Skip to content

Landscape Note — Startups in AI for Automotive Moulded-Parts Engineering

Purpose and Method

This study's research identified nine startup-opportunity areas; this note maps the startup landscape onto them, answering the follow-up question for each: who is already trying? Startups (and a few relevant recent acquisitions and failures) are organized by opportunity area, each with an honest-AI verdict — does it learn from data, or compute from rules/physics?

Method note: the seed list was verified and extended via web search and direct company-site checks on 2026-08-01; some funding figures rest on aggregator snippets and are flagged. Prior adversarially-verified research from this study's corpus (the deployed-AI report and the SIMCON/Autodesk/Hexagon company reports) supplies several verdicts and is cited where used. [Synthesis]

Research caveat (Aug 2026): this sweep ran under search constraints; coverage of smaller entrants may be incomplete.

1. The Benchmark Case — Emmi AI (now Mistral AI)

The one startup that fully executed the study's thesis, and the yardstick for everything below. Fully documented in the SIMCON report; summary only:

  • Founded Dec 2024 in Linz (JKU ecosystem; CSO Johannes Brandstetter); raised a €15M seed in Apr 2025 — the largest Austrian seed to date. [Evidence] 12
  • Feb 10, 2026: announced NeuralMould, a transformer ("AB-UPT" variant) neural surrogate for injection moulding, trained on 1M+ simulation trajectories from partner SIMCON; shipped Mar 18, 2026 as the Cadmould AI Solver research preview (filling only). Passes the honest-AI test outright. [Evidence] 35
  • May 19, 2026: acquired by Mistral AI — ~18 months from founding to exit, before the moulding product reached general availability. [Evidence] 6

What it proves: (a) a small team can build a genuinely learned moulding solver if an incumbent supplies the physics engine and training corpus; (b) the exit path is real and fast; (c) the acquirer was an LLM company, not a CAE vendor — moulding-domain AI talent is being valued as generic physics-AI talent. Whether NeuralMould survives inside Mistral is publicly unanswered — the moulding-surrogate seat may be vacant again. [Synthesis]

2. Area 1 — Learning from Production Outcomes

The most populated area — but all activity is machine/process-side (sensors, telemetry, per-mould trials). Nobody learns from geometry + outcome history across a part library, which is the actual opportunity.

  • eMoldino (Seoul, South Korea). IoT sensors on 100,000+ moulds/tooling assets streaming shot-by-shot telemetry into monitoring, predictive- maintenance and supplier-performance analytics; wireless sensor co-developed with Samsung (2017). Named users include Continental, PACCAR, Honeywell, Eaton, Dyson — real automotive-supplier traction. [Evidence] 8 Honest-AI verdict: tooling telemetry analytics; the depth of learned ML behind "predictive" claims is not publicly evidenced. It watches moulds, not designs — no geometry in the loop. [Inference]
  • STASA QC (Steinbeis spin-off, Germany; distributed by Kistler). "AI + machine learning" process setup: systematic DoE plus model fitting from quality data to find optimal machine settings. [Evidence] 9 Verdict: statistical models fitted per mould, per trial — genuine data-fitting at small scale, same class as SIMCON's Varimos Real; carries nothing between projects. [Synthesis] 35
  • Symate / Detact (Germany). AI platform correlating process-chain and quality data; case studies in plastics, injection moulding, die casting, automotive lightweighting. [Evidence] 10 Verdict: plant-data ML, no geometry. [Inference]
  • plus10 (Augsburg, Fraunhofer spin-off). Self-learning shopfloor assistance (SHANNON) on high-frequency machine data; explicitly targets faster commissioning and less waste on injection-moulding systems; Siemens ecosystem partner; claims 5–15% OEE (overall equipment effectiveness) gains. [Evidence + Marketing] 11
  • Moldsonics (Linz, founded 2021, JKU polymer-institute spin-off — same ecosystem as Emmi AI). In-mould ultrasonic sensing of flow front, solidification and wear. [Evidence] 12 Verdict: sensing infrastructure, not ML — but exactly the kind of in-process signal an outcome-learning product would need. Potential partner/acquisition target rather than competitor. [Inference]
  • † iMFLUX (Ohio, P&G subsidiary, 2013–2023) — the cautionary tale. P&G's adaptive melt-pressure process control; shut down June 30, 2023 (122 laid off), technology moved to royalty-free licensing through machine vendors. P&G judged it "not scalable as an investment." [Evidence] 1314 Lesson: even superb process technology with a deep-pocketed parent failed as a business selling process control into moulders — a warning for any Area-1 model that monetizes at the machine rather than in engineering. [Inference]

Area verdict: crowded with small process-side players; empty where the real opportunity points — learning defect/warpage/scrap outcomes against part geometry across a library. [Synthesis]

3. Area 2 — Geometry Retrieval / Design Reuse

  • Physna (Cincinnati, founded 2015; ~$86M raised incl. a $56M Series B — Tiger Global, Sequoia, GV). The flagship geometric-search startup ("world's leading geometric search engine"; Thangs consumer platform). [Evidence — funding via aggregator snippets, verify] 15 Strategic pivot: in Jan 2026 it unveiled "Physical AI Search" and an early-access program for AI companies to train models on 3D product data — drifting from enterprise search toward 3D-data supply for AI labs. [Evidence — search-snippet sourced; primary URL not captured, verify] Verdict: general-purpose geometric search is real and funded, but there is no moulding vertical, no CATPart-native offering, no DFM linkage — the study's specific retrieval thesis (search your own mould history, return the tooling/process knowledge attached) remains unclaimed. [Synthesis]
  • CADENAS / 3Dfindit (Augsburg, est. 1992) — incumbent, not a startup: geometric similarity search over supplier catalogue parts, not a customer's own history. Context only. [Synthesis]
  • Backflip AI (US; founded by Markforged founders Greg Mark and David Benhaim; $30M Series A, NEA + a16z). Foundation model turning 3D scans, photos and text into editable CAD (.STEP). [Evidence] 16 Verdict: genuine learned geometry AI, adjacent to retrieval (it creates CAD, doesn't search it); a capability marker that geometry foundation models are venture-fundable. [Inference]
  • Spread AI (Berlin). "Engineering ontology" unifying requirements, design, simulation, test and field data; AI agents over it (Product Explorer, Error Inspector). Customer logos: VW, Mercedes-Benz, BMW, Audi, Ford, Daimler Truck, Bosch, Rheinmetall; 40+ connectors including CATIA and Teamcenter. [Evidence] 17 Verdict: the knowledge half of the retrieval thesis built CAD-agnostically at OEM scale — but document/data-graph-centric, not geometric. The strongest adjacent player: a geometry-retrieval startup might end up as a Spread feature, or its partner. [Inference]

Area verdict: lukewarm — funded generalists exist; the moulding-specific, CATPart-native, knowledge-attached version has no startup. [Synthesis]

4. Area 3 — Learned DFM from Geometry

  • Nobody found. No startup ships a learned manufacturability check for moulded parts. [Synthesis — from this study's research corpus] 18
  • The near-misses are the quoting engines (see §11): aPriori's engine is explicitly physics/rules ("Digital Factories", 440+ process models); Protolabs ProDesk (Feb 2026) markets "AI-driven DFM" but its engine is unverified; Xometry genuinely uses ML trained on millions of manufacturing data points — but for price/lead-time prediction, with DFM feedback remaining geometric rules. [Evidence] 1819

Area verdict: empty. The strongest whitespace signal in this note, especially combined with the Creo 13 vendor-appetite signal from this study's research. [Synthesis]

5. Area 4 — Warpage-First Neural Surrogates

The generic ML-CAE surrogate category is the best-funded corner of engineering AI — but no startup has claimed moulding since Emmi's exit.

  • Neural Concept (Lausanne, EPFL spin-out). Learned 3D deep-learning CAE surrogates; $27M Series B (Jun 3, 2024, Forestay); $100M Series C (Dec 18, 2025, Growth Equity at Goldman Sachs). [Evidence] 20 Verified automotive use: GM (crash), MAHLE (EV blower: +15% efficiency, −4 dB), JLR, Renault, Subaru — and, notably for this study, moulded- parts Tier-1s Antolin, OPmobility and Kautex as customers. [Synthesis] 21 Two of its own marketing claims (PINN (physics-informed neural network) framing; a GM "co-pilot") were refuted in this study's verification research — do not cite. [Synthesis] 22 Verdict: real learned ML, deployed in automotive — but on structures/CFD/thermal. No moulding physics product. The single most capable company to move into this area. [Inference]
  • PhysicsX (London). "Large Physics Models"; $135M Series B (Jun 22, 2025, led by Atomico — Siemens and Applied Materials participating) plus a Nov 19, 2025 NVentures extension to >$155M at ~$1B valuation. [Evidence] 23 Automotive aero models (PXNetCar) with honestly-reported out-of-distribution degradation; no named OEM production deployment, no moulding. [Evidence] 24
  • Luminary Cloud (US). "Physics AI Model Factory"; SHIFT-SUV automotive aero foundation model built with Honda; NVIDIA GeoTransolver architecture; $72M Series B (N47, Sutter Hill, NVentures). [Evidence] 25
  • Monolith AI (London). ML on engineering test data; £8.5M Series A (2021, Insight Partners); vendor-claims BMW, Honda, BAE, Siemens as clients — the only detailed public case is Rolls-Royce (aerospace); treat the automotive names as logo-level. [Evidence + Marketing] 26
  • † Navasto (Berlin, founded 2014; AI aero surrogates for VW, Audi, GM, Airbus) — acquired by Autodesk (announced Dec 10, 2024). The talent behind Autodesk's moulding-AI ambitions now sits inside a platform vendor. [Evidence] 27

Area verdict: hot money, generic focus. Warpage-first moulding surrogacy has zero dedicated startups — Emmi held the seat and was bought out of it. The window this study's research calls "open but short" is, today, open. [Synthesis]

6. Area 5 — Independent Validation / Benchmarking

Nobody. No startup, consultancy-as-product, or lab offers independent accuracy validation of neural moulding (or even neural CAE) solvers; every accuracy number in this study is vendor-published, and SIMCON's downloadable benchmark set remains the only public artefact. [Evidence of absence] 5 Zero competition; the open question is whether it's a venture-scale business or a niche service. [Inference]

7. Area 6 — Metrology-Feedback Intelligence

No startup closes measure→learn→correct for moulded parts (deviation maps → tool-correction suggestions). Hexagon owns the data pipes and ships no learning product. [Evidence of absence] 33 Closest adjacents: Moldsonics (in-mould sensing, §2) and the vision-QA players (§12) — all measure, none feed corrections back into tooling or design. [Synthesis]

8. Area 7 — Cold-Start / Pre-Trained Moulding Models

  • Emmi's NeuralMould was exactly this — a pre-trained model claiming validation on "1000+ tested real company products" (vendor claim, thin) [Marketing] 4 — and it is now inside Mistral, status unknown. [Evidence] 6
  • Luminary's SHIFT models prove the pattern commercially in the adjacent domain: domain-specific foundation physics models (aero SUV) sold to users without their own simulation archives. [Evidence] 25

Area verdict: pattern proven, moulding instance in limbo — one credible (wildcard) occupant, no independent startup. [Synthesis]

9. Area 8 — Assistant-Native Moulding Copilots

  • Zoo (formerly KittyCAD) (Inglewood, CA). AI-native CAD (Design Studio) plus Zookeeper, a "conversational CAD agent" advertising "manufacturing-aware feedback," and ML/CAD APIs. [Evidence + Marketing] 28 CAD-general; nothing moulding- or simulation-results-aware. [Inference]
  • Spread AI's agents (§3) answer engineering questions over enterprise data — the nearest shipped "engineering copilot" with automotive OEM logos, but not a moulding-results reader. [Evidence] 17
  • Foundation EGI (LLM-based engineering copilot, reportedly MIT-linked) — could not be verified in this research pass; watch-list (§13).

Area verdict: copilot startups exist for CAD and for engineering data; none reads moulding simulation/process results. Empty in the specific sense that matters. [Synthesis]

10. Area 9 — Open / On-Prem / Multi-CAD AI Tooling

No startup occupies the counter-position. Emmi open-sourced architectures (AB-UPT) but not trained moulding weights [Evidence] 7; every surrogate startup in §5 sells closed SaaS and the platform vendors are sovereign-cloud/ecosystem-locked. [Synthesis] Zoo is the only startup with a partially open, API-first posture — but not in moulding. [Inference]

11. Category Note — DFM / Cost / Quoting Incumbents and Marketplaces

One paragraph, per scope: aPriori (the incumbent to beat; physics/rules Digital Factories, explicitly not learned ML — "AI" branding is layered on top)18; Tset (Vienna; cost + CO₂ should-cost (bottom-up cost estimation) with named automotive users BMW, Brose, KTM, Bosch Rexroth; site shows no learned-ML claims — model-based costing)29; Paperless Parts (Boston; job-shop quoting — its process list omits injection moulding entirely, itself a datum)30; marketplaces Xometry (learned ML for pricing, rules for DFM)19, Protolabs (ProDesk, Feb 2026 — "AI-driven DFM" marketing, engine unverified) and Fictiv (no verified claims survived prior research)18. Net: this category monetizes quoting, computes with rules/physics, and learns (at most) prices — it does not threaten Areas 3 or 4 technically, but owns the customer relationship a DFM startup would want. [Synthesis]

12. Category Note — Vision QA on Moulded Parts (outside the nine)

A genuinely active startup category the nine areas don't cover: Maddox AI (Germany) publishes an injection-moulded-parts inspection case (four-camera surface-defect detection)31; Inspekto (Heilbronn; self-learning "QA-in-a-box" since 2018, with an in-line injection-moulding use case) was acquired by Siemens in Feb 202432. These learn from images at the end of the line — they do not touch design or engineering. Relevant to the study mainly as (a) proof that moulded-parts ML businesses can be built and exited, and (b) a potential data source for Area ⅙ feedback loops. [Synthesis]

13. Watch-List / Unverified

Leads seen but not verified (see the research caveat under Purpose and Method) — do not cite until checked: eMoldino funding history (PitchBook profile exists, amounts unseen); Physna's Jan 2026 "Physical AI Search" primary source; Foundation EGI (LLM engineering copilot); nTop and Leap71 (computational design — no moulded-parts evidence found, likely out of scope); SimScale / Flexcompute AI features; Symate funding/size; Monolith's BMW/Honda work beyond logos. [Synthesis]

14. Heat Map — Crowded vs. Empty

# Opportunity area Startups active Heat
1 Learning from production outcomes eMoldino, STASA/Kistler, Symate, plus10, (Moldsonics); † iMFLUX Warm process-side; empty geometry→outcome
2 Geometry retrieval / design reuse Physna (generic, pivoting), Backflip (adjacent), Spread (knowledge, not geometry) Lukewarm; moulding vertical empty
3 Learned DFM from geometry none (quoting engines are rules/physics) Empty
4 Warpage-first neural surrogates Emmi→Mistral (filling, in limbo); Neural Concept / PhysicsX / Luminary / Monolith (generic, no moulding) Hot money, vacant seat
5 Independent validation / benchmarking none Empty
6 Metrology-feedback intelligence none (Moldsonics/vision-QA adjacent) Empty
7 Cold-start pre-trained moulding models NeuralMould (inside Mistral); SHIFT pattern proven in aero Sparse
8 Assistant-native moulding copilots Zoo, Spread (neither moulding-aware) Sparse
9 Open / on-prem / multi-CAD AI none Empty

The absences: Areas 3, 5, 6 and 9 have literally no startup; Area 4's only occupant was acquired away. These are the strongest whitespace signals — with the standard caveat that an empty area can mean either "unnoticed" or "no buyer." Area 5 in particular may be empty because it isn't venture-scale. [Synthesis + Inference]

15. What This Means for a New Entrant

  1. The moulding-surrogate seat is vacant again. Emmi proved the build and the exit — built and shipped in ~15 months, exited in 18 (one incumbent partner, €15M); Mistral may or may not pursue moulding. A warpage-first entrant partnering with a solver vendor (Moldex3D- or Moldflow-tier, per this study's opportunity analysis) replays a proven playbook — consistent with the external-entrants note's warning against going frontier-scale alone (see the external AI entrants note). [Inference]
  2. Incumbents buy, not build. Emmi→Mistral, Navasto→Autodesk, Inspekto→Siemens within ~28 months: platform vendors and AI labs acquire domain-AI teams early. Good for exits; bad for long independent runways — the competitive clock in Areas 4/7 is set by acquisition, not by product. [Synthesis]
  3. Geometry+outcome learning (Area 1's real version) has no competitor — but iMFLUX warns against monetizing at the machine. Sell into engineering (design/tooling decisions), source outcomes via partners like the sensing and vision-QA players. [Inference]
  4. Retrieval whitespace is narrower than it looks but still real. Physna is drifting toward 3D-data supply; Spread owns the OEM knowledge layer without geometry. The defensible position is the moulding-vertical combination — geometric search over a company's own CATPart/mould history with attached tooling/process knowledge — which nobody ships. [Synthesis]
  5. Don't fight the quoting incumbents for DFM distribution — aPriori and the marketplaces own the workflow; a learned-DFM engine may be worth more as their component (or channel play) than as a rival product. [Inference]
  6. Funding climate is favourable: more than $325M went into physics-AI surrogate startups in 2025 (PhysicsX $155M, Neural Concept $100M, Luminary $72M) — with adjacent raises continuing into 2026 (Synera, an engineering-automation platform, $40M in Apr 2026). Investors have priced the category; moulding is an unclaimed vertical within it. [Synthesis] 23202534

Challenges to the nine areas worth noting: the active-but-uncovered vision-QA category (§12); the process-side cluster in Area 1 suggesting the area should be split (machine-side learning = crowded; geometry-side = empty); and Spread AI as evidence that the knowledge-graph half of Area 2 is being solved CAD-agnostically at OEM level. [Synthesis]

Footnotes

All URLs verified accessible 2026-08-01 unless noted.

  1. Emmi AI, "Building the Frontier Lab for Industrial Engineering" (founded Dec 2024; Brandstetter CSO) — https://www.emmi.ai/news/building-the-frontier-lab-for-industrial-engineering ↩

  2. Emmi AI €15M seed, Apr 25 2025 (3VC, Speedinvest, Serena, PUSH) — https://www.emmi.ai/news/emmi-ai-raises-eur-15m-to-bring-ai-to-the-heart-of-industrial-engineering ↩

  3. Emmi AI, "NeuralMould: Our First Digital Engineer for Injection Moulding," Feb 10 2026 — https://www.emmi.ai/news/neuralmould-our-first-digital-engineer ↩

  4. Emmi AI NeuralMould model page ("consistent 5% relative errors," "1000+ tested real company products") — https://www.emmi.ai/models/neuralmould ↩

  5. SIMCON Cadmould AI Solver launch, Business Wire, Mar 18 2026 — https://www.businesswire.com/news/home/20260318680159/en/SIMCON-Unveils-Worlds-First-Large-Engineering-Model-for-Plastic-Injection-Moulding ↩↩

  6. "Mistral AI Acquires Emmi AI," May 19 2026 — https://www.emmi.ai/news/mistral-ai-acquires-emmi-ai ↩↩

  7. Emmi AI AB-UPT open-source repository — https://github.com/Emmi-AI/anchored-branched-universal-physics-transformers ↩

  8. eMoldino homepage (100k+ assets; Dyson/Eaton/Continental/Honeywell/PACCAR references), accessed 2026-08-01 — https://www.emoldino.com; HQ Seoul per Crunchbase profile — https://www.crunchbase.com/organization/emoldino ↩

  9. Kistler STASA QC product page (AI/ML + DoE process optimization) — https://www.kistler.com/INT/en/cp/software-stasa-qc-optimize-2820b/P0001388; STASA — https://stasa.de/stasaqc/index_en.html ↩

  10. Detact by Symate GmbH — case studies in plastics/injection moulding/automotive — https://www.detact.com/en/case-studies/ ↩

  11. plus10 GmbH, "About us" (Fraunhofer spin-off; injection-moulding commissioning/waste claims) — https://en.plus10.de/about-plus10; Siemens ecosystem listing — https://www.siemens.com/en-gb/ecosystem/plus10/ ↩

  12. Moldsonics (Linz, founded 2021; inline ultrasonic sensing) — https://www.moldsonics.com/en/ ↩

  13. Cincinnati Business Courier, "Procter & Gamble closes iMFLUX subsidiary," May 2 2023 — https://www.bizjournals.com/cincinnati/news/2023/05/02/procter-gamble-closes-imflux-subsidiary.html ↩

  14. Plastics Technology, "iMFLUX Goes Royalty-Free and Transitions to Mainly R&D Role," Apr 2023 — https://www.ptonline.com/news/imflux-goes-royalty-free-and-transitions-to-mainly-rd-role ↩

  15. Physna homepage — https://physna.com; funding ($56M Series B, ~$86M total; Tiger Global/Sequoia/GV) per aggregator search snippets, 2026-08-01 — verify against primary press before citing onward. ↩

  16. Backflip AI — https://backflip.ai; $30M raise and Markforged-founder provenance: 3D Printing Industry, "Markforged Founders Launch New AI 3D Model Generator Backflip with $30M Funding" — https://3dprintingindustry.com (article surfaced in search 2026-08-01). ↩

  17. SPREAD AI homepage (engineering ontology; VW/Mercedes/BMW/Audi/Ford/Daimler Truck/Bosch logos; CATIA/Teamcenter connectors), accessed 2026-08-01 — https://www.spread.ai ↩↩

  18. This study's research corpus (unpublished working notes; adversarially verified: aPriori = physics/rules Digital Factories; Protolabs ProDesk Feb 2026; Xometry/Fictiv claims unverified), 2026. ↩↩↩↩

  19. Xometry, "Machine Learning for Manufacturing" — https://www.xometry.com/machine-learning-for-manufacturing/; "Trained on millions of manufacturing data points" per Xometry instant-quoting page — https://xometry.in/instant-quoting-engine/ ↩↩

  20. Neural Concept $100M Series C, Dec 18 2025 (Growth Equity at Goldman Sachs Alternatives) — https://www.neuralconcept.com/press-release/neural-concept-closes-100m-funding-round; Series B $27M, Jun 3 2024 (Forestay) per funding-press search results 2026-08-01. ↩↩

  21. Verified deployments (GM crash; MAHLE +15%/−4dB; Antolin, OPmobility, Kautex as customers): this study's research corpus (unpublished working notes) incl. https://www.neuralconcept.com/post/neural-concept-accelerates-integrated-ai-adoption-with-100-enterprise-growth ↩

  22. Refuted Neural Concept marketing claims (PINN framing; GM "co-pilot") — this study's research corpus (unpublished working notes). ↩

  23. PhysicsX $135M Series B, Jun 22 2025 — https://www.physicsx.ai/newsroom/physicsx-raises-135m-series-b-to-usher-in-a-new-era-of-ai-native-engineering-and-manufacturing; NVentures extension, Nov 19 2025 — https://www.physicsx.ai/newsroom/physicsx-announces-extension-to-series-b-round ↩↩

  24. PhysicsX automotive-aero methods announcement (PXNetCar; OOD degradation figures) — https://www.physicsx.ai/newsroom/scaling-physics-ai-for-automotive-aerodynamics ↩

  25. Luminary Cloud SHIFT models (SHIFT-SUV with Honda) — https://luminary.ai/resources/introducing-luminary-shift-models; $72M Series B — https://luminary.ai/resources/luminary-cloud-secures-72m-series-b ↩↩↩

  26. Monolith AI Series A release (clients "BMW, Honda, BAE Systems and Siemens" — vendor-claimed) — https://www.monolithai.com/press-release/monolith-series-a-funding; Rolls-Royce case — https://monolithai.com/post/surrogate-model-accuracy-rolls-royce ↩

  27. Autodesk, "Welcome NAVASTO," Dec 10 2024 — https://blogs.autodesk.com/design-studio/2024/12/10/welcome-navasto/ ↩

  28. Zoo (formerly KittyCAD) — Design Studio, Zookeeper agent, ML APIs, accessed 2026-08-01 — https://zoo.dev ↩

  29. Tset homepage (Vienna; BMW, Brose, KTM, Bosch Rexroth among customers), accessed 2026-08-01 — https://www.tset.com ↩

  30. Paperless Parts homepage (process list without injection moulding), accessed 2026-08-01 — https://www.paperlessparts.com ↩

  31. Maddox AI, "Inspection of Injection Molded Parts" success story — https://www.maddox.ai/en/ ↩

  32. Photonics Media, "Siemens Acquires Inspekto," Feb 2024 — https://www.photonics.com/Articles/Siemens-Acquires-Inspekto/a69757; Siemens Inspekto page incl. injection-moulding in-line QA use case — https://www.siemens.com/en-us/company/artificial-intelligence/industrial-ai/inspekto-ai-inspection/ ↩

  33. This study: A_companies/1_platform_vendors/1_07_hexagon.md §19 (owns PC-DMIS/Q-DAS data pipes, ships no learning product). ↩

  34. Synera $40M Series B, Apr 14 2026 (Revaia; BMW i Ventures, Capgemini/ISAI, UVC; deployed at NASA, BMW, Airbus, Volvo Trucks, Hyundai) — https://www.synera.ai/press; TNW coverage — https://thenextweb.com/news/synera-40m ↩

  35. Why DoE/meta-model fitting is classical optimization, not modern learned AI — concept note What is Optimization technology. ↩