Competitive Landscape

Home / Company / Competitive Landscape

Understanding where others have succeeded and failed shapes our approach. The CNC machining market has seen multiple attempts at automation and platform plays — each one teaches us something about what works and what doesn't.

Market Map

graph TD
  subgraph Marketplaces
    XOM[Xometry]
    FIC[Fictiv]
  end
  subgraph "Automated Manufacturing"
    HAD[Hadrian]
    PLE["Plethora ❌"]
    FRM["Formlogic ❌"]
  end
  subgraph "Software Tools"
    CNC[CloudNC]
    TP[Toolpath]
  end
  subgraph "Legacy Automated"
    PRO[Protolabs]
  end
  MAC[Anvil] --> |"AI-CAM + Vertical Integration"| MAC

How They Compare

Company	Model	What They Proved	Their Lesson for Us
Xometry / Fictiv	Marketplace (third-party shops)	Massive demand for fast prototyping	Aggregation alone doesn't fix the CAM bottleneck
Protolabs	Owned facilities, constrained geometry	Automation works when scope is narrow	Restricting geometry caps your market — we need AI that handles complexity
Hadrian	Vertical integration, defense focus	Software-defined factory is viable	Orchestration without CAM automation still requires expert labor per part
CloudNC / Toolpath	Software sold to shops	AI-CAM technology is real	Selling into heterogeneous environments kills adoption — control your own stack
Plethora	Vertical integration, broad scope	Full-stack vision was correct	Don't build facilities before software can handle the work
Formlogic	Custom hardware, scale-first	Hardware automation has limits alone	Prove unit economics on one cell before scaling

Deeper Look

MarketplacesProtolabsHadrianCloudNC / ToolpathFailed Attempts

Xometry and Fictiv proved that hardware teams are desperate for faster, easier access to CNC parts. They built demand aggregation platforms that route orders to third-party job shops.

The problem: they don't own machines, don't control quality, and don't generate manufacturing data. Lead times are stuck at 7-10 days because the underlying shops still use manual CAM programming. They're structurally disincentivized to build owned manufacturing — their business model depends on shop network effects, not production efficiency.

What we take from this: The demand is real. But you can't fix manufacturing from the demand side alone.

Protolabs automated CAM early — but only for simple 2.5D and 3-axis parts. Their automation works precisely because they restrict the geometry they'll accept. That caps their addressable market to basic prototypes and simple parts.

They can't handle complex geometries, tight tolerances, or 5-axis work. The segment they serve is real but limited.

What we take from this: Automation works when you constrain scope. But constraining geometry forever means you never reach the larger market.

Hadrian is building software-defined factories for aerospace and defense. They've raised significant capital and have real facilities. Their center of gravity is factory orchestration — scheduling, workflow, MES.

But CAM still involves meaningful manual labor. They employ CAM programmers and CMM programmers. Their optimization is for defense procurement cycles, not rapid prototyping.

What we take from this: Vertical integration works. But orchestration without CAM automation still requires expert humans per part.

CloudNC (now Toolpath) built strong AI-CAM technology. The core tech is real — automated feature recognition, toolpath generation, optimization.

They tried selling it to heterogeneous job shops. Nearly impossible to generalize across hundreds of different machines, tool libraries, fixtures, and materials. Long sales cycles, heavy implementation overhead per customer.

What we take from this: AI-CAM is technically feasible. But selling into diverse environments is an intractable go-to-market problem. Control your own stack.

Plethora had the right vision — own machines, automate CAM, serve prototyping demand. But they built facilities before their software could reliably handle complex geometry. Software lagged hardware, creating operational chaos. The timing was wrong.

Formlogic went heavy on custom hardware and large-scale operations before proving a single profitable cell. They bet that scale would fix model efficiency. It didn't — they burned capital trying to make economics work at volume without first proving them at unit level.

What we take from this: Sequence matters. Software capability first, then facilities, then scale.

How Our Approach Differs

The pattern we must avoid

Every failed attempt in this space made the same mistake: scaling operations before proving that software could reliably automate CAM for complex parts. We sequence the other way — prove automation first, then scale.

Our approach synthesizes the lessons from every player in this space:

• Start from AI-CAM as the core, not orchestration or marketplace. The programming bottleneck is where the margin lives. Solve that first.
• Control all variables. One machine (DVF 5000), one tool library (120 HSK-A63 tools), one material (6061-T651 aluminum). This makes the AI problem tractable — unlike CloudNC, we don't need to generalize across environments.
• Prove economics on real customer parts before adding capacity. The Alpha Facility exists to validate unit economics on actual orders, not to impress with scale. See Targets & Scaling.
• Data compounds. Every part we cut generates labeled training data — geometry, toolpaths, cycle times, inspection results. This flywheel doesn't exist for marketplaces or software vendors. It only exists if you own the machines and close the loop.