Skip to content

3.3.1 Platform Vendors

Platform vendors are the companies that sell the CAD, CAE and simulation software on which automotive moulded parts are actually designed and validated: Dassault Systèmes, Siemens, Autodesk, PTC, and the moulding-simulation specialists (Moldex3D, SIGMASOFT, SIMCON), alongside the broader engineering platforms (Hexagon, Cadence). If any group in this study is going to turn AI/ML into a working part of the injection-moulding design workflow, it is most likely to appear here first: these vendors own the geometry, the solvers, the material databases and the day-to-day tools that engineers already use.

That is why this group leads the study. It sets the baseline for everything that follows. When a service provider, an OEM or a Tier-1 supplier talks about "AI in moulding," they are almost always talking about a capability that either ships inside one of these platforms or competes with it. Reading the platform vendors first makes the rest of the study legible.

The company reports are written around a single, deliberately strict question, the honest-AI test: does the system learn from data, or does it compute from equations? Decades of optimisation, design-of-experiments and rule-checking are now routinely relabelled "AI," so each report separates true machine learning (models that improve from examples) from classical numerical methods wearing a new adjective. Marketing framing is quoted, tagged [Marketing], and cross-referenced to the concept note What is Optimization technology rather than taken at face value.

Applying that test across the nine vendors studied below produces a consistent through-line, the spine that runs through every report in this section:

  • No vendor learns from measured production outcomes. Every "learning" capability found is either simulation-taught (models trained on solver output, not on real defects, warpage or scrap) or document-bound (an LLM assistant retrieving from manuals and prior projects). The feedback loop from the moulding floor back into the design tool does not yet exist in any shipping product. This negative finding is a first-class result, not a gap in the research. [Synthesis + Evidence of absence]
  • Every platform vendor rents its language-model layer. The conversational and copilot features are built on third-party foundation models (Mistral, Microsoft/OpenAI, AWS), not on models the vendors train themselves. [Synthesis]
  • The genuine, dated "firsts" are narrow but real, and worth naming up front so the reader can watch for them:
    • Autodesk: first AI assistant embedded inside a mould-flow product from a generalist CAD/PLM vendor.
    • Siemens (+Altair): first ML trained on mould-flow simulation results found by this study (Jun 2026).
    • Moldex3D: first shipping moulding retrieval knowledge base.
    • SIMCON: first claimed neural moulding solver (research preview, Mar 2026).

Read together, the section makes a case that is easy to miss vendor-by-vendor: the industry has begun to bolt learning onto simulation, but no one has yet closed the loop back to real manufacturing outcomes. Keep that in mind as a lens while reading each report: where a vendor sits relative to that line is usually the most important thing about its AI story.

3.3.1.1 Selection criteria

This study gives a full report only to vendors that pass three tests:

  1. (a) Real automotive presence: named OEM/supplier customers or a documented automotive business, not just "we serve all industries."
  2. (b) Real moulded/plastic-parts capability: a product that actually addresses injection-moulded thermoplastic parts (mould-flow simulation, mould tooling design, moulding DFM), not merely generic CAD/CAE that could touch a plastic part.
  3. (c) Concrete, dated AI capability: a shipping or announced AI/ML product with a name and a date, not an "AI-powered" adjective.

A vendor that fails (b) is excluded even when its automotive and AI credentials are strong; that asymmetry is itself a study finding (see "Considered but not studied in detail" below, and the two scope notes). [Synthesis]

3.3.1.2 Companies studied in detail

Nine vendors passed all three tests and receive a full report, each following the study's standard report template.

# Vendor One-line role
01 Dassault Systèmes 3DEXPERIENCE / CATIA platform; rents its LLM layer from Mistral.
02 Siemens (+Altair) NX / Simcenter; first ML trained on mould-flow results found by this study (Jun 2026).
03 Autodesk Moldflow; first AI assistant inside a mould-flow product from a generalist CAD/PLM vendor.
04 PTC Creo / Windchill; copilots rented via Microsoft and AWS.
05 Moldex3D Dedicated moulding solver; first shipping moulding retrieval knowledge base.
06 SIGMASOFT Moulding simulation with autonomous optimisation.
07 Hexagon Metrology + Digimat (before the Cadence divestment).
08 SIMCON Cadmould; first claimed neural moulding solver (research preview, Mar 2026).
09 Cadence Acquired Hexagon's Digimat/MSC design-and-engineering business (Feb 2026).

3.3.1.3 Considered but not studied in detail

3.3.1.3.1 Scope notes (researched, written up, excluded)

Two famous simulation companies keep coming up in this space because they have genuine automotive businesses. Both were researched and both were excluded: each fails exactly one test, (b) the moulded-parts test. Each has its own short scope note (steelman, the test it fails, what would change our mind):

Vendor Why excluded
ESI Group (now Keysight) No thermoplastic injection-moulding solver; moulding-adjacent work is composites RTM and metal casting.
Ansys (now Synopsys) No native injection-moulding solver; a host other vendors' moulding tools plug into. Strong AI (SimAI), but not in moulding.

3.3.1.3.2 Dismissed briefly

Names that surfaced during screening and were set aside without a full note. Each fails test (b), test (c), or is out of scope as a company:

  • Keysight and Synopsys themselves: the new parents of ESI and Ansys. Test/measurement and EDA companies respectively; neither has any moulded-parts product of its own. Assessed only as owners of the two vendors above. [Evidence] 12
  • CYBERNET (PlanetsX): it does make an injection-moulding solver, but as a regional (Japan-centric) Ansys Workbench add-on with no documented AI capability found. Fails test (c). [Evidence — thin] 3
  • MAGMA (MAGMASOFT): metal-casting simulation specialist. No thermoplastic injection-moulding solver; casting is a different process. Fails test (b). [Evidence] 4
  • Flow Science (FLOW-3D): general free-surface CFD covering some casting/moulding-adjacent flow, but no dedicated thermoplastic injection-moulding DFM product. Fails test (b). [Evidence] 5
  • OpenFOAM: ESI stewards this open-source CFD project6; it is a community codebase, not a vendor, and has no moulding-specific or AI product offering relevant here. Out of scope as a company. [Evidence]

The section's core finding — the AI-moulding asymmetry that sits above the individual reports — is drawn together with the other four groups in Insights from DFM Companies → Conclusion.

3.3.1.4 Things to watch

The findings above are a 2026 snapshot. Three developments are most likely to move it, and are worth keeping in view while reading the nine company reports.

  • Who closes the AI-moulding gap first. A moulding specialist acquiring real AI, or an AI-strong platform acquiring moulding physics: whichever happens first reshapes this group. This is the single most consequential move to watch. [Synthesis]
  • Whether anyone closes the production-feedback loop. No vendor yet learns from measured factory outcomes (defects, warpage, scrap); every "learning" capability today is simulation-taught or document-bound. The first product that trains on real moulding results (not solver output) would overturn the spine finding of this section. [Synthesis + Evidence of absence]
  • Whether the early "firsts" mature into everyday tools. Autodesk's in-solver assistant, Siemens/Altair's ML-on-mould-flow (Jun 2026), Moldex3D's retrieval knowledge base, and SIMCON's claimed neural solver (research preview, Mar 2026) are all young. Which of them hardens into something engineers rely on daily is the near-term signal to track. [Synthesis]

3.3.1.5 References

  1. Keysight announces result of cash tender offer for remaining ESI shares, Jan 2024 (ESI delisted from Euronext Paris Jan 26 2024): https://investor.keysight.com/investor-news-and-events/financial-press-releases/press-release-details/2024/Keysight-Announces-Result-of-Cash-Tender-Offer-for-Shares-of-ESI-Group/default.aspx ↩

  2. Synopsys completes acquisition of Ansys, Jul 17 2025 (announced Jan 16 2024): https://news.synopsys.com/2025-07-17-Synopsys-Completes-Acquisition-of-Ansys ↩

  3. PlanetsX, injection-moulding CAE for Ansys Workbench, by CYBERNET: https://cybernet.co.jp/ansys/product/lineup/planetsx/en ↩

  4. MAGMASOFT, casting process simulation, MAGMA Giessereitechnologie GmbH: https://www.magmasoft.de ↩

  5. FLOW-3D, computational fluid dynamics software, Flow Science Inc.: https://www.flow3d.com ↩

  6. ESI business unit strengthens commitment to OpenFOAM: https://www.esi-group.com/news/esi-business-unit-as-part-of-keysight-technologies-strengthens-commitment-to-openfoam ↩