Build vs Buy: Should Your Equipment Company Develop In-House Design Automation?

The Temptation of “We’ll Just Build It Ourselves”

Every equipment OEM with an ambitious CTO has had the same conversation in the past 18 months: “AI is everywhere now — can’t we just build a tool that automates our SolidWorks design process?”

It’s a reasonable question. Large language models are commoditized. Open-source computer vision frameworks are mature. SolidWorks has a published API. In theory, the building blocks are available to anyone.

In practice, the gap between “available building blocks” and “a system that reliably converts a P&ID into a manufacturable SolidWorks assembly” is roughly $1.2 million and 24 months of development — and that’s if everything goes well.

This article breaks down the real costs, timelines, and talent requirements so you can make an informed build-vs-buy decision.

What You’re Actually Building

P&ID-to-3D automation is not a single AI model. It is a pipeline of at least six distinct technical systems that must work together flawlessly:

• P&ID parsing and symbol recognition — extracting valve types, instruments, line sizes, and connectivity from 2D drawings (often scanned PDFs, not clean CAD exports)
• Bill of materials generation — mapping recognized symbols to real component catalogs with correct specifications
• 3D component library management — maintaining parametric SolidWorks models of thousands of fittings, valves, regulators, and manifolds
• Spatial layout optimization — determining where each component physically sits within the enclosure constraints
• Intelligent tube/pipe routing — generating collision-free routing paths that respect bend radii, service access, and manufacturing constraints
• Native SolidWorks assembly output — producing .sldasm and .sldprt files with proper mates, configurations, and drawing-ready metadata

Each of these is a non-trivial engineering challenge. Together, they represent one of the hardest applied AI problems in mechanical engineering.

The True Cost of Building In-House

Talent: The $600K+ Annual Problem

The core challenge is talent. You need engineers who understand all three domains simultaneously: machine learning, mechanical design, and the SolidWorks API. Based on LinkedIn data and industry salary surveys from Robert Half and Hays, here’s what the team costs:

• ML/Computer Vision Engineers (2-3 required): $130K-$180K each in the US, EUR 90K-130K in Europe
• SolidWorks API Developer (1-2 required): $100K-$140K each — a rare specialty, as most SolidWorks users never touch the API
• Mechanical Engineer with AI literacy (1-2 required): $90K-$120K each — someone who can validate that AI-generated assemblies are actually manufacturable
• DevOps/Infrastructure (1 required): $120K-$160K — GPU compute, model training pipelines, version control for 3D models

Minimum viable team: 5-7 people. Minimum annual fully-loaded cost (salary + benefits + compute + tools): $600K-$900K in the US. Even in lower-cost markets, you’re looking at $300K-$500K.

Timeline: 18-30 Months to MVP

Based on published case studies from companies that have attempted similar automation projects (Siemens’ Xcelerator development, Autodesk’s generative design rollout), expect this timeline:

• Months 1-6: P&ID parsing achieves 70-80% symbol recognition accuracy. Unusable for production.
• Months 7-12: Accuracy reaches 90-95%. Tube routing generates valid but non-optimal paths. Component library covers 40% of your catalog.
• Months 13-18: First end-to-end assemblies produced. Engineers spend 4-6 hours fixing each output. Net time savings: minimal.
• Months 19-24: System handles “standard” designs with 1-2 hours of manual cleanup. Edge cases still fail.
• Months 24-30: Production-ready for 60-70% of your design volume. Ongoing maintenance and model retraining required.

Total investment to reach production: $1.2M-$2.0M. And you haven’t generated a single dollar of revenue from this investment — it’s pure internal tooling.

The Hidden Costs Nobody Budgets For

• Training data creation: You need hundreds of validated P&ID-to-assembly pairs. Each pair takes a senior engineer 2-3 days to validate. Budget 1,000+ hours of senior engineer time.
• Component library maintenance: Every time a vendor releases a new valve or fitting, someone must create and validate the parametric SolidWorks model. This is a permanent, ongoing cost.
• SolidWorks API version changes: Dassault Systemes updates the API annually. Your integration will break. Budget 2-4 weeks of developer time per year for compatibility updates.
• Talent retention: ML engineers in 2026 have a median tenure of 2.1 years (LinkedIn Workforce Report). When your lead ML engineer leaves, they take critical institutional knowledge with them.

The Buy Option: What You Actually Get

NeuroBox D exists specifically because MST’s founders understood this problem from the inside — having spent years at the intersection of semiconductor equipment design and AI research.

When you subscribe to NeuroBox D, you get:

• Immediate capability: Upload a P&ID, receive a native SolidWorks assembly. No 24-month development wait.
• Continuously improving models: MST’s ML team trains on thousands of real-world P&ID-to-assembly conversions across multiple equipment categories. Your in-house team would train on only your own limited dataset.
• Maintained component libraries: Thousands of parametric components from major valve, fitting, and instrument manufacturers — kept current by a dedicated team.
• No talent risk: You don’t need to hire, retain, or replace ML engineers. Your mechanical engineers use the tool; MST maintains the AI.

When Building In-House Actually Makes Sense

To be fair, there are scenarios where building internally is defensible:

• You have 5+ ML engineers who already understand CAD. This almost never happens outside of Dassault, Siemens, and PTC.
• Your designs are so proprietary that no external tool can handle them. In practice, P&ID conventions are highly standardized (ISA 5.1), so this is rarer than companies think.
• Design automation is your product, not your tool. If you’re selling automation software to others, building makes sense. If you’re using it to design gas panels faster, it doesn’t.

The Decision Framework

Ask yourself three questions:

• Do we have the ML + mechanical engineering + SolidWorks API talent on staff today? (If no: buy.)
• Can we afford to wait 24 months for production capability while competitors adopt AI now? (If no: buy.)
• Is design automation our core business, or a tool that serves our core business? (If it’s a tool: buy.)

For 95% of equipment OEMs, the answer to all three questions points in the same direction: subscribe to a purpose-built solution, and allocate your engineering talent to what actually differentiates your company — your process expertise, customer relationships, and application knowledge.