Unveils-Industrys-First-Fully-Autonomous-Virtual-Engineer-for-Chip-Design-powered-by-NVIDIA ↩
Cadence-Design-Systems on GitHub (no public repositories) — https://github.com/Cadence-Design-Systems ↩↩
This study's Hexagon report — full pre-acquisition history of Digimat, ODYSSEE and the Digimat-RP 2023.2 Quasar UQ plugin: 1_07_hexagon.md ↩↩↩↩↩
What "optimization technology" means and why it is not generative AI — see the concept note What is Optimization technology: topology optimization and design-of-experiments compute from physics equations and search; they learn nothing from data, so they are classical optimization, not AI. ↩↩↩↩↩↩↩