An Update on the State of AI in Manufacturing

Meta is putting $600 billion dollars into AI infrastructure over the next five years. If you bought $600 billion dollars worth of CNC machines, how many parts would you have to make—and what would you have to charge—to pay them off?

In the latest episode of the Cutting Through the Noise podcast, Paperless Parts Co-Founders Jason Ray and Scott Sawyer sat down to unpack what all the AI hype actually means for manufacturers. Yes, the technology is moving fast. Yes, it is incredibly powerful. And yes, there is a real argument that AI may be as cheap as it is ever going to be right now. But that does not mean manufacturers should become overly dependent on it, start replacing critical people, or ignore the processes and business fundamentals that actually make technology valuable in the first place.

Here were the biggest takeaways from the episode:

1. AI is moving fast, but manufacturing value still depends on context.

A model can sound smart about a part, a print, or a process. That does not mean it understands your shop well enough to make reliable business decisions. In manufacturing, useful output still depends on real context: your equipment, your workflows, your people, your costs, and your customer requirements.

2. The most powerful AI is not always the most practical AI.

Frontier models are impressive because they can handle a wide range of unfamiliar tasks. But in a manufacturing environment, many workflows are repetitive and specific. That makes smaller, domain-specific models a much better fit in many cases because they can be more affordable, more accurate, and easier to deploy in a durable way.

3. This is the cheapest AI will ever be.

The technology is improving quickly, but the economics are not that simple. The best AI models are becoming more capable while also becoming more expensive to train and run. Infrastructure, energy, hardware, and usage costs all matter. That makes it risky to build a strategy on the assumption that advanced AI will always be cheap enough to use everywhere.

4. Using one giant model for every task in the business is not a realistic long-term answer.

There is a temptation to believe one general-purpose model will eventually handle everything: reading drawings, interpreting emails, routing work, writing code, and making decisions across the business. The more realistic path is targeted use. Repeated, business-critical workflows need tools that are trained and structured for those jobs, not just a powerful model pointed at everything all at once.

5. Security, compliance, and trust still matter more than flashy output.

For manufacturers working with technical drawings, customer data, and sensitive information, AI has to be secure, compliant, and reliable. A tool that produces a slick answer but mishandles data or cannot be trusted at scale creates more risk than value.

6. Replacing critical thinking with AI enthusiasm is a mistake.

There is a difference between using AI to strengthen a business and using it as a shortcut around judgment, process, and expertise. Manufacturers still need sound systems, knowledgeable people, and clear operating discipline. Technology can accelerate good fundamentals, but it does not replace them.

7. Durable value > novelty.

The companies that will get the most from AI are the ones applying it where it creates repeatable value, controlling costs, and making sure the technology can hold up in a real business over time. That is a much less flashy approach, but it is the one most likely to matter.