Why Not Incumbents
Analyst note
Four of the world's largest manufacturing technology companies — Autodesk, Sandvik, Hexagon, and Siemens — independently evaluated building autonomous CAM. All four concluded it was not viable and chose partnership with or investment in CloudNC instead. The universal kernel provider, ModuleWorks (90% market share in CAM toolpath algorithms), has no AI team, no AI roadmap, and a business model that shares everything it builds with every competitor simultaneously. Multiple industry executives have told CloudNC they do not expect ModuleWorks to solve this. The question is not whether incumbents will try to build this. They already tried. They chose CloudNC.
They already tried. The economics are highly contained for any single vendor. They don't own the technology they'd need. And the market structure means only a cross-platform player can win.
1. The Ones Who Tried
Several incumbents have attempted AI-driven CAM. All have failed or retreated.
| Who | What They Did | What Happened |
|---|---|---|
| Autodesk | The most technically capable CAM vendor, with $246M in annual CAM revenue across Fusion, PowerMill, and FeatureCAM. Ran internal research attempts at AI-driven CAM. | Saw that CloudNC was solving it and led CloudNC's Series B instead. Their 2026 Fusion roadmap lists "AI-Assisted CAM" as automatic hole recognition and pattern matching. For AI quoting, they partnered with Toolpath Labs, a startup. Autodesk is outsourcing the hard part. |
| Sandvik (Mastercam) | The largest CAM vendor by brand. Launched "Prism" in partnership with ModuleWorks as next-generation touch-based CNC programming. | Prism was killed.1 Sandvik's strategy pivoted to distribution consolidation: acquired CNC Software (Mastercam) in 2024, buying US resellers through 2025. Their CTO publicly stated the target for taking Mastercam to the cloud is 2030. Their "Manufacturing Copilot" is a Microsoft Azure OpenAI wrapper that helps users navigate menus. |
| Sandvik (again) | Launched "Manufacturing Copilot" across GibbsCAM, Cimatron, and SigmaNEST in 2024. | An LLM chatbot inside the software. It explains features and links to documentation. It does not generate toolpaths. This is autocomplete, not autonomy. |
| Hexagon | The second-largest CAM vendor by revenue ($253M). Launched "ProPlanAI" as their AI CAM offering. | ProPlanAI was killed. Hexagon subsequently partnered with CloudNC.2 |
| Autodesk (Copilot) | Shipped "AI-Assisted CAM" features in Fusion. | Automatic hole recognition. Feature-based pattern matching. These are useful workflow accelerators. They are not autonomous toolpath generation for complex 3D geometry. The gap between recognising a hole and generating a full machining strategy for a freeform aerospace part is the entire problem. |
"It would take $100 million and 7 years for Autodesk to build this."
Carl Bass — Former CEO, Autodesk. CloudNC investor.
2. Every Layer is Disincentivized
The CNC manufacturing value chain has five layers. Each has a structural reason to resist autonomous CAM.
| Layer | Who | Why They Won't |
|---|---|---|
| Machine Tool OEMs | DMG MORI, Mazak, Haas, Okuma | Revenue scales with machines sold. AI-driven utilization gains mean fewer machines purchased. They also make neither the controllers nor the software. |
| Controller Manufacturers | Siemens (SINUMERIK), Fanuc, Mitsubishi, Heidenhain | They don't sell CAM software and don't have any CAM R&D capability. Controllers are a different technical domain entirely: real-time motion control, not toolpath strategy. |
| Toolpath Kernel | ModuleWorks (Aachen) | 250 employees, $24M revenue, ~90% kernel market share. They are the intelligence layer, but they are a component vendor. Their business model is licensing the same kernel to everyone simultaneously — anything they build gets shared with every competitor. Multiple CAM vendor executives have told CloudNC they do not expect ModuleWorks to solve autonomous CAM. No AI roadmap. No AI team. No end-user product. |
| CAM Software Vendors | Mastercam, Autodesk, Siemens NX, SolidCAM, hyperMILL, etc. | Most license core algorithms from ModuleWorks. Software architectures are decades old. Teams primarily sell and maintain. One major release per year at best. |
| CAM Resellers | Regional distributors | Post-processor revenue disappears with automation. Training revenue at risk. Their business model depends on the software being complex enough to require integration services. |
3. The Intelligence is Rented, Not Owned
The CAM industry's core toolpath algorithms are not proprietary to the vendors who sell the software.
ModuleWorks (Aachen, Germany) licenses toolpath generation algorithms to over 60% of the world's CAM vendors. In the CAD/CAM kernel market, they hold approximately 90% share. Revenue: $24M. Employees: 250. They power over 500,000 installed seats.
This means: Mastercam, Siemens NX, GibbsCAM, Cimatron, Edgecam, ESPRIT, and dozens of others all run the same underlying toolpath engine. The competitive differentiation between CAM vendors is largely in the user interface, post-processor library, and sales channel. The intelligence is shared.
To build autonomous CAM, a vendor would need to either (a) convince ModuleWorks to build AI capabilities, or (b) build their own toolpath intelligence from scratch.
Option (a) is structurally difficult. ModuleWorks' business model is to license the same kernel to everyone simultaneously: anything they build gets shared with every competitor. They have no AI team and no published AI roadmap. Their 2025 releases are incremental kernel updates. Industry executives from multiple CAM vendors have briefed CloudNC that they do not expect ModuleWorks to solve this.
Option (b) is what CloudNC did. It took a decade.
4. Only a Cross-Platform Layer Wins
Even if a single vendor built autonomous CAM, they would only serve their own installed base. The market is too fragmented for any one vendor to capture it.
CAM market revenue share (CIMdata, 2022):
| Vendor | Revenue ($M) | Share |
|---|---|---|
| Hexagon (Edgecam, ESPRIT, VISI, Radan) | $253M | 14.9% |
| Dassault Systèmes (CATIA/DELMIA) | $249M | 14.6% |
| Autodesk (Fusion, PowerMill, FeatureCAM) | $235M | 13.8% |
| Sandvik (Mastercam, GibbsCAM, Cimatron) | $220M | 12.9% |
| Siemens (NX CAM) | $189M | 11.1% |
| Open Mind (hyperMILL) | $96M | 5.6% |
| Everyone else (54+ vendors) | $460M | 27.0% |
Five vendors between 11% and 15%. Nobody above 15%. The long tail (54+ smaller vendors) accounts for 27% of revenue. This is a structurally fragmented market.
In installed seats, the picture is even more dispersed: 2.4 million industrial seats across 60+ CAM providers. Even the largest vendor serves only a fraction of the world's machinists.
Critically, switching costs between traditional CAM packages are extremely high. Each package requires its own trained operators and its own post-processor library (the machine-specific translation layer between CAM output and CNC controller). A shop that has spent years training machinists on Mastercam and building post-processors for their specific machines is not going to switch to Fusion because Autodesk built a better AI. The installed base is locked in. An incumbent's AI CAM would only reach their own captive users.
CloudNC sits on top of the existing CAM package. No retraining. No new post-processors. The machinist stays in the software they already know. That is why only a cross-platform layer can address the full market.
An incumbent building autonomous CAM would automate at most 15% of the market. The other 85% would remain manual.
5. Why Not Acquire?
If incumbents cannot build it, the natural question is: why not buy a startup that has?
The short answer: there is very little to buy. The competitive landscape contains five identifiable competitors. All rely on supervised machine learning trained on historical machining data. The structural limitation is not just part similarity — the model must have seen the specific combination of part geometry, tooling, and factory setup before. This limits every competitor to months of on-premise deployment, training on a narrow segment of the market with self-similar parts. None have built their own CAM kernel. None operate at CloudNC's scale.
An acquisition would also not solve the cross-platform problem. Autodesk acquiring an AI CAM startup would give Autodesk users autonomous CAM. Mastercam users, Siemens NX users, and the 54+ other CAM packages would remain manual. The acquirer captures at most their own installed base — the same 11-15% ceiling.
This is why four of the largest incumbents have chosen to partner with CloudNC rather than acquire a competitor: the cross-platform model serves their users without requiring them to build (or buy) it themselves.
Summary
Four of the world's largest manufacturing technology companies independently evaluated whether to build autonomous CAM. All concluded it was not viable and chose to partner with or invest in CloudNC instead.
The startup landscape offers no credible acquisition targets — every competitor uses supervised ML, none have built their own CAM kernel, and an acquisition would be locked to a single vendor's installed base anyway.
The question is not whether incumbents will try. They already did. The question is whether the structural incentives have changed. They have not.
Sources:
CIMdata CAM Market Analysis Report, 2023 edition (CY2022 data). Revenue figures are direct provider revenues. Market share based on total direct CAM revenues of $1.70B.
1 Sandvik Prism cancellation: private briefing between Sandvik and CloudNC executives.
2 Hexagon ProPlanAI cancellation: private briefing between Hexagon executives and CloudNC.