Skip to content

Hyundai Motor Group (HMG) — incl. Hyundai Motor Co., Kia Corp., Hyundai Mobis

Company Overview

Business Overview

Hyundai Motor Group is a family-led Korean industrial group (chaebol) headquartered in Seoul. The affiliates relevant here: Hyundai Motor Company (incl. the premium brand Genesis), Kia Corporation, Hyundai Mobis (the group's captive Tier-1 — modules, lamps, electronics), plus Hyundai Wia, Hyundai Steel and Hyundai Transys. Robotics holdings include Boston Dynamics — explicitly out of scope for this report, as no public evidence connects it to parts-design engineering. [Synthesis — corporate structure is common knowledge; volume ranking and shareholding percentages left unverified, see Appendix C]

Euisun Chung (Executive Chair) frames the group's current direction around AI-powered mobility and smart factories: "Deepening our collaboration with NVIDIA marks a pivotal step forward in AI-powered mobility and smart factories" (Oct 30, 2025). [Evidence + Marketing] 4

Automotive Business

HMG is the automotive business — this section inverts for an OEM. What matters for this study is where moulded parts sit organisationally:

  • Interior/exterior modules, cockpit assemblies and lamps are largely engineered and supplied by Hyundai Mobis, the group's captive Tier-1. Mobis maintains an in-house "Mold Technology Team" — attested by a 2016 peer-reviewed paper on fluid-assisted injection moulding authored from that team. [Evidence] 10
  • Vehicle-level plastic closure design sits inside brand engineering teams — a "Genesis Closure Engineering Design Team" appears as an author affiliation in
  • [Evidence] 9
  • Actual mould design/build involves external Korean mould makers (e.g. Q-JEN Mold Tech, co-author on the same paper). [Evidence] 9

Engineering Business

HMG is a buyer of engineering platforms, not a seller. The two structural facts:

  • Siemens NX + Teamcenter were selected as the group's next-generation engineering platform — Hyundai Motor Company and Kia named Siemens "preferred partner" on Nov 29, 2021 (the same citation used in this study's Siemens report). CAD and PLM across the group are therefore Siemens-based. [Evidence] 1
  • Siemens Simcenter (Amesim system simulation, Reduced Order Modeling, HEEDS optimization) is used in vehicle-attribute engineering, with Teamcenter integration planned. [Evidence] 2

Role in Automotive Moulded Parts Workflow

For an OEM the table reads "who executes this stage, and is any AI visible there?"

Workflow Stage Executed by AI visible? (as of Aug 2026)
Requirements HMG brand engineering No public evidence
Industrial Design HMG design centres No public evidence (styling AI not searched in depth)
Concept Design HMG engineering No
CAD Modelling HMG on Siemens NX No moulded-parts AI; platform is conventional CAD1
Engineering Review HMG on Teamcenter No
Simulation (vehicle attributes) HMG virtual development teams Yes — neural-net ROM + HEEDS (chassis)2
DFM (moulded parts) Mobis + mould makers No AI evidence — the report's central negative
Tool Design External mould makers (e.g. Q-JEN) No; Bayesian-optimization collaboration is academic9
Mould Flow Mobis / mould makers (tools unknown) No public evidence of which solver, or any ML
Prototype / Validation HMG + suppliers No moulding-specific AI
Manufacturing Engineering HMG smart factories Omniverse digital twins, robotics — process, not part design3
Production Release HMG on Teamcenter No

Structural takeaway. The AI that exists clusters at the two ends — vehicle-level simulation upstream, factory/robotics downstream. The moulded-parts middle (DFM, tool design, mould flow) is delegated to the supplier tier and shows no public AI adoption. [Synthesis]

AI Strategy

Public AI Vision

HMG's public AI programme has four visible strands. [Synthesis]

  1. Physical AI at scale. With NVIDIA: first a strategic partnership (CES, Jan 9, 2025 — Omniverse digital-twin smart factories, Isaac robotics, autonomous-driving simulation)3, then an "AI factory" with 50,000 NVIDIA Blackwell GPUs and ~US$3 billion of investment (Oct 30, 2025), including a "Physical AI Application Center" in Korea. [Evidence] 4
  2. AI for materials discovery. Partnership with CuspAI (Cambridge, UK — generative AI plus physics for on-demand materials design), signed Nov 6, 2025. [Evidence] 5
  3. Enterprise generative-AI assistants. Hyundai Mobis launched MoAI ("Mobis one AI"), an in-house conversational generative-AI service wired into work systems for R&D, quality and sales tasks (Oct 21, 2025). [Evidence — trade press]7
  4. Open innovation. The ZER01NE platform (creators + startups; 2025 ZER01NE Day ran Sept 17–21, 2025 with 8 creator projects and 11 startups, including 5 via the group's global CRADLE offices) showcased generative AI, robotics and AI process automation. [Evidence] 6

Investor / Executive Statements

  • Euisun Chung on the AI factory: see §1. [Marketing] 4
  • Heung-Soo Kim (EVP, Head of Global Strategy Office), on NVIDIA: "This partnership is set to accelerate our progress, positioning the Group as a frontrunner in driving AI-empowered mobility innovation." [Marketing] 3
  • Chul Park (VP, Head of New Business Strategy Group), on CuspAI: "Hyundai Motor Group is driving transformative change in mobility through fundamental materials innovation." [Marketing] 5

Engineering AI Strategy

Where AI touches design engineering specifically, the pattern is surrogate modelling over its own simulation output, delivered by a vendor:

  • Siemens Simcenter Engineering Services trained a neural network on Hyundai's own batch of 200,000+ Amesim simulation results, then coupled it to the HEEDS optimizer so requirement evaluation runs on the surrogate instead of the solver. This is machine learning by this study's honest test — the system learns from (simulation) data — but note the recurring pattern: learning is simulation-taught, not production-outcome-taught, exactly as with every company in the platform vendors group. [Evidence + Synthesis] 2
  • The optimization layer around it (HEEDS parameter search; the tailgate paper's Bayesian optimization) is the classical search family — see the study concept note What is Optimization technology. Bayesian optimization does fit a statistical surrogate from sampled simulations, but it learns within a single job and carries nothing between programmes. [Synthesis] 9

Timeline of AI Evolution

Date Event
Nov 29, 2021 Siemens NX + Teamcenter chosen as next-gen engineering platform1
Nov 28, 2023 Simcenter ROM chassis case published — 1 week → 15 min2
Jan 9, 2025 NVIDIA strategic partnership (Omniverse, Isaac, AV simulation)3
Jan 2025 Bayesian tailgate-rib optimization paper (HMG + KAIST + Q-JEN)9
Jun 2025 Hyundai Mobis generative-AI paper honoured at CVPR 20258
Sep 17–21, 2025 2025 ZER01NE Day (AI, robotics, generative-design startups)6
Oct 21, 2025 Hyundai Mobis launches MoAI internal generative-AI service7
Oct 30, 2025 NVIDIA AI factory — 50,000 Blackwell GPUs, ~$3B4
Nov 6, 2025 CuspAI materials-discovery partnership5

Engineering Programmes Relevant to this Study

Simcenter ROM chassis programme — Purpose: set chassis parameters (mass distribution, suspension kinematics, mounting system) against comfort/handling targets for the Genesis GV80 electric vehicle. AI: neural-network Reduced Order Model trained on 200,000+ Amesim models; coupled to HEEDS; 52 KPIs, 350+ input parameters. Moulded-parts relevance: none directly — but it is the proof that HMG values and pays for exactly the surrogate-model pattern a moulding-AI product would sell. [Evidence] 2

NVIDIA smart-factory / physical-AI programme — Omniverse factory digital twins, Isaac robotics, AV simulation; scaled by the Oct 2025 AI factory. Moulded-parts relevance: none at the part-design level; manufacturing-floor AI only. [Evidence] 34

CuspAI materials programme — generative AI + physics for materials discovery; "framework for collaboration across multiple domains." No material classes are named in the announcement — polymers/plastics are not mentioned. Moulded-parts relevance: potential (future polymer discovery) but currently unproven. [Evidence] 5

Mobis MoAI — in-house conversational generative-AI service connected to work systems (R&D, quality, sales), launched Oct 21, 2025. Document-bound assistant work, consistent with the cross-vendor finding that all shipped "learning" at this layer is document-bound. [Evidence — trade press] 7

ZER01NE / CRADLE — the group's startup and creator platform; 2025 edition showcased generative AI, robotic automation and generative-design projects. [Evidence] 6

AI Capabilities

1. Neural-net Reduced Order Modeling of vehicle dynamics (with Siemens). Description: surrogate model trained on Hyundai's own simulation campaign; HEEDS searches on the surrogate. Stage: vehicle-attribute simulation/target setting. Inputs: 350+ chassis parameters. Outputs: 52 KPIs per requirement. Quantified outcome: optimization 1 week → 15 minutes; per-evaluation 2 min → 0.1 s. Quote: "We can now find the optimal parameter set very quickly by searching through the neural network" — Ilsoo Jeong, comfort engineer, Driving Comfort Virtual Development Team. Limitation: vendor-delivered (Simcenter Engineering Services trained the network); simulation-taught; nothing to do with moulding. [Evidence] 2

2. Bayesian optimization of injection-moulded rib structures (with academia). Description: tailgate rib layout optimized for stiffness under injection- moulding manufacturing constraints, using Bayesian optimization (surrogate-guided search). Stage: DFM/part design. Limitation: an academic collaboration (J. Manuf. Processes, Jan 2025), not an internal HMG tool; classical optimization-family machinery — see the concept note What is Optimization technology. [Evidence] 9

3. Mobis MoAI generative-AI assistant. Internal LLM-based service for R&D, quality and sales work. Limitation: productivity assistant; no evidence it touches part design or moulding data; underlying model undisclosed. [Evidence — trade press]7

4. Mobis generative AI for virtual driving environments (CVPR 2025). A generative model recreating virtual driving environments, honoured at CVPR (June 2025) — evidence of genuine in-house ML research capability at the Tier-1, aimed at autonomous-driving validation, not parts. [Evidence] 8

5. CuspAI materials discovery. Generative AI + physics search over candidate materials. Limitation: announcement-stage; no materials named, no quantified goals. [Evidence] 5

6. Factory digital twins and robotics (NVIDIA). Omniverse-simulated plants, Isaac-based robots, an AI factory of 50,000 Blackwell GPUs. Limitation: manufacturing-floor and mobility AI; no announced application to part or tool design. [Evidence] 34

What was NOT found (first-class negative): no HMG or Mobis system that applies AI/ML to injection-moulding DFM, mould-flow prediction, mould design, or moulding defect prediction; no moulding-AI procurement announcement; no HMG-affiliated deep-learning moulding paper. The one moulding-adjacent optimization work (item 2) was executed through universities and a mould maker. [Evidence of absence]12

Engineering Workflow Contribution

As a demand-side actor, HMG's workflow runs on bought platforms:

requirements and programme data in Teamcenter → CAD in NX → vehicle attribute simulation in Simcenter Amesim (now ROM-accelerated for chassis) → moulded interior/exterior modules engineered at Hyundai Mobis → moulds built by external Korean mould makers → production in HMG plants (Omniverse digital twins arriving via NVIDIA).

The AI insertions to date sit at the simulation stage (ROM surrogates) and the factory stage (digital twins/robotics). Nothing AI-shaped is publicly inserted between CAD and mould build — the moulded-parts DFM/tooling gap this study tracks. [Synthesis] 123

Public Customer Evidence

(For an OEM this inverts: HMG appears as the customer in vendors' evidence.)

Case Studies

  • Siemens Simcenter ROM chassis case (Nov 28, 2023) — the study's strongest quantified OEM engineering-AI outcome: 200,000+ Amesim models, 1 week → 15 min, 2 min → 0.1 s per evaluation, 52 KPIs, 350+ parameters, Genesis GV80 EV comfort and handling. Named engineer quoted (Ilsoo Jeong). Published by the vendor — treat the framing as Siemens marketing over a real, specific, dated engagement. [Evidence + Marketing] 2
  • Siemens NX/Teamcenter selection (Nov 29, 2021) — platform-level deployment evidence; no quantified outcome; nothing plastics-specific. [Evidence] 1
  • NVIDIA partnership + AI factory (Jan 9 and Oct 30, 2025) — infrastructure commitment (~$3B, 50,000 GPUs); outcomes not yet demonstrable. [Evidence] 34

Conference / Academic Evidence

  • CVPR 2025 — Mobis generative-AI paper honoured (June 2025). [Evidence] 8
  • Journal of Manufacturing Processes, Jan 2025 — the Bayesian tailgate-rib paper with the Genesis closure-team co-author. [Evidence] 9

The gap

No public case combines HMG + AI + moulded plastic parts + a quantified outcome. The nearest miss is the tailgate paper (moulding constraints, no AI-learning, academic venue) and the ROM case (real ML, no moulding). [Evidence of absence]

Technical Architecture (Inferred)

Evidence

  • Siemens NX + Teamcenter as the engineering backbone (2021).1
  • Simcenter Amesim + Simcenter ROM + HEEDS in attribute engineering; Teamcenter integration "planned" as of the 2023 blog.2
  • NVIDIA Omniverse (factory twins), Isaac (robots), Blackwell AI factory (training/validation/deployment infrastructure).34
  • Mobis MoAI connected to internal work systems.7

Synthesis

HMG's AI stack is assembled from partners: Siemens supplies the engineering-AI layer, NVIDIA the compute and physical-AI layer, CuspAI the materials-AI layer, and the Tier-1 builds its own assistant layer. There is no sign of an HMG-owned engineering-AI platform product. [Synthesis]

Inference

The 50,000-GPU AI factory is sized for autonomous driving, robotics and factory AI — domains named in the release. Part-design AI is not named; if moulding AI ever runs there it will be because a supplier or partner brings the application, not because the infrastructure implies it. [Inference] 4

AI Technologies

In one list: neural-network surrogates / ROM (vendor-delivered, chassis domain)2; Bayesian/DoE optimization (academic collaboration; classical search family9); enterprise LLM assistant (Mobis MoAI — underlying model undisclosed7); generative vision models (Mobis, CVPR 20258); generative AI + physics for materials (CuspAI, external5); digital twins and robotics AI (NVIDIA Omniverse/Isaac3). Nothing in this list is a moulding technology. [Synthesis]

Research Publications

The affiliation sweep (this report's original research)

Method: OpenAlex query for works matching "injection molding" whose raw author affiliation strings contain "hyundai" (and separately "kia"). Result: 34 works (Hyundai family) + 4 (Kia, all overlapping/legacy). [Evidence] 12

What the 34 contain:

  • The headline: Lee, Yeo, Kong (Genesis Closure Engineering Design Team, Hyundai Motor Group + KAIST), Myeong & Jang (Q-JEN Mold Tech13 — technical sales and mould design departments), Lee & Choi (Ulsan Technopark)14, Kim (Sogang Univ.)15, Ryu (KAIST)16: "Bayesian optimization of tailgate rib structures enhancing structural stiffness under manufacturing constraints of injection molding," Journal of Manufacturing Processes 134:739–748, January 2025, DOI 10.1016/j.jmapro.2024.12.064. An OEM closure engineer + a mould maker + two universities optimizing rib layout under moulding constraints — the exact problem a moulding-DFM product addresses, solved today by ad-hoc academic collaboration. [Evidence] 9
  • Hyundai Mobis Mold Technology Team (Hyungpil Park): fluid-assisted injection-moulding study, Int. J. Adv. Manuf. Technol., 2016 — proof of an in-house mould-technology group at the Tier-1; classical experiment, no ML. [Evidence] 10
  • Legacy classical work: fibre-orientation simulation from the "Vehicle Development & Analysis Team, R&D Division, Hyundai Motor Co. & Kia Motor Corp." (2002); a gate-location study from Hyundai Autonet's mould department (2003); Hyundai Steel metal-injection-moulding chapters (2012). All pre-AI numerical work. [Evidence] 1112

The negative that matters: Korea's universities and institutes produce a large share of the world's injection-moulding-ML literature, yet across the entire affiliation sweep not one deep-learning moulding paper carries an HMG affiliation. The group's moulding engineering publishes classical work and, at most, optimization-family collaborations. [Synthesis + Evidence of absence] 12

Other publications

Mobis publishes genuine ML research in computer vision (CVPR 2025).8 Patents were not systematically searched (see Appendix C).

Open Source

No moulding- or design-engineering-relevant open-source AI from HMG or Mobis was found. (This was not an exhaustive sweep of all group GitHub organisations; the software-defined-vehicle subsidiary 42dot's activities were not audited — see Appendix C.) [Evidence of absence — limited search]

Engineering Service / Platform Mapping

  • CAD: Siemens NX (group standard since 2021).1
  • PLM: Siemens Teamcenter.1
  • Simulation: Siemens Simcenter Amesim (+ ROM, HEEDS) in attribute engineering.2
  • Mould flow: the group's/suppliers' mould-flow toolchain is not publicly documented — a notable blank. [Evidence of absence]
  • Factory/robotics AI: NVIDIA Omniverse, Isaac, Blackwell AI factory.34
  • Assistants: Mobis MoAI (internal).7

Engineering Intelligence Stack Mapping

  1. Intent — requirements live in Teamcenter; no AI-assisted intent capture is publicly visible. [Evidence of absence]
  2. Knowledge — MoAI wires generative AI into Mobis work systems (document-bound knowledge); no evidence of a geometry- or moulding-knowledge base. [Evidence] 7
  3. Reasoning — the ROM surrogate + HEEDS loop is genuine machine-assisted reasoning over design space — for chassis attributes only; the moulding equivalent does not exist publicly. [Evidence] 2
  4. Execution — CAD/PLM execution is conventional; factory execution is where the robotics/digital-twin AI lands. [Evidence] 3
  5. Feedback — consistent with every company in this study: no public evidence of a learning loop from measured production outcomes (moulding defects, warpage, scrap) back into design. [Evidence of absence]

Strengths

  • Proven appetite for surrogate-model AI with hard ROI numbers — the 1-week → 15-minute chassis case is the strongest quantified OEM number in this study's research on the automotive OEMs group. [Evidence] 2
  • Compute and capital commitment — ~$3B / 50,000 GPUs makes infrastructure a non-obstacle for any future engineering-AI application. [Evidence] 4
  • Real in-house ML research talent at the Tier-1 (Mobis at CVPR). [Evidence] 8
  • Formal open-innovation doors — ZER01NE and CRADLE actively showcase external AI startups, including generative-design projects. [Evidence] 6
  • Willingness to partner rather than build (Siemens, NVIDIA, CuspAI) — good news for outside vendors and new entrants. [Synthesis]

Weaknesses

  • No AI reaches moulded-parts engineering — no DFM AI, no mould-flow ML, no moulding knowledge base, in any public source found. [Evidence of absence] 12
  • Moulding know-how is organisationally fragmented — split between brand closure teams, Mobis's mould-technology group and external mould makers; the tailgate paper shows cross-boundary problems being solved through academia. [Synthesis] 9
  • The flagship engineering-AI capability is vendor-delivered — Simcenter Engineering Services trained the ROM network; the capability lives in the Siemens relationship, not (yet demonstrably) in-house. [Evidence] 2
  • Announcement-stage items (CuspAI, AI factory) have no engineering outcomes yet. [Evidence] 54

Current Gaps (largely manual today)

On public evidence: moulded-part DFM review, mould-flow interpretation, rib/draft/ wall-thickness design decisions, and mould design remain human, rules-and-experience work distributed across HMG teams, Mobis and mould makers. The tailgate paper is direct evidence that even a premium-brand closure team resorts to an external academic collaboration to get moulding-constrained optimization done. [Evidence + Inference]9

Future Direction

  • Scale "physical AI" (factories, robots, autonomy) on the Blackwell AI factory; Korea-based ecosystem build-out with a Physical AI Application Center. [Evidence] 4
  • Materials discovery with CuspAI — watch whether polymers/plastics ever get named. [Evidence] 5
  • Deepen the Siemens engineering stack (Teamcenter-integrated ROM was already "planned" in 2023). [Evidence] 2
  • My read: engineering AI will keep arriving vendor-first. The moulded-parts gap will be filled by whichever supplier or vendor brings a credible product — HMG shows no sign of building one itself. [Inference]

Relevance to Automotive Moulded Parts

Capability State at HMG Notes
Plastic Part Design Conventional NX-based; closure teams (Genesis) do rib/stiffness work manually or via academic collabs9
Surface Design (Class A) Conventional No public AI evidence
CAD Automation None visible No public NX-automation or CAD-AI announcements found
DFM Manual / rules — no AI The report's central negative12
Tool Design External mould makers Q-JEN-type suppliers; no AI evidence
Mould Flow Toolchain not public No solver or ML evidence either way
Manufacturing Engineering AI arriving Omniverse twins, robotics — process-side only3
Quality No design-loop No defect-data feedback into design found
Engineering Knowledge Reuse Document-bound only MoAI (Mobis) is an LLM work assistant, not a geometry/moulding KB7

Critical finding. A group widely ranked among the world's three largest automotive groups, with a $3B AI compute programme and a surrogate-model speed-up of roughly three orders of magnitude in chassis engineering, shows zero public AI adoption in moulded-parts engineering — and its one moulding-optimization effort was outsourced to universities and a mould maker. Demand exists; supply does not. [Synthesis] 249

Key Takeaways

  1. HMG's engineering platform is Siemens NX + Teamcenter (Nov 2021) — moulded parts live in a Siemens world. [Evidence] 1
  2. The Simcenter ROM chassis case (Nov 2023) is the strongest quantified OEM engineering-AI number found: 200k+ models, 1 week → 15 min, 2 min → 0.1 s. [Evidence] 2
  3. That ML is simulation-taught and vendor-delivered — consistent with the study's cross-vendor spine. [Synthesis] 2
  4. No AI reaches HMG's moulded-parts engineering — no DFM AI, no mould-flow ML, no moulding knowledge base, anywhere public. [Evidence of absence] 12
  5. The one moulding-adjacent optimization effort — Bayesian tailgate-rib design, Jan 2025 — was done with KAIST, Sogang, Ulsan Technopark and a mould maker, not with an internal tool. [Evidence] 9
  6. Mobis is the group's moulding centre of gravity (Mold Technology Team, 2016) and now has real ML talent (CVPR 2025) plus an internal GenAI platform (MoAI, Oct 2025). [Evidence] 1087
  7. Compute is a solved problem for HMG: 50,000 Blackwell GPUs, ~$3B (Oct 2025). [Evidence] 4
  8. Materials AI (CuspAI, Nov 2025) is announcement-stage; polymers are not named. [Evidence] 5
  9. Route into the opportunity: supplier tier + ZER01NE/CRADLE doors, carrying the already-validated surrogate pattern into the moulding gap. [Inference]

References

Primary Sources — HMG and partners

  • Siemens press release, Hyundai Motor Company & Kia partner with Siemens, Nov 29, 2021 — https://press.siemens.com/global/en/pressrelease/hyundai-motor-company-and-kia-corporation-partners-siemens-digital-mobility
  • Siemens Simcenter blog, "Hyundai Motor Group uses AI to reduce the parameter optimization process from one week to 15 minutes," Nov 28, 2023 — https://blogs.sw.siemens.com/simcenter/hyundai-motor-group-uses-ai-to-reduce-the-parameter-optimization-process-from-one-week-to-15-minutes/
  • HMG × NVIDIA partnership, Jan 9, 2025 — https://www.hyundai.com/worldwide/en/newsroom/detail/0000000893
  • NVIDIA × HMG AI factory, Oct 30, 2025 — https://nvidianews.nvidia.com/news/hyundai-motor-group-ai-factory
  • HMG × CuspAI partnership, Nov 6, 2025 — https://www.hyundaimotorgroup.com/en/news/hyundai-motor-group-and-cuspai-partner-to-accelerate-material-innovation-using-ai
  • 2025 ZER01NE Day press release, Sep 16, 2025 — https://www.hyundai.news/eu/articles/press-releases/2025-zer01ne-day.html

Research Papers

  • Lee, Yeo, Kong, Myeong, Jang, Lee, Choi, Kim, Ryu, "Bayesian optimization of tailgate rib structures enhancing structural stiffness under manufacturing constraints of injection molding," Journal of Manufacturing Processes 134: 739–748, Jan 2025 — https://doi.org/10.1016/j.jmapro.2024.12.064
  • Park, H. et al. (Mold Technology Team, Hyundai Mobis), fluid-assisted injection moulding study, Int. J. Adv. Manuf. Technol., 2016 (located via the OpenAlex affiliation sweep below).
  • OpenAlex affiliation sweep (34 works, methodology in §10) — https://api.openalex.org/works?search=injection%20molding&filter=raw_affiliation_strings.search:hyundai

Secondary Sources

  • MarkLines, Hyundai Mobis generative-AI model at CVPR 2025, Jun 2025 — https://www.marklines.com/en/news/328299
  • Chosun Biz (English), Hyundai Mobis launches MoAI, Oct 21, 2025 — https://biz.chosun.com/en/en-industry/2025/10/21/LJSD2GX4WFGVRLVRKWEMIYRVOQ/

Appendix A — Timeline

Siemens NX/Teamcenter selection (Nov 2021) → Simcenter ROM chassis case (Nov 2023) → NVIDIA partnership (Jan 2025) → Bayesian tailgate paper (Jan 2025) → Mobis at CVPR (Jun 2025) → ZER01NE Day (Sep 2025) → Mobis MoAI (Oct 2025) → NVIDIA AI factory, 50k GPUs (Oct 2025) → CuspAI (Nov 2025).

Appendix B — Glossary

  • HMG — Hyundai Motor Group: Hyundai Motor, Kia, Genesis (brand), Hyundai Mobis and other affiliates; a Korean family-led conglomerate (chaebol).
  • Reduced Order Model (ROM) — a small, fast stand-in model (here a neural network) trained on many runs of a full simulation, so new variants can be evaluated near-instantly.
  • Simcenter Amesim — Siemens's system-simulation tool (vehicle behaviour as connected 1-D physics models).
  • HEEDS — Siemens's design-space exploration/optimization tool; searches parameter combinations against targets.
  • Bayesian optimization — an optimization method that fits a statistical surrogate to sampled results and uses it to pick the next promising design; learns within one job, carries nothing between jobs.
  • Closure — a vehicle's openable panels (tailgate, doors, hood); tailgate inner structures are commonly injection-moulded composites with rib patterns.
  • Tier-1 — a supplier delivering complete modules directly to the OEM; Hyundai Mobis is HMG's captive Tier-1.
  • MoAI — "Mobis one AI," Hyundai Mobis's internal conversational generative-AI service (Oct 2025).
  • ZER01NE / CRADLE — HMG's open-innovation platform (creators + startups) and its global venturing offices.
  • Physical AI — HMG/NVIDIA framing for AI embodied in factories, robots and vehicles.
  • DFM — design for manufacturability: checking a design against the rules that make it mouldable (draft angles, wall thickness, undercuts…).

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

  1. "Genesis GV80 electric vehicle" is the Siemens blog's phrasing; Genesis's own electrified-model naming conventions differ ("Electrified GV70/G80"). Confirm the exact programme name against the blog/case-study before quoting in print.
  2. MoAI facts come from English-language Korean trade press (Chosun Biz, Maeil Business); the primary Mobis press release was not directly retrievable this session. Underlying LLM undisclosed.
  3. HMG volume ranking ("world's #3 automaker") and group shareholding structure — widely reported but not verified against a primary source here; also Boston Dynamics ownership percentages, the 42dot subsidiary and the AIRS AI lab were deliberately left unaudited (out of the moulded-parts scope).
  4. Mould-flow toolchain used by HMG/Mobis (Moldflow? Moldex3D? — both plausible in Korea) could not be established from public sources — a genuine blank worth asking any HMG contact.
  5. Patent sweep not performed for HMG/Mobis moulding or design-AI patents.
  6. The claim that Korean academia produces a large share of moulding-ML literature rests on an earlier research note of this study, not on a count performed in this research pass.
  7. The 2016 Mobis paper's full citation (exact title/DOI) should be pulled from the OpenAlex record before print; only the affiliation and venue were captured here.

Executive Summary

  • Who they are. Hyundai Motor Group is a Korean automotive conglomerate whose core affiliates are Hyundai Motor Company, Kia Corporation and the Tier-1 supplier Hyundai Mobis, led by Executive Chair Euisun Chung. It is widely ranked among the world's three largest automotive groups by vehicle volume. [Synthesis — see Appendix C for the unverified volume figure] 4
  • What they are automating. HMG's engineering-AI energy is demonstrably real and quantified — but it is aimed at vehicle-attribute optimization, materials discovery, enterprise assistants and "physical AI" infrastructure, not at moulded-parts engineering. The flagship number: a neural-network Reduced Order Model built from 200,000+ Simcenter Amesim simulation models cut Genesis GV80 electric-vehicle chassis parameter optimization from one week to 15 minutes, with each design evaluation dropping from 2 minutes to 0.1 seconds (Siemens blog, Nov 28, 2023). This is the strongest quantified OEM engineering-AI outcome found anywhere in this study's research on the automotive OEMs group. [Evidence] 2
  • The moulded-parts finding. After a deliberate affiliation sweep of the academic literature (34 injection-moulding works carrying Hyundai-family affiliations), exactly one is AI/optimization-era: a January 2025 Journal of Manufacturing Processes paper on Bayesian optimization of injection-moulded tailgate rib structures, co-authored by an engineer of HMG's Genesis Closure Engineering Design Team together with KAIST, Sogang University, Ulsan Technopark and the mould maker Q-JEN Mold Tech. HMG reaches for moulding-aware design optimization through academia and its supply chain, not through any internal AI tool — the single most instructive demand signal in this report. [Evidence] 912
  • The demand-side read. HMG has proven it will pay for surrogate-model speed-ups (chassis case), that it has an unmet need in moulding-constrained design optimization (tailgate paper), and that it runs formal open-innovation doors (ZER01NE, CRADLE). But moulded-parts execution sits with Hyundai Mobis and external mould makers — a moulding-AI initiative's realistic entry is via the supplier tier or the academic/open-innovation channel, not a direct OEM tool sale. [Synthesis] 69

  1. Hyundai Motor Company & Kia partner with Siemens (NX + Teamcenter as next-generation engineering platform), Nov 29, 2021 — https://press.siemens.com/global/en/pressrelease/hyundai-motor-company-and-kia-corporation-partners-siemens-digital-mobility ↩↩↩↩↩↩↩↩↩

  2. Siemens Simcenter blog, "Hyundai Motor Group uses AI to reduce the parameter optimization process from one week to 15 minutes," Nov 28, 2023 (200,000+ Amesim models; neural-net ROM + HEEDS; 2 min → 0.1 s per evaluation; 52 KPIs; 350+ parameters; Genesis GV80 EV chassis comfort/handling; quote from Ilsoo Jeong, Driving Comfort Virtual Development Team) — https://blogs.sw.siemens.com/simcenter/hyundai-motor-group-uses-ai-to-reduce-the-parameter-optimization-process-from-one-week-to-15-minutes/ ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩

  3. "Hyundai Motor Group Partners with NVIDIA to Accelerate Development of AI Solutions for Future Mobility," Jan 9, 2025 (Omniverse digital twins, Isaac robotics, AV simulation; Heung-Soo Kim quote) — https://www.hyundai.com/worldwide/en/newsroom/detail/0000000893 ↩↩↩↩↩↩↩↩↩↩↩↩↩

  4. NVIDIA newsroom, Hyundai Motor Group AI factory (50,000 Blackwell GPUs; ~US$3B; Physical AI Application Center; Euisun Chung quote), Oct 30, 2025 — https://nvidianews.nvidia.com/news/hyundai-motor-group-ai-factory ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩

  5. "Hyundai Motor Group and CuspAI Partner to Accelerate Material Innovation Using AI," Nov 6, 2025 (signed in Cambridge, UK; Chul Park quote; no material classes named) — https://www.hyundaimotorgroup.com/en/news/hyundai-motor-group-and-cuspai-partner-to-accelerate-material-innovation-using-ai ↩↩↩↩↩↩↩↩↩

  6. 2025 ZER01NE Day press release, Sep 16, 2025 (event Sep 17–21, Hyundai Seongsu Complex Hub, Seoul; 8 creator projects, 11 startups incl. 5 via CRADLE; generative AI / robotics / generative design themes) — https://www.hyundai.news/eu/articles/press-releases/2025-zer01ne-day.html ↩↩↩↩↩

  7. Chosun Biz (English), "Hyundai Mobis launches MoAI to boost R&D, quality, and sales efficiency," Oct 21, 2025 (in-house conversational generative-AI service connected to work systems) — https://biz.chosun.com/en/en-industry/2025/10/21/LJSD2GX4WFGVRLVRKWEMIYRVOQ/ ↩↩↩↩↩↩↩↩↩↩

  8. MarkLines, "Hyundai Mobis unveils generative AI model to recreate virtual driving environment at CVPR 2025," Jun 2025 — https://www.marklines.com/en/news/328299 ↩↩↩↩↩↩↩

  9. Lee, Yeo, Kong (Genesis Closure Engineering Design Team, Hyundai Motor Group / KAIST), Myeong, Jang (Q-JEN Mold Tech), Lee, Choi (Ulsan Technopark), Kim (Sogang Univ.), Ryu (KAIST), "Bayesian optimization of tailgate rib structures enhancing structural stiffness under manufacturing constraints of injection molding," Journal of Manufacturing Processes 134:739–748, Jan 2025 — https://doi.org/10.1016/j.jmapro.2024.12.064 ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩

  10. Park, H. (Mold Technology Team, Hyundai Mobis) et al., fluid-assisted injection-moulding residual-wall-thickness study, International Journal of Advanced Manufacturing Technology, 2016 — located via the OpenAlex affiliation sweep12; pull the full record before print (Appendix C item 7). ↩↩↩

  11. Fibre-orientation simulation papers, "Vehicle Development & Analysis Team, R&D Division, Hyundai Motor Co. & Kia Motor Corp.," Polymer Composites / J. Materials Processing Technology, 2002 — located via the OpenAlex affiliation sweep12. ↩

  12. OpenAlex works query: search "injection molding," filter raw affiliation contains "hyundai" (34 results; separate "kia" query, 4 results, none ML), retrieved Aug 2, 2026 — https://api.openalex.org/works?search=injection%20molding&filter=raw_affiliation_strings.search:hyundai ↩↩↩↩↩↩↩↩↩↩

  13. Q-JEN Mold Tech Co. — Incheon, South Korea (registry; identity with the KAIST/Hyundai paper unconfirmed). ↩

  14. University of Ulsan / Ulsan Technopark — Ulsan, South Korea — https://www.ulsan.ac.kr ↩

  15. Sogang University — Seoul, South Korea — https://www.sogang.ac.kr ↩

  16. KAIST — Daejeon, South Korea — https://www.kaist.ac.kr ↩