Last week, I wrote about AI supply chains and risk. I planned for this week to discuss the diversification of AI tools. Then a commenter suggested I consider the differences between key person risk, key supplier risk, and key infrastructure risk. So, that’s just what I’m going to do. We will bump the diversification of AI Tools to next week. So, stay tuned.
In manufacturing, we don’t confuse risk categories. We distinguish between them, but in AI adoption most companies are blending them together or worse, ignoring them entirely. Let’s separate them.
1. Key Person Risk
You’ve seen this before. The engineer who “knows the system.” The salesperson who “owns the account.” The operations manager who “understands the ERP.” If they leave, productivity drops and tribal knowledge walks out the door.
Now translate that to AI.
If one marketing manager built all your prompts…
If one operations leader configured your AI forecasting…
If one tech-savvy salesperson is the only one using the AI copilot effectively…
You don’t have AI integration. You have AI dependency on a person. That’s fragile.
Documented workflows. Shared prompt libraries. Cross-training. Governance. That’s how you reduce variation. AI is no different.
2. Key Supplier Risk
This is where companies are starting to wake up. If your CRM automation, content generation, forecasting, quoting logic, and customer support all sit inside one ecosystem — whether that’s OpenAI, Anthropic, or another platform — you have supplier concentration risk.
What happens if:
• Pricing changes
• Access tiers shift
• Data policies evolve
• Model performance degrades
• Regulatory pressures alter availability
Manufacturers learned after resin shortages and chip constraints that dual sourcing isn’t paranoia. It’s prudence. Diversification doesn’t mean tool sprawl. It means intentional architecture, including core system, secondary capability, data portability, and exit optionality. That’s strategy. Not fear.
3. Key Infrastructure Risk
This is the one almost no one is discussing. AI is powered by:
• Advanced semiconductors
• Hyperscale cloud providers
• Massive energy consumption
• Geographically concentrated data centers
When OpenAI expands data centers or Anthropic deepens cloud partnerships, that’s not just scale. That’s infrastructure dependency. And infrastructure concentration behaves differently than supplier concentration. You can swap vendors. You can’t easily swap global compute ecosystems.
Manufacturers understand this instinctively. When energy grids fail, ports shut down, or geopolitical tension affects raw materials, the impact cascades. AI sits on similar structural layers.
If AI becomes embedded in demand forecasting, production planning, pricing models, lead qualification, or customer communication then infrastructure instability becomes operational instability. That deserves board-level visibility.
So What Does Diversification Actually Look Like?
Not 17 AI subscriptions. Not chaos. Diversification looks like:
• Clear process mapping before automation
• Defined ownership and governance
• Data backed up outside the tool environment
• Multi-model testing (where appropriate)
• Scenario planning for provider disruption
• Cross-functional alignment between sales, marketing, and operations
AI is now part of your value stream. Don’t run and hide from it, or ignore it. In Lean Six Sigma terms, every value stream must be evaluated for variation, bottlenecks, failure points, and cost of poor quality. If AI reduces waste but increases systemic fragility, you’ve shifted risk not eliminated it.
The Strategic Leadership Question
The question isn’t: “Are we using AI?” It’s: “Have we applied supply chain discipline to our AI ecosystem?” Because in manufacturing, we don’t wait for the second disruption to take risk seriously. We redesign the system. If you wouldn’t single-source your most critical raw material, why would you single-source the intelligence layer of your business?