The make-anything factory

I once shared the idea that Custom manufacturing should be as scalable as the web. I spent eight years trying to solve that problem. We made some useful contributions, but didn’t get all the way there. Why?

The idea is simple and I still think it’s a good one: what if manufacturing capacity could be accessed as easily as computing capacity?

The internet covers the planet like a nervous system. Can it be connected to our planetary network of factories, and the logistics that move the goods they produce? What if you could assemble and move atoms as easily as bits?

It’s a funny time to reflect on these questions. As I write, some forms of computing capacity are severely bottlenecked. There are not enough physical GPUs in the world to run all the training and inference needed for artificial intelligence. Even something that seems as fungible as computing has a caloric cost that we can’t get away from. A hundred thousand people need to put pants on every morning to help make more GPUs, connect them to the internet, and train the models.

To build programmable planetary-scale manufacturing infrastructure, several assumptions need to be made:

  1. People know what they want to make.
  2. What people want to make can unambiguously be translated into manufacturable specifications.
  3. The supply is fungible, many factories can make the same thing.
  4. The supply is scalable, can easily be expanded or contracted.
  5. People can trust that the output will be fairly priced and accurately made.

None of these assumptions can be relied upon. And we’re not really close.

Both supply and demand want a competitive advantage. Suppliers don’t want to be fungible and easily replaced. Buyers want to have access to exclusive supply that differentiates their product in the market.

When you buy server capacity from cloud providers, the options are fairly simple. Computers can handle many different types of tasks, so all you need to specify is how powerful you want your servers to be. Factories, on the other hand, are highly specialized. They can typically only make one kind of thing. If we had the technology for general-purpose matter synthesis, then factories would be more comparable to servers.

Is it possible to create programmatic manufacturing infrastructure without such as sci-fi idea as matter synthesis? I see two paths:

  1. general-purpose factories
  2. a cooperative model for specialized factories

Make-anything factories focus on using general-purpose robots, intelligent planning systems, and other technologies that are agnostic of the output. They have the potential to be an open, upgradable template for a new kind of general-purpose factory that can compete with specialized factories. Makers of make-anything factories abstract building of general-purpose factories.

In the beginning, make-anything factories will only be competitive with factories that rely on low-skill human labor. Over time it may be possible to modularize equipment.

I have never been a believer in

Whereas software thrives on open-source code and a culture of concise explanations, the supply chain industry thrives on proprietary information that slows overall progress. The supply chain industry’s bias towards specialization is intertwined with its culture of secrecy and opacity. Finding arbitrage is how you remain competitive.

Flipping the switch from opacity to transparency is the most important catalyst I can imagine for the progress of manufacturing infrastructure. But that change will not occur without a shared incentive. Without a reward for cooperation, there is no incentive for supply chains to become transparent.

Both buyers and suppliers would need to value shared, fungible infrastructure. Factories would need to see value in being fluid nodes in the manufacturing network. Both parties would need to have confidence that there is no significant arbitrage to be gained from fluctuations in quality and price across factories.