I feel somewhat conflicted about this--on the one hand, as a software nerd, the idea of bringing modularity into physical domains is super neat. On the other, it gives me a hint of 'solution in search of a problem'.
It would be useful to compare and contrast this to the evolution of the 3D printing space: there's a lot of initial promise, a tremendous speed up to product prototyping, some niche applications where it's well suited, but broadly failing to displace traditional high-volume manufacturing. Particularly, the 'high mix' use-case seems rarer in practice. Handy for prototyping, but once you've nailed it, you optimize the line and shave every penny off the high-volume part.
How would you see this approach emerging differently? I'd imagine that, if there is some cross-over point where sheer bulk parallelization overtakes traditional volume manufacturing, there's a large initial gulf. Cheaper than a prototyping firm, but still sufficiently more expensive than what an agile Chinese manufacturing center can turn around. What scales this across the gap?
If this did totally succeed, does that just move the value up the supply chain to the material processing? I don't see shipping-container-sized steel foundries or aluminum smelters being viable. Can this address American reindustrialization without tackling the supply chains? Right now, even if my factory were free, I think if I wanted to make, say, motors, it might actually be cheaper to buy a motor off Alibaba and break it into parts than to try to secure the magnets alone. Especially if I weren't capable of securing a single-part high-volume contract because my software-defined factory doesn't build anything specific in large quantities.
Anyway, I like the ideas, I've thought along similar lines in the past. Show me I'm wrong!
These are all great points that deserve a longer response but for the sake of speed, some thoughts in no particular order:
1. Nothing is a silver bullet -- it doesn't address material costs (but I do think it will make it easier to adapt to new materials so it will be easier to sub things out and create new processes around new materials)
2. There are absolutely many processes that are "physics dominated" like aluminum smelting and many other things that require large volumes of hot stuff. (That being said, I suspect that we may also be able to discover ways around that over time with tools like lasers, continuous processing, etc but that is lower confidence)
3. My bet is that the demand for "high mix" is elastic -- if it was cheaper and faster, people would do more of it
4. It certainly isn't cheaper than a chinese manufacturer right now [in part because it barely exists] -- the crux is going to be how quickly it can get there and for what; even a chinese manufacturer needs to retool a factory and dial in a process
(This all being said these are hypotheses that I feel confident about but could be wrong)
On a tangent (speaking of physics-dominated processes), when I was thinking through the space, one of the more interesting problem domains is allocating rigidity. Many industrial machines begin on a bedrock of bolting some sufficiently large mass of cast iron or epoxy granite to a concrete foundation. Assuming that foundation is deep enough, you have a limitless sink for vibrations. It also runs pretty counter to easy modularization.
If you place a bunch of machines into a shipping container that's intended to be movable, you now have a potentially vibrating box whose dynamics are going to be contingent on exactly which machines are in use and how they're being used (e.g. a part size and the machine's speed is going to determine a resonant frequency). So you have to figure out what your weight budget is, how to allocate it based on projected workloads, and how to plan those workloads given the parts/precision/feed rates needed. Not impossible, but a fun new dimension to work through when it comes to planning.
I think you have the wrong names on the pictures of the solar system. The pictures are of the geocentric and heliocentric models, associated with Ptolemy and Copernicus.
Multi-domain optimization would be an awesome way to bootstrap this. It feels like it could possibly be within reach of current AI, an extension of vibe coding. If a non-expert today could design anything, there are plenty of contract fab services today that can make various pieces, and owning those would be just an upgrade. MDO scales like software, so could pay for itself today, and rigorous backends avoid some of the AI reliability problems.
I feel somewhat conflicted about this--on the one hand, as a software nerd, the idea of bringing modularity into physical domains is super neat. On the other, it gives me a hint of 'solution in search of a problem'.
It would be useful to compare and contrast this to the evolution of the 3D printing space: there's a lot of initial promise, a tremendous speed up to product prototyping, some niche applications where it's well suited, but broadly failing to displace traditional high-volume manufacturing. Particularly, the 'high mix' use-case seems rarer in practice. Handy for prototyping, but once you've nailed it, you optimize the line and shave every penny off the high-volume part.
How would you see this approach emerging differently? I'd imagine that, if there is some cross-over point where sheer bulk parallelization overtakes traditional volume manufacturing, there's a large initial gulf. Cheaper than a prototyping firm, but still sufficiently more expensive than what an agile Chinese manufacturing center can turn around. What scales this across the gap?
If this did totally succeed, does that just move the value up the supply chain to the material processing? I don't see shipping-container-sized steel foundries or aluminum smelters being viable. Can this address American reindustrialization without tackling the supply chains? Right now, even if my factory were free, I think if I wanted to make, say, motors, it might actually be cheaper to buy a motor off Alibaba and break it into parts than to try to secure the magnets alone. Especially if I weren't capable of securing a single-part high-volume contract because my software-defined factory doesn't build anything specific in large quantities.
Anyway, I like the ideas, I've thought along similar lines in the past. Show me I'm wrong!
These are all great points that deserve a longer response but for the sake of speed, some thoughts in no particular order:
1. Nothing is a silver bullet -- it doesn't address material costs (but I do think it will make it easier to adapt to new materials so it will be easier to sub things out and create new processes around new materials)
2. There are absolutely many processes that are "physics dominated" like aluminum smelting and many other things that require large volumes of hot stuff. (That being said, I suspect that we may also be able to discover ways around that over time with tools like lasers, continuous processing, etc but that is lower confidence)
3. My bet is that the demand for "high mix" is elastic -- if it was cheaper and faster, people would do more of it
4. It certainly isn't cheaper than a chinese manufacturer right now [in part because it barely exists] -- the crux is going to be how quickly it can get there and for what; even a chinese manufacturer needs to retool a factory and dial in a process
(This all being said these are hypotheses that I feel confident about but could be wrong)
On a tangent (speaking of physics-dominated processes), when I was thinking through the space, one of the more interesting problem domains is allocating rigidity. Many industrial machines begin on a bedrock of bolting some sufficiently large mass of cast iron or epoxy granite to a concrete foundation. Assuming that foundation is deep enough, you have a limitless sink for vibrations. It also runs pretty counter to easy modularization.
If you place a bunch of machines into a shipping container that's intended to be movable, you now have a potentially vibrating box whose dynamics are going to be contingent on exactly which machines are in use and how they're being used (e.g. a part size and the machine's speed is going to determine a resonant frequency). So you have to figure out what your weight budget is, how to allocate it based on projected workloads, and how to plan those workloads given the parts/precision/feed rates needed. Not impossible, but a fun new dimension to work through when it comes to planning.
I think you have the wrong names on the pictures of the solar system. The pictures are of the geocentric and heliocentric models, associated with Ptolemy and Copernicus.
Worthless drivel written by someone who has never manufacturing anything in their life.
Multi-domain optimization would be an awesome way to bootstrap this. It feels like it could possibly be within reach of current AI, an extension of vibe coding. If a non-expert today could design anything, there are plenty of contract fab services today that can make various pieces, and owning those would be just an upgrade. MDO scales like software, so could pay for itself today, and rigorous backends avoid some of the AI reliability problems.