Autodesk, Inc.¶

Company Overview¶

Business Overview¶

Autodesk was founded in 1982 and is headquartered in San Francisco (NASDAQ: ADSK). Its fiscal year ends January 31, so "fiscal 2026" is roughly calendar 2025. Andrew Anagnost has been CEO since 2017. The company sells by subscription across three industry groups: AECO (architecture, engineering, construction, operations), Design & Manufacturing, and Media & Entertainment. [Evidence + Synthesis] ³

Key financials:

Fiscal 2026 (reported February 26, 2026): revenue $7.2 billion, up 18% as reported; Q4 revenue $1.96 billion, up 19%; billings $7.8 billion, up 30%; non-GAAP operating margin 38%; free cash flow $2.4 billion, up 54%. [Evidence] ³
Fiscal 2025 (reported February 27, 2025): revenue $6.13 billion, up 12%; recurring revenue 97% of total; Design revenue $5.10 billion. [Evidence] ⁴

The fiscal-2026 jump partly reflects a billings/transition model change and strong AECO performance — treat the 18% as reported, not as pure organic demand. [Inference] ³

Automotive Business¶

Autodesk's automotive presence is real but narrower than Dassault's or Siemens'. It centres on the design studio, not the vehicle program: [Synthesis] ³²

Alias — sketching, concept modelling, Class-A surfacing and visualization; a long-standing standard in automotive styling studios. [Evidence] ³³
VRED — automotive 3D visualization for design reviews, virtual prototypes and perceived-quality work. [Evidence] ³⁴
Moldflow — used throughout automotive plastics supply chains (interior trim, lighting, mirrors, tanks) to validate parts and moulds before tooling. [Evidence — partner-reported for named users] ³⁰³¹
BlankAI / Form Explorer — generative AI for automotive concept design (see §5). [Evidence] ¹⁹²⁰

Autodesk does not publish automotive revenue, and no current automotive OEM vehicle-program PLM deployment on Autodesk tools was found. The verified OEM touchpoints are project-level: the General Motors generative-design partnership (2018) is the strongest primary-source OEM story. [Evidence of absence + Evidence] ²⁹

Engineering Business¶

For product engineering, the flagship is Fusion (formerly Fusion 360) — a cloud-native CAD/CAM/CAE/PCB platform aimed at mid-market manufacturing — plus Inventor (established mechanical CAD), Vault/Fusion Manage (data and PLM, mid-market), and the simulation portfolio in which Moldflow sits. [Synthesis]

Role in Automotive Moulded Parts Workflow¶

Workflow Stage	Relevant?	Notes (product)
Requirements	Weak	No serious requirements/PLM story at automotive-program scale
Industrial Design	Strong	Alias (Class-A studio surfacing), VRED; Form Explorer AI concepts
Concept Design	Strong	Alias, Fusion; BlankAI-derived generative concepts
CAD Modelling	Yes	Fusion, Inventor (mid-market; not the automotive OEM CAD standard)
Engineering Review	Partial	Fusion cloud collaboration; VRED design reviews
Simulation	Yes	Fusion Simulation, Inventor Nastran; Moldflow for plastics
DFM	Partial — rules/physics based	Fusion draft analysis + moulding checks; Moldflow Adviser guidance
Tool Design	Partial	Inventor/Fusion mould tooling; weaker than CATIA/NX dedicated apps
Mould Flow	Core strength — owns the category	Moldflow Insight/Adviser; also embedded in Fusion (part-only)
Prototype	Yes	Fusion CAM/additive
Validation	Yes	Moldflow solver validation program; measured-vs-simulated warp tools
Manufacturing Engineering	Partial	Fusion CAM; no DELMIA-class digital factory
Production Release	Partial	Vault / Fusion Manage (mid-market PLM)

What makes Autodesk structurally different: it owns the single most on-topic product (Moldflow) but not the surrounding automotive program backbone. Dassault owns the whole chain and no category-leading moulding brand; Autodesk owns the category-leading moulding brand and only parts of the chain. [Synthesis]

AI Strategy¶

Public AI Vision¶

Autodesk branded its AI push as "Autodesk AI" at Autodesk University on November 13, 2023 — "intelligent assistance and generative capabilities" native to its Design and Make Platform. [Evidence + Marketing] ¹³

Three strands are visible underneath the branding: [Synthesis]

Assistants and agents — the Autodesk Assistant, rolled out across products (AutoCAD, Revit, Civil 3D from 2023; Fusion from January 2025; Moldflow from the 2027 release) and described by 2025 as agentic — able to execute commands, not just answer questions. [Evidence] ¹⁵¹⁴⁸
Learned models for design tasks — transformer-based AutoConstrain, AI-assisted automated drawings, BlankAI's GAN-based concept generation, and "proprietary foundation models for manufacturing" announced September 2025. [Evidence] ¹⁵¹⁹¹⁴
Open research — the Autodesk AI Lab publishes at CVPR/ICCV and releases code and datasets (UV-Net, BRepNet, Fusion 360 Gallery); Project Bernini (2024) is its public generative-3D research flagship. [Evidence] ²⁵¹⁸

Investor Statements¶

CEO Andrew Anagnost has framed the position as: industrial AI needs "data, context and domain expertise," with Autodesk's advantage being probabilistic generation combined with deterministic engineering validation; the Assistant and MCP infrastructure act as a "harness layer" that makes frontier models controllable. (Q3 fiscal-2026 call, November 25, 2025 — trade-transcript sourced.) [Evidence — transcript] ³⁵
The September 2025 Fusion announcement disclosed a Microsoft collaboration (Azure OpenAI Service, GPT-image-1) for design visualization — i.e., like Dassault with Mistral, the frontier LLM layer is partnered, not owned, though Autodesk does claim its own manufacturing foundation models. [Evidence] ¹⁴

Engineering AI Strategy¶

For engineering specifically, the pattern is: ship narrow, well-scoped learned features inside existing workflows (constraints, drawings, toolpaths), put an LLM assistant everywhere, and keep the ambitious generative-geometry work (Bernini, foundation models) in research or preview until it is reliable. This is more incremental and more honest-by-construction than a "companions will co-engineer" pitch — but it also means nothing learned has reached the physics of moulding yet. [Synthesis + Inference]

Timeline of AI Evolution¶

Date	Event
1978	Moldflow founded in Melbourne, Australia (Colin Austin)⁶
Jun 25, 2008	Autodesk completes Moldflow acquisition (~$297M)⁵
2018	Generative design ships in Fusion 360; GM seat-bracket partnership²⁹
Nov 2, 2023	BlankAI acquisition surfaces — automotive generative concept design¹⁹
Nov 13, 2023	"Autodesk AI" branding launched at AU 2023¹³
Jan 2024	AI-enhanced Automated Drawings in Fusion¹⁵
May 2024	Project Bernini — generative 3D research (not a product)¹⁸
Oct 2024	AU 2024: AutoConstrain + Fusion Assistant announced¹⁵
Jan 2025	AutoConstrain and Autodesk Assistant ship in Fusion¹⁵
Apr 2025	Moldflow 2026 release; Moldflow R&D shows NN-surrogate research¹⁰⁷
Sep 16, 2025	Fusion: manufacturing foundation models; Microsoft/Azure OpenAI collab¹⁴
Nov 25, 2025	Q3 FY26 call: agentic "harness layer" positioning³⁵
Mar 27, 2026	Moldflow 2027 released ("Moldflow 2027 is live!", Autodesk forum): Autodesk Assistant in Moldflow (Tech Preview) + `aiassistant` data utility⁸⁹¹

Products Relevant to Engineering¶

Moldflow (Insight / Adviser) — the injection-moulding simulator. Recent releases: Moldflow 2026 (April 2025) made STAMP the default 3D shrinkage model, added RTM mould-thermal analysis, Digimat data exchange, and a warpage result designed for comparison against 3D-scanned real parts.¹¹ Moldflow 2027 (2026) added GPU-accelerated graphics, corner-effect warp accuracy, cooling-channel optimization for drill-and-plug moulds (Tech Preview), and — the AI first — the Autodesk Assistant (Tech Preview) plus the aiassistant utility that exports study data as JSON "to help AI tools work with Moldflow data." AI: an LLM assistant beside a classical solver — see §5. [Evidence] ⁷⁸⁹

Fusion — cloud CAD/CAM/CAE. AI: AutoConstrain (transformer ML), automated drawings, Assistant with toolpath generation, September 2025 generative "editable geometry" foundation models. Moulded parts: Injection Molding Simulation inside the Simulation Extension — explicitly powered by Moldflow solvers, cloud-solved, and limited to part-only analysis (no runners, cooling channels or mould). Draft analysis lives in the Product Design Extension. Note: there is no "Molding Extension" — the moulding capability sits inside the Simulation Extension. [Evidence] ²³²⁴¹⁴

Fusion Generative Design — "outcomes"-based shape generation for weight and stiffness. This is topology optimization plus cloud-scale option generation — optimization technology wearing an AI label³⁸; Autodesk itself has published a blog titled "Topology Optimization is not Generative Design," arguing the difference is breadth of exploration, not learning. [Evidence + Marketing] ²¹²²

Alias / VRED — automotive studio surfacing and visualization; Alias now hosts Form Explorer, the productized descendant of BlankAI. [Evidence] ³³²⁰

Inventor / Vault / Fusion Manage — mechanical CAD and mid-market data management; Assistant reaches Vault with natural-language search (September 2025). [Evidence] ¹⁴

AI Capabilities¶

1. Autodesk Assistant in Moldflow (Tech Preview, Moldflow 2027, 2026). An "intelligent AI partner" that "answers questions, helps interpret results, and performs a limited set of tasks" inside Moldflow [Evidence — indexed, unfetched].¹⁷ Enabled by default; agentic actions can be opted out; availability varies by country and language. Supported by the aiassistant command-line utility, which extracts geometry, material, machine and result summaries from Moldflow's binary studies into JSON for AI consumption. Why it matters: this is the first AI shipped in a mould-flow product by any of the generalist CAD/PLM platform vendors. Limitation: it is a language-model layer for Q&A and result interpretation — the fill/pack/warp physics it explains is still computed classically; it predicts nothing itself. [Evidence] ⁸⁹

2. Moldflow neural-network surrogate — research only. Autodesk's Moldflow R&D update (Dr. Franco Costa, April 2025) shows a multi-layer perceptron (Python scikit-learn v1.5.0, two hidden layers of 470 and 255 neurons) trained on a 3×3 grid of full simulations to predict deflection magnitude and variance across a 9×9 process window (melt temperature × flow rate), for one geometry and one material — motivated by recycled-material property variation. The deck's own caveat: "training data should be closely related to the problem to be solved." Limitation: a two-input demo on one part, presented under a research disclaimer — not a product, and far from a general fill/warp surrogate. [Evidence] ¹⁰

3. Moldflow Cooling Channel / Circuit Optimization. Automatic generation and iterative improvement of cooling layouts (42% combined-metric improvement, 13.4°C average mould-temperature reduction in the shown example); extended to drill-and-plug moulds in 2027 (Tech Preview). This is classical optimization — simulate, score, move channels, repeat — not machine learning³⁸. [Evidence + Marketing] ¹⁰⁸

4. STAMP shrinkage model (default in Moldflow 2026). "Shrinkage Test Adjusted Mechanical Properties" — uses measured shrinkage data to calibrate anisotropic thermal-expansion and modulus values, with markedly better measured-vs-predicted scatter across 277 polymers. Honest classification: data-calibrated physics, not machine learning — the answer still comes from equations; the data tunes their coefficients. It is patent-tracked (US application 17/959,221), not AI-branded, to Autodesk's credit. [Evidence] ⁷¹⁰

5. AutoConstrain (Fusion, shipped January 2025). A transformer-based ML model that predicts and applies sketch constraints; grew out of published Autodesk Research work (SketchGraphs-style learning). One of the few shipped, architecture-disclosed ML features in mainstream CAD. [Evidence] ¹⁵¹⁶

6. Automated Drawings (Fusion, AI-enhanced since January 2024). Generates 2D drawing views and dimensions from 3D models and "intelligently decides" what to include. [Evidence] ¹⁵

7. BlankAI / Form Explorer (automotive concept design). Acquired 2023. A generative model (custom GAN architecture, "XDGAN") trained on a studio's historical 3D designs, generating editable vehicle exterior concepts in near-real-time from semantic and natural-language controls; productized as Form Explorer for Alias. Limitation: styling volumes only — no engineering, no interiors, no manufacturability. But it passes the honest AI test cleanly: it learns from data. [Evidence] ¹⁹²⁰

8. Project Bernini (research, May 2024). Generative-3D model trained on ~10 million shapes; text/image/sketch/point-cloud input; generates shape and texture separately. Explicitly experimental, not available for use. [Evidence] ¹⁸

9. Fusion generative design. Topology optimization with cloud-parallel outcome generation — see §4 and the concept note³⁸. [Evidence + Marketing] ²¹

10. DFM for moulding = rules and physics, not AI. Fusion's moulding-related checks (draft analysis in the Product Design Extension; part-only injection-moulding simulation with "guided results"; Moldflow Adviser's guidance) are deterministic rules plus classical simulation. No learned DFM checker ships anywhere in the portfolio. [Evidence of absence] ²³²⁴

Engineering Workflow Contribution¶

An automotive moulded part can touch Autodesk tools like this:

concept and Class-A surfacing in Alias (with Form Explorer AI concepts) → visualization/review in VRED → detailed CAD in Fusion/Inventor → part-only moulding sanity check in Fusion (Moldflow solvers) → full mould-flow, tooling and process validation in Moldflow Insight → CAM in Fusion → data in Vault.

But unlike Dassault, this chain is not one platform: the studio tools, Fusion, and Moldflow are separate products with separate data models, and the program-level PLM backbone typically belongs to another vendor (Teamcenter, ENOVIA, Windchill). The Assistant layer is being stretched across products (Fusion, Vault, Moldflow) as the connective AI tissue — in preview. [Synthesis] ¹⁴⁸

Public Customer Evidence¶

Case Studies¶

General Motors (May 2018) — generative design in Fusion 360 for a seat bracket: >150 design options; the chosen one-piece design was 40% lighter and 20% stronger than the eight-piece original; framed as the start of a multi-year partnership. Primary Autodesk source. Note: this is topology-optimization generative design³⁸, and additive-manufacturing oriented — not moulded plastics, and not learned AI. [Evidence + Marketing] ²⁹
Varroc Polymers (India) — automotive plastics maker (headlamp covers, mudguards); used Moldflow with partner Neilsoft to simulate rapid heat-and-cool moulding and catch part issues before tooling. Partner success story hosted on autodesk.com; no quantified outcome, and not AI. [Evidence — partner-reported] ³⁰
FEKA Automotive (Turkey) — Tier-1/OEM supplier (mirrors, lighting, tanks, trim); uses Moldflow Insight to eliminate moulding problems before tooling; reports shortened project completion. Partner success story; unquantified, not AI. [Evidence — partner-reported] ³¹
Synventive Molding Solutions — supplied the physical moulding case study that validated Moldflow 2025's variable-speed valve-gate simulation (correct weld-line position vs. photographed parts). Validation evidence, not a deployment story. [Evidence] ¹⁰

Conference Demonstrations¶

Autodesk University 2023 (Autodesk AI launch, BlankAI car-concept demos) and AU 2024 (AutoConstrain, Fusion Assistant); Moldflow's annual Advanced Manufacturing Summit decks (2025) are unusually candid technical disclosures. [Evidence] ¹³¹⁵¹⁰

White Papers / Testimonials¶

Moldflow publishes an annual solver validation report comparing releases against measured mouldings — a scientific-culture artifact worth noting. [Evidence] ¹²

The gap. As with Dassault, a public automotive story combining Autodesk AI + moulded plastic parts + a quantified outcome does not exist as of August 2026. The AI stories are styling (BlankAI) or general CAD; the moulding stories are classical simulation. [Evidence of absence]

Technical Architecture (Inferred)¶

Evidence¶

Fusion's injection-moulding simulation runs Moldflow solvers in Autodesk's cloud. [Evidence] ²⁴
The Assistant layer spans products; September 2025 disclosed Azure OpenAI Service integration for image generation and "proprietary foundation models for manufacturing" for editable geometry. [Evidence] ¹⁴
Moldflow 2027 ships a JSON export bridge (aiassistant) between solver studies and AI tooling — a deliberate data-plumbing move. [Evidence] ⁹
The AI Lab's stack is public: B-rep graph networks (UV-Net/BRepNet), large shape-generation models (Bernini), CAD-history datasets (Fusion 360 Gallery). [Evidence] ²⁵¹⁸

Synthesis¶

Bottom to top: Autodesk cloud (Fusion platform + simulation compute) → partnered frontier LLMs (Microsoft/Azure OpenAI) plus in-house manufacturing foundation models → per-product Assistants with tool-calling ("harness layer," MCP) → narrow shipped ML features (constraints, drawings) at the edges. Moldflow connects to this stack today only through the Assistant and the JSON bridge — its solvers are untouched by ML. [Synthesis]

Inference¶

The aiassistant utility looks like groundwork: exposing solver results as structured data is precisely what an agent (or a future surrogate-training pipeline) needs. The R&D surrogate slides suggest Moldflow's team sees ML as a process-window-exploration accelerator, not a solver replacement. Expect Assistant-driven result interpretation to be productized before any learned physics is. [Inference] ⁹¹⁰

AI Technologies¶

In one list: LLM assistants (partnered frontier models + Autodesk harness; agentic tool-calling); transformer ML for sketch constraints; GAN-based generative styling (BlankAI/XDGAN); large generative-3D research models (Bernini, ~10M training shapes); B-rep graph neural networks (UV-Net, BRepNet — research/open source); topology optimization marketed as generative design³⁸; classical optimization for cooling layouts; data-calibrated physics (STAMP); and one small-scale NN surrogate research demo in moulding. [Evidence + Synthesis]

Research Publications¶

Papers¶

UV-Net: Learning from Boundary Representations — CVPR 2021. Graph neural network over B-rep face-adjacency with UV-grid features; classification and segmentation on public datasets. [Evidence] ²⁵
BRepNet — topological message passing directly on B-rep coedges; reported 92.5% accuracy on modelling-operation segmentation. [Evidence] ²⁶
Fusion 360 Gallery — dataset + environment for programmatic CAD reconstruction from human design sequences (~8,600 sequences from ~20,000 designs). [Evidence] ²⁶
Project Bernini (2024) — generative 3D from multimodal input. [Evidence] ¹⁸
Third-party researchers have built ML surrogates on top of Moldflow outputs (e.g., arXiv:2107.14574, fill-time and deflection prediction from Moldflow projects) — evidence of Moldflow as a research substrate, but not Autodesk work. [Evidence] ²⁸

Patents¶

US patent application 17/959,221 — the STAMP shrinkage-calibration method disclosed in the Moldflow R&D deck. [Evidence] ¹⁰

Notes¶

The research bench directly relevant to this study's retrieval thesis (B-rep GNN embeddings for searching part libraries) is Autodesk's, and it is open. The research bench relevant to moulding physics ML is thin and early. [Synthesis]

Open Source¶

Substantial for AI — the sharpest contrast with Dassault. [Evidence] ²⁷

The AutodeskAILab GitHub organisation publishes the code of its flagship CAD-ML papers: UV-Net (MIT license), UVStyle-Net (ICCV 2021), BRepNet, and the Fusion 360 Gallery Dataset (data under a non-commercial research license). [Evidence] ²⁵²⁶
These artifacts are foundational to the wider field: UV-Net and the Gallery dataset are the standard starting points for anyone building B-rep learning or CAD-retrieval systems. [Synthesis]
The Dassault Systèmes report documented Dassault's near-nil open-source AI footprint (4 graphics repos, no models, no datasets). Autodesk is the opposite pole among platform vendors: it publishes the papers, the code, and the training data. The frontier LLMs are still partnered (Microsoft), but the CAD-domain ML is genuinely open. [Evidence + Synthesis] ²⁷
Caveat: openness stops at research. Shipped features (AutoConstrain, Assistant, Form Explorer) and their trained weights are proprietary, and the Gallery data licence excludes commercial use. [Evidence] ²⁶

Engineering Service / Platform Mapping¶

Fusion (cloud CAD/CAM/CAE + Simulation Extension) · Inventor (mechanical CAD) · Moldflow Insight/Adviser (moulding simulation) · Alias + VRED (automotive studio) · Vault / Fusion Manage (data/PLM, mid-market) · Autodesk Platform Services (APIs) · Autodesk Assistant (cross-product AI layer, preview). Interoperates with — rather than replaces — CATIA/NX/Teamcenter in automotive programs; Moldflow imports neutral geometry from any CAD, which is exactly why it is ubiquitous in supplier chains. [Evidence + Synthesis] ²³⁷

Engineering Intelligence Stack Mapping¶

Intent — weak at program level (no requirements backbone); emerging at task level via Assistant prompts ("generate a toolpath," "explain this warp result"). [Synthesis]
Knowledge — Moldflow's material database and 45+ years of moulding know-how are a deep, proprietary knowledge asset; the aiassistant JSON bridge is the first step toward making it AI-legible. The Fusion 360 Gallery shows Autodesk knows how to turn design history into training data. [Evidence + Inference] ⁹
Reasoning — classical solvers and optimizers for moulding; learned reasoning only in CAD authoring (constraints, drawings) and research (surrogate demo). [Evidence] ¹⁰
Execution — Fusion Assistant executes commands and toolpaths; Moldflow's Assistant "performs a limited set of tasks" (Tech Preview). [Evidence] ⁸
Feedback — Moldflow 2026's scan-comparison warp result and the solver validation program close the sim-vs-reality loop manually; there is no learning loop from production outcomes (defects, scrap) anywhere in the product line. [Evidence + Evidence of absence] ⁷¹²

Strengths¶

Owns Moldflow — the reference injection-moulding simulator, with a live R&D program (STAMP, corner effects, cooling optimization) and annual public validation. [Evidence] ⁵¹²
First of the generalist CAD/PLM platform vendors to ship AI in a mould-flow product (Assistant Tech Preview, 2026) plus deliberate AI data-plumbing (aiassistant). [Evidence] ⁸
Genuinely open AI research — UV-Net, BRepNet, Fusion 360 Gallery; unmatched among platform vendors. [Evidence] ²⁷
Real shipped ML in CAD (AutoConstrain, automated drawings) and real learned generative AI in automotive styling (Form Explorer). [Evidence] ¹⁵²⁰
Strong automotive studio franchise (Alias, VRED). [Evidence] ³³

Weaknesses¶

The AI in Moldflow explains; it does not predict. No shipped ML surrogate, no learned defect/warp prediction, no AI DFM — the physics is classical and the Assistant is a preview chatbot layer. [Evidence] ⁸¹⁰
Competitors are ahead on learned or AI-branded moulding physics: Simcon (with startup Emmi AI) markets an "AI Solver" for Cadmould claiming ~1,000× faster design exploration, and Moldex3D 2025 ships an AI Optimization Wizard. Moldflow, the market leader, has shipped neither. [Evidence] ³⁶³⁷
No automotive program backbone — PLM/requirements/Class-A production workflows belong to Dassault/Siemens; Autodesk's OEM evidence is project-level. [Synthesis] ²⁹
Fragmented portfolio — Alias/VRED, Fusion, Inventor and Moldflow are separate products, not one platform; the Assistant-as-glue story is still preview. [Synthesis]
Frontier LLMs are partnered (Microsoft), not owned; named-country and language limits already apply to the Moldflow Assistant. [Evidence] ¹⁴⁸

Current Gaps (largely manual today)¶

Interpreting Moldflow results still requires an expert — the Assistant helps explain, but the judgment loop (is this weld line acceptable? which gate moves?) remains human. Choosing process windows is manual or brute-force DoE; the NN surrogate that would accelerate it is a research demo. There is no product that learns from a moulder's historical defect/scrap records, and no geometry-retrieval product over a company's own part library — despite Autodesk owning the best open-source building blocks for exactly that. [Evidence + Inference] ¹⁰²⁵

Future Direction¶

Assistant everywhere, deepening from Q&A to agency — Moldflow's Tech Preview is explicitly the "first agentic system implementation" with basic action and knowledge-access pieces; expect result-interpretation and setup automation next. [Evidence + Inference] ⁸
Manufacturing foundation models (September 2025) point toward generative, editable geometry inside Fusion. [Evidence] ¹⁴
Moulding ML stays research-paced: the R&D deck's surrogate work, recycled- material variation studies, and automatic conformal cooling generation are the visible pipeline — all pre-product, all under a research disclaimer. [Evidence] ¹⁰
My read: Autodesk will productize AI around Moldflow (assistants, data bridges, optimization wrappers) well before it puts learning inside the solver — and competitive pressure from Cadmould/Emmi AI and Moldex3D is the most likely accelerant. [Inference] ³⁶

Relevance to Automotive Moulded Parts¶

Capability	Strength	Notes
Plastic Part Design	Moderate	Fusion/Inventor plastic features; rules-based, not AI
Surface Design (Class A)	Strong (studio)	Alias is the styling-studio standard; production Class-A often finishes elsewhere
CAD Automation	Strong, AI shipping	AutoConstrain (transformer), automated drawings, Assistant toolpaths
Mould Flow	Strongest — owns the category	Moldflow Insight/Adviser: fill, pack, cool, warp, fibre, over-moulding, RTM; embedded (part-only) in Fusion
DFM	Partial — rules/physics	Draft analysis, guided moulding results; no learned DFM
Tool Design	Partial	Inventor/Fusion tooling; cooling-layout optimization (classical) in Moldflow
Manufacturing Engineering / Quality	Partial	Fusion CAM; scan-vs-sim warp comparison; no production-data learning loop
Engineering Knowledge Reuse	Emerging	`aiassistant` JSON bridge; Assistant knowledge access; open CAD-ML research

Critical finding. Autodesk is the one platform vendor whose AI has touched the injection-moulding domain — but only just: a Tech Preview LLM assistant that answers questions and interprets results (Moldflow 2027, 2026), plus a single-part neural-surrogate research demo (April 2025). Everything that actually computes moulding outcomes — fill, pack, warp, cooling — remains classical physics, classical optimization, and data-calibrated (not data-learned) material models. The honest test — "does the system learn from data, or compute from equations?" — is passed today only by features far from the mould: sketch constraints, drawings, and car-styling concepts. [Evidence — the report's central point]⁸¹⁰

Net. Dassault owns the widest moulded-parts platform with no AI in the moulding domain; Autodesk owns the deepest moulding product with the first — still shallow — AI beside it, and by far the openest AI research bench. [Synthesis]

Key Takeaways¶

Autodesk owns Moldflow, the industry-standard injection-moulding simulator (founded 1978; acquired 2008 for ~$297M). [Evidence] ⁵⁶
Moldflow 2027 (2026) shipped the first AI in the moulding domain from any of the generalist CAD/PLM platform vendors — the Autodesk Assistant, in Tech Preview — but it interprets results; it does not predict them. [Evidence] ⁸
Autodesk's internal moulding-ML is a research demo: a two-input scikit-learn perceptron on one part (April 2025), shown under a research disclaimer. [Evidence] ¹⁰
Moulding competitors (Simcon/Emmi AI, Moldex3D) are shipping learned or AI-branded solvers ahead of the market leader. [Evidence] ³⁶³⁷
Autodesk's genuinely shipped ML lives in Fusion CAD (AutoConstrain, automated drawings) and automotive styling (BlankAI → Form Explorer). [Evidence] ¹⁵²⁰
Fusion generative design is topology optimization, not learned AI — [Evidence + Marketing], see the concept note³⁸. STAMP is data-calibrated physics, honestly not branded as AI. [Evidence] ²¹⁷
Autodesk's open-source AI footprint (UV-Net MIT, BRepNet, Fusion 360 Gallery) is the largest of any platform vendor — the direct opposite of Dassault (see the Dassault Systèmes report). [Evidence] ²⁷
The automotive story is studio tools + supply-chain Moldflow, not program PLM; the strongest OEM evidence (GM, 2018) is generative design for metal, and the moulding customer evidence (Varroc, FEKA) is partner-reported and not AI. [Evidence] ²⁹³⁰³¹
White space: production-grade moulding surrogates, learned DFM, outcome-learning, and part-library retrieval — the last standing on Autodesk's own open research. [Inference]

References¶

Primary Sources — Autodesk¶

FY2026 Q4/full-year results, Feb 26 2026 — https://investors.autodesk.com/news-releases/news-release-details/autodesk-inc-announces-fiscal-2026-fourth-quarter-results
FY2025 Q4/full-year results, Feb 27 2025 — https://investors.autodesk.com/news-releases/news-release-details/autodesk-inc-announces-fiscal-2025-fourth-quarter-and-full-year
Moldflow acquisition completed, Jun 25 2008 — https://investors.autodesk.com/node/13416/pdf
Autodesk AI launch, Nov 13 2023 — https://adsknews.autodesk.com/en/pressrelease/autodesk-university-2023/ and https://investors.autodesk.com/news-releases/news-release-details/introducing-autodesk-ai-design-and-make
Fusion AI investments (foundation models, Microsoft), Sep 16 2025 — https://adsknews.autodesk.com/en/news/new-investments-in-fusion-bring-ai-powered-transformation-to-manufacturing/
Project Bernini, May 2024 — https://adsknews.autodesk.com/en/news/research-project-bernini/ and https://www.research.autodesk.com/projects/project-bernini/
BlankAI introduction (Autodesk Design Studio blog), Nov 2 2023 — https://blogs.autodesk.com/design-studio/2023/11/02/welcome-to-blankai/
Moldflow Insight — What's New in 2026 — https://help.autodesk.com/cloudhelp/2026/ENU/MoldflowInsight-CLC-WhatsNew/files/MFLO-WHATS-NEW-2026-0.html
Moldflow Insight — What's New in 2027 (Autodesk Assistant) — https://help.autodesk.com/cloudhelp/2027/ENU/MoldflowInsight-CLC-WhatsNew/files/MFLO-WHATS-NEW-2027-0.html
Moldflow Insight 2027 — New Features (aiassistant utility) — https://help.autodesk.com/cloudhelp/2027/ENU/MoldflowInsight-CLC-WhatsNew/files/whats-new-2027-0/MFLO-WHATS-NEW-2027-0-FEATURES.html
Moldflow R&D update (Dr. Franco Costa), Apr 2025 (PDF) — https://damassets.autodesk.net/content/dam/autodesk/www/campaigns/advmfg-summit/2025/Moldflow%20Research%20and%20Development%20Updates_Franco%20Costa.pdf
What's new in Moldflow 2026 (Hanno van Raalte, PDF) — https://damassets.autodesk.net/content/dam/autodesk/www/campaigns/advmfg-summit/2025/Whats%20new%20in%20Moldflow%202026%20-%20Summit_2025_Hanno%20van%20Raalte.pdf
Moldflow 2026 solver validation report (PDF) — https://damassets.autodesk.net/content/dam/autodesk/www/pdf/simulation-hub/Autodesk%20Moldflow%202026%20Solver%20Validation%20Report.pdf
Autodesk Assistant page — https://www.autodesk.com/solutions/autodesk-ai/autodesk-assistant
Plastic injection molding solutions — https://www.autodesk.com/solutions/simulation/plastic-injection-molding
Fusion Simulation Extension — https://www.autodesk.com/products/fusion-360/simulation-extension
Fusion generative design for manufacturing — https://www.autodesk.com/solutions/generative-design/manufacturing
"Topology Optimization is not Generative Design" (Fusion blog) — https://www.autodesk.com/products/fusion-360/blog/topology-optimization-is-not-generative-design/
Automotive industry page — https://www.autodesk.com/industry/automotive
Automotive design software (Alias/VRED) — https://www.autodesk.com/solutions/design-manufacturing/automotive-design-software
VRED overview — https://www.autodesk.com/products/vred/overview
AutoConstrain research blog — https://www.research.autodesk.com/blog/using-ai-to-power-autoconstrain-in-fusion-automated-sketching/
GM generative design story — https://www.autodesk.com/customer-stories/general-motors-generative-design and https://adsknews.autodesk.com/en/news/gm-autodesk-using-generative-design-vehicles-future/

Open Source / Research¶

Autodesk AI Lab (GitHub org) — https://github.com/AutodeskAILab
UV-Net (MIT license; CVPR 2021) — https://github.com/AutodeskAILab/UV-Net
Fusion 360 Gallery Dataset (incl. BRepNet) — https://github.com/AutodeskAILab/Fusion360GalleryDataset
Third-party ML surrogate on Moldflow data — https://arxiv.org/abs/2107.14574

Customer Evidence (partner-reported)¶

Varroc Polymers (Neilsoft partner story) — https://www.autodesk.com/support/partners/success-stories/moldflow-for-rapid-heating-and-cooling/8542
FEKA Automotive partner story — https://www.autodesk.com/support/partners/success-stories/feka-automotive-achieves-success-with-the-right-simulation-solutions/8265

Secondary Sources¶

Moldflow — Wikipedia — https://en.wikipedia.org/wiki/Moldflow
Engineering.com, "3 new AI features in Autodesk Fusion," Oct 17 2024 — https://www.engineering.com/3-new-ai-features-in-autodesk-fusion/
DEVELOP3D on Autodesk AI / Form Explorer — https://develop3d.com/cad/autodesk-lays-the-foundations-for-ai/
Motley Fool, Autodesk Q3 FY2026 earnings-call transcript, Nov 25 2025 — https://www.fool.com/earnings/call-transcripts/2025/11/25/autodesk-adsk-q3-2026-earnings-call-transcript/
Plastics Today on Simcon/Emmi AI "AI Solver" — https://www.plasticstoday.com/injection-molding/large-engineering-model-software-slashes-injection-molding-simulation-time
SPE Injection Molding Division on Moldex3D 2025 AI — https://www.injectionmoldingdivision.org/2025/04/22/moldex3d-2025-molding-intelligence-embracing-the-ai-revolution-in-injection-molding/

Appendix A — Timeline¶

1978 Moldflow founded (Melbourne) → Jun 2008 Autodesk acquires Moldflow (~$297M) → 2018 generative design ships; GM partnership → Nov 2023 Autodesk AI branding + BlankAI acquisition → Jan 2024 AI automated drawings → May 2024 Project Bernini (research) → Oct 2024 AU: AutoConstrain + Fusion Assistant announced → Jan 2025 both ship in Fusion → Apr 2025 Moldflow 2026 release; R&D deck shows NN-surrogate research → Sep 2025 manufacturing foundation models + Microsoft/Azure OpenAI → Feb 2026 FY26 results ($7.2B) → ~Mar 27, 2026 Moldflow 2027: Autodesk Assistant Tech Preview + aiassistant JSON bridge.

Appendix B — Glossary¶

Moldflow Insight / Adviser — the advanced (analyst-grade) and simplified (designer-grade) tiers of Autodesk's injection-moulding simulator.
Solver — the numerical engine that computes the physics (how molten plastic fills, packs, cools and warps).
Midplane / Dual Domain / 3D — Moldflow's three meshing approaches for thin-walled parts (shell idealizations) versus full solid meshes.
STAMP — Shrinkage Test Adjusted Mechanical Properties: calibrating material parameters from measured shrinkage tests; data-calibrated physics, not ML.
Surrogate model — a machine-learning model trained on simulation results that then predicts outcomes near-instantly for new inputs.
Autodesk Assistant — Autodesk's cross-product LLM/agentic AI; Tech Preview in Moldflow 2027.
Tech Preview — Autodesk's label for shipped-but-not-final features; may change or be withdrawn.
Agentic AI — an assistant that can execute actions (run commands, generate toolpaths), not just answer questions.
B-rep — boundary representation, the faces/edges data structure of CAD solids; what UV-Net/BRepNet learn from.
Topology optimization — iterative physics-based material removal; the engine behind "generative design" (see concept note).
Class-A surface — the visible, aesthetically controlled outer surface of a vehicle part.
RTM — resin transfer moulding, a composite-moulding process Moldflow also simulates.
DFM — design for manufacturability: checking a design against moulding rules (draft angles, wall thickness, undercuts…).

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

RESOLVED 2026-08-02: Moldflow 2027 released ~March 27, 2026 — the Autodesk Moldflow Insight forum announcement "Moldflow 2027 is live!" is dated Mar 27, 2026 (page 403-blocks fetchers; date consistent across two independent search-index snapshots), corroborated by the live 2027 help pages and the "What's New" deck uploaded May 2026. Cite as "released late March 2026." §3 timeline updated.
The claim that Moldflow 2027's Assistant is Autodesk's "first agentic system implementation" in Moldflow with basic action/knowledge pieces comes via search summaries of community/summit material; the help-page wording ("limited set of tasks," Tech Preview) is the safely citable core.
Whether the Moldflow Assistant runs on Azure OpenAI, Autodesk's own models, or both is not disclosed. Re-checked 2026-08-02, upgraded from "unfetched" to primary-source-read: Autodesk's official AI Transparency Card for "Autodesk Assistant – Tech Preview" (PDF, read in full) states model source "Licensed, pre-trained", primary technique "Agentic AI", and "Prompts shared for model provider training: No" — a third-party provider exists but is never named; no Autodesk, Microsoft or Anthropic source names it. Correct phrasing: "Autodesk rents an unnamed third-party LLM (per its AI Transparency Card)." Consistent with the cross-vendor every-platform-rents-its-LLM finding. Provider identity stays OPEN.²
PARTIALLY RESOLVED 2026-08-02: the Autodesk Assistant transparency card PDF itself was fetched and read (see item 3); the index page (autodesk.com/trust/trusted-ai) remains 403 to fetchers, and other per-feature cards (AutoCAD: "Combination / Transformer"; Construction Cloud: "Licensed / RAG") are known from that sweep — none names a vendor.
Named automotive users of Moldflow (Varroc, FEKA) are partner success stories hosted by Autodesk, undated and unquantified — treat accordingly. No recent OEM-level Moldflow press release was found.
The Simcon/Emmi AI "1,000× faster" and Moldex3D AI-wizard claims are vendor/ trade-press figures used here only as competitive context, not verified.
Form Explorer's shipping status/tier within Alias (which customers can buy it, since when) needs confirmation from an Alias release note.
FY2026's 18% growth includes billings-model effects; check the 10-K before quoting it as organic growth.

Executive Summary¶

Who they are. Autodesk is the world's best-known design-software company (AutoCAD, Revit, Fusion, Inventor, Maya). Revenue in fiscal 2026 (year ended January 31, 2026) was $7.2 billion, up 18%. Manufacturing is one of its three business pillars, alongside architecture/construction and media. [Evidence] ³
Why they matter for moulded parts. Autodesk owns Moldflow — for decades the de-facto industry-standard injection mould-flow simulator, founded in Melbourne in 1978 and acquired by Autodesk in 2008 for about $297 million. No other product in any platform vendor's portfolio is as squarely on this study's topic. [Evidence] ⁵⁶
The central honest finding — a first, but a modest one. With the Moldflow 2027 release (2026), Autodesk became the first of the generalist CAD/PLM platform vendors to ship any AI inside an injection-moulding tool (moulding specialist Moldex3D shipped moulding AI in 2024–25): the LLM-based Autodesk Assistant, in Tech Preview, which "answers questions, helps interpret results, and performs a limited set of tasks." That is real, dated, primary-source evidence — but it is a chat-and-interpret assistant bolted beside the solver, not machine learning doing the physics. The solvers themselves remain classical. An in-house neural-network surrogate for warpage exists only as early research: a two-input scikit-learn perceptron demonstrated on a single part in April 2025. [Evidence] ⁸¹⁰
Genuine AI elsewhere, and unusually open. Autodesk's real shipped ML lives in Fusion (transformer-based AutoConstrain, AI-assisted automated drawings, the Assistant) and in automotive styling (the BlankAI acquisition, now Form Explorer for Alias). Its research arm publishes and open-sources aggressively — UV-Net, BRepNet, the Fusion 360 Gallery dataset — the exact opposite of Dassault's closed stance (see the Dassault Systèmes report). [Evidence] ¹⁵²⁵¹⁹
The automotive caveat. Autodesk's automotive position is design-studio tools (Alias, VRED) plus Moldflow in the supply chain — not vehicle-program PLM, where Dassault and Siemens dominate. Its strongest automotive-moulding customer evidence is partner-reported (Varroc Polymers, FEKA Automotive). [Synthesis] ³⁰³¹

"Moldflow 2027 is live!", Autodesk Moldflow Insight forum, Mar 27 2026 (date via search-index snapshots; page 403-blocks fetchers) — https://forums.autodesk.com/t5/moldflow-insight-forum/moldflow-2027-is-live/td-p/14066785; "Moldflow 2027 What's New" deck (uploaded May 2026) — https://moldflow.b-cdn.net/wp-content/uploads/2026/05/Moldflow-2027_Final.pdf ↩
Autodesk AI Transparency Card, "Autodesk Assistant – Tech Preview" (model source "Licensed, pre-trained"; provider unnamed; fetched 2026-08-02) — https://damassets.autodesk.net/content/dam/autodesk/www/pdfs/autodesk-assistant-tech-preview.pdf ↩
Autodesk fiscal 2026 Q4/full-year results, Feb 26 2026 — https://investors.autodesk.com/news-releases/news-release-details/autodesk-inc-announces-fiscal-2026-fourth-quarter-results ↩↩↩↩
Autodesk fiscal 2025 Q4/full-year results, Feb 27 2025 — https://investors.autodesk.com/news-releases/news-release-details/autodesk-inc-announces-fiscal-2025-fourth-quarter-and-full-year ↩
Autodesk completes Moldflow acquisition (~$297M), Jun 25 2008 — https://investors.autodesk.com/node/13416/pdf ↩↩↩↩
Moldflow history (founded 1978, Melbourne; Colin Austin) — https://en.wikipedia.org/wiki/Moldflow ↩↩↩
Moldflow Insight "What's New in 2026" (STAMP default, RTM thermal, Digimat, scan-comparison warp result) — https://help.autodesk.com/cloudhelp/2026/ENU/MoldflowInsight-CLC-WhatsNew/files/MFLO-WHATS-NEW-2026-0.html ↩↩↩↩↩↩
Moldflow Insight "What's New in 2027" — Autodesk Assistant Tech Preview: "answers questions, helps interpret results, and performs a limited set of tasks" — https://help.autodesk.com/cloudhelp/2027/ENU/MoldflowInsight-CLC-WhatsNew/files/MFLO-WHATS-NEW-2027-0.html ↩↩↩↩↩↩↩↩↩↩↩↩↩↩
Moldflow Insight 2027 New Features — aiassistant utility (JSON export of geometry/material/machine/result summaries "to help AI tools work with Moldflow data"); cooling optimization for drill-and-plug; 3D corner effects — https://help.autodesk.com/cloudhelp/2027/ENU/MoldflowInsight-CLC-WhatsNew/files/whats-new-2027-0/MFLO-WHATS-NEW-2027-0-FEATURES.html ↩↩↩↩↩↩
"Injection Molding Simulation: Research and Development Updates," Dr. Franco Costa, Autodesk, April 2025 (PDF) — NN surrogate slides (scikit-learn v1.5.0 MLP, layers 470/255, melt-temp + flow-rate inputs, deflection outputs, 3×3-of-9×9 training grid), cooling circuit optimization (42% metric improvement), STAMP (US pat. app. 17/959,221), Synventive validation, research disclaimer — https://damassets.autodesk.net/content/dam/autodesk/www/campaigns/advmfg-summit/2025/Moldflow%20Research%20and%20Development%20Updates_Franco%20Costa.pdf ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩
"What's new in Moldflow 2026," Hanno van Raalte, Autodesk summit deck 2025 (PDF) — https://damassets.autodesk.net/content/dam/autodesk/www/campaigns/advmfg-summit/2025/Whats%20new%20in%20Moldflow%202026%20-%20Summit_2025_Hanno%20van%20Raalte.pdf ↩
Autodesk Moldflow Insight 2026 Solver Validation Report (PDF) — https://damassets.autodesk.net/content/dam/autodesk/www/pdf/simulation-hub/Autodesk%20Moldflow%202026%20Solver%20Validation%20Report.pdf ↩↩↩
"AI takes main stage at Autodesk University" / "Introducing Autodesk AI," Nov 13 2023 — https://adsknews.autodesk.com/en/pressrelease/autodesk-university-2023/; investor release — https://investors.autodesk.com/news-releases/news-release-details/introducing-autodesk-ai-design-and-make ↩↩↩
"New investments in Autodesk Fusion to bring AI-powered transformation to manufacturing," Sep 16 2025 (foundation models for editable geometry; Assistant in Fusion & Vault; Microsoft Azure OpenAI + GPT-image-1) — https://adsknews.autodesk.com/en/news/new-investments-in-fusion-bring-ai-powered-transformation-to-manufacturing/ ↩↩↩↩↩↩↩↩↩↩
Engineering.com, "3 new AI features in Autodesk Fusion," Oct 17 2024 (AutoConstrain transformer model, ship Jan 2025; Automated Drawings AI since Jan 2024; Assistant in Fusion manufacturing workspace) — https://www.engineering.com/3-new-ai-features-in-autodesk-fusion/ ↩↩↩↩↩↩↩↩↩↩↩
Autodesk Research blog, "Using AI to Power AutoConstrain in Fusion Automated Sketching" — https://www.research.autodesk.com/blog/using-ai-to-power-autoconstrain-in-fusion-automated-sketching/ ↩
Autodesk Assistant product page ("your agentic AI partner"; Tech Preview across products including Moldflow) — https://www.autodesk.com/solutions/autodesk-ai/autodesk-assistant (page listed in search results; direct fetch blocked — see Appendix C) ↩
Autodesk Research, Project Bernini, May 2024 ("strictly experimental… not available for public use"; ~10M training shapes) — https://adsknews.autodesk.com/en/news/research-project-bernini/ ↩↩↩↩↩
"AI in the design studio: Welcome to BlankAI," Autodesk Design Studio blog, Nov 2 2023 (XDGAN trained on historical 3D designs; automotive-targeted; not a CAD product) — https://blogs.autodesk.com/design-studio/2023/11/02/welcome-to-blankai/ ↩↩↩↩↩
DEVELOP3D on Autodesk AI foundations / Form Explorer for Alias (from the BlankAI acquisition) — https://develop3d.com/cad/autodesk-lays-the-foundations-for-ai/ ↩↩↩↩↩
Autodesk Fusion generative design for manufacturing — https://www.autodesk.com/solutions/generative-design/manufacturing ↩↩↩
Autodesk Fusion blog, "Topology Optimization is not Generative Design" — https://www.autodesk.com/products/fusion-360/blog/topology-optimization-is-not-generative-design/ ↩
Fusion Simulation Extension (injection-moulding study type; manufacturability focus) — https://www.autodesk.com/products/fusion-360/simulation-extension; Autodesk plastic injection molding solutions — https://www.autodesk.com/solutions/simulation/plastic-injection-molding ↩↩↩
MF Software (Autodesk Moldflow partner) on Fusion plastic analysis: "using the high-performance Moldflow computational solvers"; part-only; draft analysis in Product Design Extension — https://www.moldflow.eu/en/autodesk-fusion/ ↩↩↩
UV-Net repo (MIT license; CVPR 2021, Jayaraman et al.) — https://github.com/AutodeskAILab/UV-Net; paper page — https://www.research.autodesk.com/publications/uv-net-learning-from-boundary-representations/ ↩↩↩↩↩↩
Fusion 360 Gallery Dataset (non-commercial research use; ~8,600 design sequences; BRepNet 92.5% segmentation accuracy) — https://github.com/AutodeskAILab/Fusion360GalleryDataset ↩↩↩↩
Autodesk AI Lab GitHub organisation — https://github.com/AutodeskAILab ↩↩↩↩
"Surrogate Modelling for Injection Molding Processes using Machine Learning" (third-party, uses Moldflow project data), arXiv:2107.14574 — https://arxiv.org/abs/2107.14574 ↩
GM × Autodesk generative-design seat bracket (40% lighter, 20% stronger; >150 options), 2018 — https://adsknews.autodesk.com/en/news/gm-autodesk-using-generative-design-vehicles-future/; customer story — https://www.autodesk.com/customer-stories/general-motors-generative-design ↩↩↩↩↩
Varroc Polymers rapid heat-and-cool Moldflow story (Neilsoft partner success story on autodesk.com) — https://www.autodesk.com/support/partners/success-stories/moldflow-for-rapid-heating-and-cooling/8542 ↩↩↩↩
FEKA Automotive Moldflow Insight partner success story — https://www.autodesk.com/support/partners/success-stories/feka-automotive-achieves-success-with-the-right-simulation-solutions/8265 ↩↩↩↩
Autodesk automotive industry page — https://www.autodesk.com/industry/automotive ↩
Autodesk automotive design software (Alias for Class-A/concept; VRED for visualization) — https://www.autodesk.com/solutions/design-manufacturing/automotive-design-software ↩↩↩
Autodesk VRED overview — https://www.autodesk.com/products/vred/overview ↩
Autodesk Q3 fiscal-2026 earnings call, Nov 25 2025 (Motley Fool transcript) — https://www.fool.com/earnings/call-transcripts/2025/11/25/autodesk-adsk-q3-2026-earnings-call-transcript/ ↩↩
Plastics Today on Simcon Cadmould "AI Solver" built with Emmi AI (~1,000× exploration claim) — https://www.plasticstoday.com/injection-molding/large-engineering-model-software-slashes-injection-molding-simulation-time ↩↩↩
SPE Injection Molding Division, "Moldex3D 2025 Molding Intelligence," Apr 22 2025 — https://www.injectionmoldingdivision.org/2025/04/22/moldex3d-2025-molding-intelligence-embracing-the-ai-revolution-in-injection-molding/ ↩↩
What "optimization technology" means and why it is not generative AI — see the concept note What is Optimization technology: topology optimization and cooling-layout search compute shapes from physics equations in an iterative loop; they learn nothing from data, so they are classical optimization, not AI. ↩↩↩↩↩↩