11 months ago, it was only Steve Yegge. now, thanks to the power of software and factories, it could be just about anybody but if i see them i'll tell them you're interested in seeing a prospectus
This feels more like a surface level googling about who assembles the machines. If your castings, precision ball-screws, ways, and controls are being imported, then your import policy needs to take that into consideration or you won't get far.
Can America build the precision machine parts needed to manufacture useful robots ?
Browsing Michelle Suns other articles gives a great insight into how un-viable it is for American manufacturing to catch up with Chinese manufacturing.
There are pretty massive benefits of scale / density in Shenzen to support every aspect of manufacturing, which are similar to the benefits of VC funding scale in Silicon Valley.
Its hard to reproduce either ecosystem, and impossible to do it quickly.
Even if we use a self-improvement loop that uses AI to train industrial robots to run machine tools, thats going to take some time - maybe more than a decade, even if we hit AGI / ASI in the next 5 years.
and the race to AGI / ASI is already being throttled by compute, which is being limited by the rate at which we can build datacenters and get power sources for them - some GPUs are sitting in warehouses, they might even lose half their value before they get installed.
Perhaps a better outcome for everyone is a detente, a re-invigoration of the free trade cooperation/competition that gave us this incredible growth and prosperity the past 40 years, with its intimately co-mingled supply chains ?
Its probable that engineers in Shenzen can innovate faster in using AI to run robots to run machine tools to manufacture machine tools and parts for robots, than America can. Their AI chips and LLM models are good enough, only getting better.
I guess the counter-argument would be the rapid innovation of SpaceX engines and catching both starship stages and the scaling up of Tesla car and robot production.
Not to mention the very successful Artemis Moon loop mission - clearly America _can_ innovate in manufacturing / physical tech.
China is the counter argument to China. Their rise was improbable and hard won but it proves that you can overcome an unimaginable lead if you really care.
On incredibly short timescales — we're talking less than 25 years, really, for the major part of the ramp. You could argue "2000-2010" with a straight face. It's 'just' will & well-regulated markets.
It was done with a different political system though. The 'move everything to China' approach made huge fortunes for many now wealthy and powerful Americans, at the cost of the wider country. China would not make the same mistake.
exactly re your last point. While there's a lot of opportunities to optimize factories with AI, Shenzhen factories can do the same with agents and cheaper LLMs.
My guess is that the US (and western countries) can likely build a decent supply base for price-insensitive verticals like defense, through a mix of venture capital support and govt budget. And it's still worth doing for AI / physical AI sovereignty.
> Perhaps a better outcome for everyone is a detente
> It's probable that engineers in Shenzen can innovate faster in using AI to run robots to run machine tools to manufacture machine tools and parts for robots, than America can. Their AI chips and LLM models are good enough, only getting better.
These are directly contradictory statements. If China does not need the US, but the US needs China, then detente is not in China's interest, and therefore not a 'better outcome for everyone'.
Im assuming a detente where America leads in innovation in the algorithms of AI, and China leads in the production of robotics : Shenzen and Silicon Valley play to their strengths resulting in mutual co-dependence.
I understand what you are saying, but it isn't possible.
They already have complimentary roles, and this hasn't lead to a deepening co-dependence. The dependence is becoming more one sided. You've already said as much in your own words - China's AI are good enough.
If China needed the US to provide models, then things could work (for some time) as you suggest.
The issue is that, without a Chinese dependence on the US, detente is not the best strategy for China - they can instead leverage their position for better terms. A < 12 month lag in models is only a temporary setback.
China's robots will lead unless you start building robots yourself.
To stimulate that, you need to nurture a long-term stable climate where (for manufacturers & investors) it pays to build robots & max out factories with those.
Over in Europe, that long-term vision is lacking. Or isn't executed upon. Anti-robotization sentiment "think of those poor worker's jobs!" doesn't help either.
I am old so I grew up with the panic that the USA does not make the most steel anymore. At the time I thought: wow: we're probably way behind in brick and concrete too.
We still manufacture the most rockets, the most satellites, the most jets, the most missiles etc. I think it is easier to relearn or reinvent the machines that build the machines. Being cutting edge is hard. Tesla essentially had to go back to assembly lines, a technique everyone said Americans can't do anymore and they invented new ways of assembling autos (I'm think building in pieces, not off the frame) that many companies imitated.
I like this perspective. I think right now there's a lot of exuberance of reindustrializing _everything_ but perhaps the highest ROI is to focus on what America does best, vs. trying to reinvent everything as if we're an island of our own.
It seems like part of the tension is between the question of what has the highest financial ROI vs the idea of manufacturing self-sufficiency being seen as important from a geopolitical standpoint.
Your example kinda belies your thesis. Tesla is producing a substantially worse product than its Chinese competitors and surviving only thanks to protectionism.
A notable milestone was the heavy press program --- but research where Tesla bought its Gigapress and where the now largest/heaviest was recently brought online.
First time hitting the front page. Definitely made my day as a long time lurker!
When I look at the reindustrialization efforts, the bottleneck keeps showing up in the machines themselves (and the machines that make those machines). The question then shifts from “we need more factories” or "we need robots that run the factories", to the capex, precision components, process knowledge, and industrial base that make modern manufacturing possible.
Curious what people see as the highest-leverage places to tackle.
>Curious what people see as the highest-leverage places to tackle.
Economies of scale obviously
Until we figure out how to scale economically China will continuously undercut every disorganized effort we come up with, whether it's robots, AI, energy, EVs, rare earth minerals
It's the same story every time. We make a small twitch, they set up a small government office (it was only 2 people for REMs) that helps organize the entire industry into one unified movement. It's poor governance, simple as that
Interesting you say that. I think it's a bit deeper than just poor governance. I think the system itself doesn't lend well to longer-term planning and execution.
Auckland, New Zealand's mayor recently got into a bit of controversy for saying “Democracy’s had its day. It’s time for something else, really”
I don't know about other countries, but in NZ we have had many projects stop-start-stop etc, being stuck in endless planning cycles etc, because the election cycle is every three years, and people will end up voting a government out partly because they haven't seen any progress from the current government. Enter the next govt, and the things that were being planned will be canned.
Not suggesting the CCP model should be adopted. But I do think there needs to be a rethink on how long-term projects and infra can be supported without being affected by short election cycles.
Germany does great at planning and was a democracy last I looked.
Some authoritarian states might do well at planning but many haven't. And things vary over time. The Chinese that seems to be doing well at planning now is the descendant of the Maoist state that gave us the (utterly disastrous) Great Leap Forward.
The US is bad on educating and hiring machine shop as well as other factory workers. Industrial workers aren't culturally respected, don't make a good living until they're quite experienced and don't have job security until then. Long, split shifts are standard here along with high intensity work. But that makes these jobs less desirable than a similarly paid less intensive job and thus make them just not desirable. All the efforts to change this stuff are piecemeal or involve importing workers.
The US also doesn't educate regular engineers anymore or emphasize hands-on results with the engineers. And, well, look at how Boeing treated it's highly skilled workforce in Washington. If policy makers cared about this sh*t, they'd push back on maneuvers like that.
Extremely illuminating in the same context as your piece.
I am not particularly moved by arguments against globalization but giving away our tool and die + heavy industry base load now seems like a huge loss, and I'm not sure how you rebuild an advanced manufacturing base without first rebuilding that.
Maybe there's a way to skip some generations of development and volume here, and I'd love to read more about it.
I was going to mention Dustin as well as the old Tim Cook interview about how the tool making skills Apple needs for their products primarily exist in China.
The US needs to rebuild both the tools AND the lost knowledge base/know how.
Why think in terms of bottlenecks rather than incentives?
The Americans with the capital to fix these problems, invest that capital in China.
If you run a factory in America, then lots of your profit ends up going to China anyway. You pay high rent for the factory, high interest on your bank loans, high wages to workers who pay high mortgages themselves. This money goes to landlords/bankers who then invest it in Chinese factories instead because they get a better return. This is all standard economic theory. Only if/when investing in America is incentivized will you get large investment in America.
What is the solution? Tariffs perhaps but in the interim the consumer is forced to pay them and manufacturing seems to rather wait it out for them to be revoked under consumer/voter pressure than invest in onshoring. I’m not sure another lever. Outright embargo would also see similar economic pressure to lift the embargo. The economy is so codependent on china now in just about every sector. It is almost like it is too late to do much of anything about it.
Then again at the rate of china’s income growth, maybe latin america would one day actually prove cheaper from a labor and overhead perspective. Far closer too and just as mineral rich.
It is crazy how rather down to earth and low tech some of the machine and machine tool manufacturers operate.
We did a data analytics project for a company that builds strand press tools for manufacturers world wide. It is basically about understanding the impact of material, temperature, etc on the production process and looping back precision metrology.
Interesting thing is that they did not even test the tools but rather improve based on feedback from the their customers thousands of kilometers away.
It a lot about controlling the process and people's experience over years. A lot of Germany's 'hidden champions' are specialized precision machine and machine tool manufacturers. Unfortunately for Germany particularly China is catching up in terms of controling the whole value chain...
Most Haas components are made in China, aren't they? Final assembly may be in the US but I don't think that's where the castings are made.
But at any rate, is there a problem with the US buying machine tools from Japan and Europe? Those are friendly allied countries and they do an excellent job.
And the machine tool industry is famously cyclical. This is one reason why builders don't invest in increasing capacity during boom times, because inevitably they are followed by a bust.
Nice. I didn't realize that the US had lost its position in high-end machine tools, too.
Volume in the machine tool business is dependent on the derivative of product production. If product production is trending downward, machine tool sales drop to near zero. They're long-lasting machines, usually good for many decades. There's a sizable used market. The new replacement market is thus small. Machine tools become obsolete, but often, the big iron gets a new controller and goes on cutting metal.
Conversely, if product production is growing, as in China until recently, there's a huge demand for new machine tools.
All those US medium-sized machine tool makers that made gear hobbing machines and such seem to be gone. Along with most of the people who were really into precision machining.
You can't do this stuff with Amazon warehouse level employees.
Robotic joints are an interesting case. They need a small 3-phase motor, which is a lot like a drone motor. They usually have a harmonic drive. And of course they have sensors and a controller. For a long time, robotic joints for mobile robots were put together mostly by grad students in tiny quantities using off the shelf components intended for other things. But at last, they're a volume business. The practical engineering problems have been solved, and there are at least 18 humanoid robots good enough to have Youtube videos.
There are a lot of tradeoffs in robot joints, and there was a good article about those on HN a few months back. Harmonic drives that can handle overloads and back-driving are tough to make, but that's now solved. Too much reduction, back-driving fails, and gear teeth break. Too little reduction, and the motors have to be too big. Motors often have to be overloaded temporarily, and need thermal monitoring to get away with that. Balancing all those issues requires components built specifically for the task. It's mostly engineering cost. Once the design is right, those things are all variants of mass produced products. Which is why, once the volume gets into reasonable numbers, humanoid robots may not be that all expensive. Probably comparable to a car.
Some of the processes are done by hand on the larger machines, and optical metrology can be used as the reference when you don't have a machine big enough to provide one.
E.g. you scrape in the ways of the grinding machine which makes the beds for the lathes which turns the camshafts or whatever.
Surface tables, lenses, etc can be bootstrapped without need a bigger more precise machine.
>"2. The US builds everyday machines and imports the precise ones.
About 55% of import value is the high-end versions of machines US plants already make, such as 5-axis machining centers. Precision machines, small-module gear grinders, precision grinders and the heaviest forging presses have no confirmed US volume builder. Imports rose 29.5% for grinders and 76.7% for gear machines in Jan to Sep 2026. These machines make the ball screws, bearings, gears and forgings that aircraft, turbines and robots depend on."
Observation: There's business value to be created in producing products in common markets that appear to be saturated and completely closed to competition -- by simply providing similar products that those markets already sell, but at higher degrees of precision...
Granpa worked the Alegaeny National Forge, plant is 1/2 mile wide and 3 miles long, the little crane is 100tons, likely dust now along with the rest, Granpa would tick off the companys sold out and moved to country with cheaper labour, which is 'optimisation' and therfor never to be questioned.
All industry requires large precision equipment. All of that equipment needs a base and body which must be very ridged and strong, nothing does this so well as a high nickle cast iron.And all of the moving parts must also be ridged and take a lot of torque, and nothing does that so well as alloy steel forgings that are the machined.
The machines make the machines, but the fine precision, design, efficiency, and longevity are done by hand useing experience and knowledge, and the finer it gets the fussyer and more arcane it is.
Can the engineering, manufacturing, set up, tuning and commishioning and then routine maintenance and repair be automated?, no.
As a metal worker who regularly makes tools, jigs, and simple equipment, I can state just how productive such things can be, the original 10 or 100 x factors of industry in action.
Current side project involves creating casting molds that start with sketches or verbal descriptions, move to 3d cad, and then into cnc wooden or possibly 3d printed polymer patterns to be used by foundry workers. The fisrts step is to do this in 2d
to establish work flows, and then move to small 3d, and then scale is only limited by
actual machine/printing and foundry equipment.
marcodiego | a day ago
stuxnet79 | 22 hours ago
0gs | a day ago
showerst | 23 hours ago
This feels more like a surface level googling about who assembles the machines. If your castings, precision ball-screws, ways, and controls are being imported, then your import policy needs to take that into consideration or you won't get far.
jgord | 22 hours ago
It might be hard to model this domain well, but she seems to be discussing it rationally, diving into the details.
michellelsun | 21 hours ago
@showerst the research is ai-assisted, the writing is not perfect but hopefully not slop :')
my hope is to invite discourse from more parties to think through this important topic for US industrial future
jgord | 22 hours ago
Browsing Michelle Suns other articles gives a great insight into how un-viable it is for American manufacturing to catch up with Chinese manufacturing.
There are pretty massive benefits of scale / density in Shenzen to support every aspect of manufacturing, which are similar to the benefits of VC funding scale in Silicon Valley.
Its hard to reproduce either ecosystem, and impossible to do it quickly.
Even if we use a self-improvement loop that uses AI to train industrial robots to run machine tools, thats going to take some time - maybe more than a decade, even if we hit AGI / ASI in the next 5 years.
and the race to AGI / ASI is already being throttled by compute, which is being limited by the rate at which we can build datacenters and get power sources for them - some GPUs are sitting in warehouses, they might even lose half their value before they get installed.
Perhaps a better outcome for everyone is a detente, a re-invigoration of the free trade cooperation/competition that gave us this incredible growth and prosperity the past 40 years, with its intimately co-mingled supply chains ?
Its probable that engineers in Shenzen can innovate faster in using AI to run robots to run machine tools to manufacture machine tools and parts for robots, than America can. Their AI chips and LLM models are good enough, only getting better.
jgord | 22 hours ago
Not to mention the very successful Artemis Moon loop mission - clearly America _can_ innovate in manufacturing / physical tech.
Groxx | 22 hours ago
orange_joe | 20 hours ago
thechao | 19 hours ago
skew-aberration | 15 hours ago
michellelsun | 21 hours ago
My guess is that the US (and western countries) can likely build a decent supply base for price-insensitive verticals like defense, through a mix of venture capital support and govt budget. And it's still worth doing for AI / physical AI sovereignty.
skew-aberration | 15 hours ago
> It's probable that engineers in Shenzen can innovate faster in using AI to run robots to run machine tools to manufacture machine tools and parts for robots, than America can. Their AI chips and LLM models are good enough, only getting better.
These are directly contradictory statements. If China does not need the US, but the US needs China, then detente is not in China's interest, and therefore not a 'better outcome for everyone'.
jgord | 13 hours ago
skew-aberration | 12 hours ago
They already have complimentary roles, and this hasn't lead to a deepening co-dependence. The dependence is becoming more one sided. You've already said as much in your own words - China's AI are good enough.
If China needed the US to provide models, then things could work (for some time) as you suggest.
The issue is that, without a Chinese dependence on the US, detente is not the best strategy for China - they can instead leverage their position for better terms. A < 12 month lag in models is only a temporary setback.
RetroTechie | 4 hours ago
To stimulate that, you need to nurture a long-term stable climate where (for manufacturers & investors) it pays to build robots & max out factories with those.
Over in Europe, that long-term vision is lacking. Or isn't executed upon. Anti-robotization sentiment "think of those poor worker's jobs!" doesn't help either.
Btw: excellent article!
m463 | 13 hours ago
I'm reminded of california housing prices. The best time to buy a house in california is either 30 years ago, or today.
georgeecollins | 22 hours ago
We still manufacture the most rockets, the most satellites, the most jets, the most missiles etc. I think it is easier to relearn or reinvent the machines that build the machines. Being cutting edge is hard. Tesla essentially had to go back to assembly lines, a technique everyone said Americans can't do anymore and they invented new ways of assembling autos (I'm think building in pieces, not off the frame) that many companies imitated.
michellelsun | 22 hours ago
apothegm | 20 hours ago
serioussecurity | 21 hours ago
WillAdams | 21 hours ago
michellelsun | 22 hours ago
When I look at the reindustrialization efforts, the bottleneck keeps showing up in the machines themselves (and the machines that make those machines). The question then shifts from “we need more factories” or "we need robots that run the factories", to the capex, precision components, process knowledge, and industrial base that make modern manufacturing possible.
Curious what people see as the highest-leverage places to tackle.
redanddead | 20 hours ago
Economies of scale obviously
Until we figure out how to scale economically China will continuously undercut every disorganized effort we come up with, whether it's robots, AI, energy, EVs, rare earth minerals
It's the same story every time. We make a small twitch, they set up a small government office (it was only 2 people for REMs) that helps organize the entire industry into one unified movement. It's poor governance, simple as that
ebbi | 20 hours ago
Interesting you say that. I think it's a bit deeper than just poor governance. I think the system itself doesn't lend well to longer-term planning and execution.
Auckland, New Zealand's mayor recently got into a bit of controversy for saying “Democracy’s had its day. It’s time for something else, really”
I don't know about other countries, but in NZ we have had many projects stop-start-stop etc, being stuck in endless planning cycles etc, because the election cycle is every three years, and people will end up voting a government out partly because they haven't seen any progress from the current government. Enter the next govt, and the things that were being planned will be canned.
Not suggesting the CCP model should be adopted. But I do think there needs to be a rethink on how long-term projects and infra can be supported without being affected by short election cycles.
joe_the_user | 19 hours ago
Some authoritarian states might do well at planning but many haven't. And things vary over time. The Chinese that seems to be doing well at planning now is the descendant of the Maoist state that gave us the (utterly disastrous) Great Leap Forward.
motionlessveloc | 18 hours ago
joe_the_user | 20 hours ago
The US is bad on educating and hiring machine shop as well as other factory workers. Industrial workers aren't culturally respected, don't make a good living until they're quite experienced and don't have job security until then. Long, split shifts are standard here along with high intensity work. But that makes these jobs less desirable than a similarly paid less intensive job and thus make them just not desirable. All the efforts to change this stuff are piecemeal or involve importing workers.
The US also doesn't educate regular engineers anymore or emphasize hands-on results with the engineers. And, well, look at how Boeing treated it's highly skilled workforce in Washington. If policy makers cared about this sh*t, they'd push back on maneuvers like that.
pj_mukh | 19 hours ago
I especially loved Destin Sandlin's attempt at making a simple grill scrubber in America: https://www.youtube.com/watch?v=3ZTGwcHQfLY
Extremely illuminating in the same context as your piece.
I am not particularly moved by arguments against globalization but giving away our tool and die + heavy industry base load now seems like a huge loss, and I'm not sure how you rebuild an advanced manufacturing base without first rebuilding that.
Maybe there's a way to skip some generations of development and volume here, and I'd love to read more about it.
lp92 | 4 hours ago
The US needs to rebuild both the tools AND the lost knowledge base/know how.
Edit: found the interview with Tim Cook: https://youtu.be/2wacXUrONUY?is=-frqJZMEq4wuCZBs
skew-aberration | 15 hours ago
The Americans with the capital to fix these problems, invest that capital in China.
If you run a factory in America, then lots of your profit ends up going to China anyway. You pay high rent for the factory, high interest on your bank loans, high wages to workers who pay high mortgages themselves. This money goes to landlords/bankers who then invest it in Chinese factories instead because they get a better return. This is all standard economic theory. Only if/when investing in America is incentivized will you get large investment in America.
asdff | 59 minutes ago
Then again at the rate of china’s income growth, maybe latin america would one day actually prove cheaper from a labor and overhead perspective. Far closer too and just as mineral rich.
riedel | 20 hours ago
We did a data analytics project for a company that builds strand press tools for manufacturers world wide. It is basically about understanding the impact of material, temperature, etc on the production process and looping back precision metrology.
Interesting thing is that they did not even test the tools but rather improve based on feedback from the their customers thousands of kilometers away.
It a lot about controlling the process and people's experience over years. A lot of Germany's 'hidden champions' are specialized precision machine and machine tool manufacturers. Unfortunately for Germany particularly China is catching up in terms of controling the whole value chain...
mitthrowaway2 | 13 hours ago
But at any rate, is there a problem with the US buying machine tools from Japan and Europe? Those are friendly allied countries and they do an excellent job.
And the machine tool industry is famously cyclical. This is one reason why builders don't invest in increasing capacity during boom times, because inevitably they are followed by a bust.
senderista | 20 hours ago
motionlessveloc | 18 hours ago
Animats | 17 hours ago
Volume in the machine tool business is dependent on the derivative of product production. If product production is trending downward, machine tool sales drop to near zero. They're long-lasting machines, usually good for many decades. There's a sizable used market. The new replacement market is thus small. Machine tools become obsolete, but often, the big iron gets a new controller and goes on cutting metal.
Conversely, if product production is growing, as in China until recently, there's a huge demand for new machine tools.
All those US medium-sized machine tool makers that made gear hobbing machines and such seem to be gone. Along with most of the people who were really into precision machining. You can't do this stuff with Amazon warehouse level employees.
Robotic joints are an interesting case. They need a small 3-phase motor, which is a lot like a drone motor. They usually have a harmonic drive. And of course they have sensors and a controller. For a long time, robotic joints for mobile robots were put together mostly by grad students in tiny quantities using off the shelf components intended for other things. But at last, they're a volume business. The practical engineering problems have been solved, and there are at least 18 humanoid robots good enough to have Youtube videos.
There are a lot of tradeoffs in robot joints, and there was a good article about those on HN a few months back. Harmonic drives that can handle overloads and back-driving are tough to make, but that's now solved. Too much reduction, back-driving fails, and gear teeth break. Too little reduction, and the motors have to be too big. Motors often have to be overloaded temporarily, and need thermal monitoring to get away with that. Balancing all those issues requires components built specifically for the task. It's mostly engineering cost. Once the design is right, those things are all variants of mass produced products. Which is why, once the volume gets into reasonable numbers, humanoid robots may not be that all expensive. Probably comparable to a car.
mrheosuper | 16 hours ago
TuringNYC | 16 hours ago
But then I guess who builds those?
skew-aberration | 15 hours ago
E.g. you scrape in the ways of the grinding machine which makes the beds for the lathes which turns the camshafts or whatever.
Surface tables, lenses, etc can be bootstrapped without need a bigger more precise machine.
peter_d_sherman | 14 hours ago
About 55% of import value is the high-end versions of machines US plants already make, such as 5-axis machining centers. Precision machines, small-module gear grinders, precision grinders and the heaviest forging presses have no confirmed US volume builder. Imports rose 29.5% for grinders and 76.7% for gear machines in Jan to Sep 2026. These machines make the ball screws, bearings, gears and forgings that aircraft, turbines and robots depend on."
Observation: There's business value to be created in producing products in common markets that appear to be saturated and completely closed to competition -- by simply providing similar products that those markets already sell, but at higher degrees of precision...
Related:
The 1751 Machine that Made Everything
https://www.youtube.com/watch?v=djB9oK6pkbA
metalman | 8 hours ago