JANNIK SCHILLING

At the close of the 20th century, the prevailing belief was that the era of manufacturing in Western countries had drawn to a close. Yet, here we stand, at the dawn of a renaissance in American manufacturing: reshoring to the US has surged in recent years. Suddenly, restoring American manufacturing and innovation in manufacturing are at the forefront for startups – even VCs such as Y-Combinator and A16Z are supporting it.

In this piece, I will argue:

1. Tacit knowledge is critical for traditional manufacturing success,

2. Outsourcing delays the destruction of manufacturing,

3. Traditional interventions to preserve manufacturing are often counterproductive, and

4. Automation is the only way to preserve tacit knowledge.

Tacit knowledge is critical for traditional manufacturing success

It is a well-established empirical observation that manufacturing prices drop with increasing manufacturing experience. In 1936, T. P. Wright introduced the concept of “experience curve effects” to describe human capital effects in the manufacturing process. Simply speaking, he assumed that the cost of production decreases by a constant percentage point every time the output produced doubles – this is mainly because learning effects from the repetition of a task. The so-called learning percent is calculated by 1-improvement per doubling and the higher the learning percent, the lower the gains from additional experience.

High-precision manufacturing in industries such as airplane manufacturing or defense technology has higher experience payoffs. For instance, airplane manufacturing costs roughly drop by 15% for every doubling in output, while the manufacturing of raw materials has only a drop in costs of 3-7%. For illustration, TSMC relies heavily on the experience of Taiwanese workers in building a new factory in Arizona to ensure a “fast ramp up”.

In traditional cost estimates, higher shares of machining assembly (lower shares of hand assembly) lead to weaker experience effects. However, modern machinery has rendered most “hand assembly” unnecessary, but large and expensive machines now require more skilled workers to run the machines and set up the production process.

In industries such as semiconductor manufacturing, there are only a few machines of one kind (think ASML EUV lithography systems) with incredibly high upfront capital needs, which implies that (1) there are only few highly-skilled workers that can operate these machines and (2) these machines only break even with low to no downtime. These factors are amplified by monopolies – there are certain machines only operated by ASML.

There are headwinds that American manufacturing has to fight against: outsourcing pressure because of cheaper labor in Asia and retiring machinists. Outsourcing might lower costs in the short term but at the cost of the West losing tacit knowledge. 

Moreover, this loss of tacit knowledge is permanent. The permanence is due to two reasons: First, as one outsources manufacturing, there is little demand for more manufacturing workers, which leads to few young people entering the field. Second, the workers that used to have manufacturing knowledge are retiring now and mostly leave without passing on their tacit knowledge to apprentices because there are none.

Contrast manufacturing knowledge to formal software or academic knowledge. Academia is well-respected and legible which ensures a growing inflow of young, talented students. Moreover, the knowledge itself is captured in universities and code bases (e.g. Github or internal documentation) and there is a robust process of senior faculty members passing on tacit knowledge to young scholars.

Manufacturing knowledge is arguably more important, but predominantly captured in the minds of retiring machinists. The knowledge required to operate these modern machines is tacit because many modern machines have so little quantities that the costs of formalizing knowledge would be high and we did not respect machinists enough to consider their knowledge as important enough to justify the cost. Moreover, there are supporting developments such as: Modern manufacturing systems are much more sensitive to minor mistakes, factories are often built with older and newer machines that are incompatible with each other, high fragmentation of the manufacturing markets require good working relationships with other factories, and better maintenance of machines drive profits because of longer depreciation periods.

Finally, Taylor’s scientific management arguably failed for high-precision manufacturing – the knowledge required to produce these components is not procedural but in the machinists’ heads. My conjecture for why it failed is that labor got so cheap that we did not invest enough in procedural manufacturing advances.

A future research question would explore whether reshoring advanced manufacturing with lower experience curve effects has lower payoffs given that the loss in tacit knowledge is less severe. However, my intuition is that reshoring the production of certain raw materials and rare earth minerals does have strategic payoffs.

Outsourcing delays the destruction of manufacturing

In a world where US companies would not have outsourced manufacturing in tech-heavy industries to Asia, I suspect we would have seen more technological innovation for manufacturing. Instead of reducing prices by hiring cheap labor in Asia, American manufacturing companies would have been forced to develop best practices, find ways to create more automation processes than previously possible, and engage in freer and more transfer of knowledge. If American labor costs are much higher, they can only win on prices if the labor productivity is higher.

To illustrate the argument, we can look to Japan or Germany – two major developed economies that found ways to build an economy based on manufacturing excellence. While both countries are protectionist, they made manufacturing in Western countries feasible by using social and technological inventions such as Germany’s apprenticeship program or Japan’s Lean Manufacturing System.

If manufacturing would have been more attended to by software engineers in Silicon Valley, there would have been more and earlier software automation. However, Silicon Valley is growing up and many software engineers are realizing that building the next SaaS company is unlikely to be important or profitable, but bringing back manufacturing and building hardtech is. In a better world, manufacturing startups like Hadrian could have existed decades ago.

Instead, Western companies focused on cutting costs as much as possible in the short-term without investing in technological innovation. What we got is critical manufacturing outside of the Western sphere of influence where China can lock up IP easily. A telling example is the Arm China IP Theft.

If companies agree to cut costs by outsourcing manufacturing from America to Asia once, they will do it again – once the labor costs in the original country are too high. I think this is one cause for a phenomenon which Dani Rodrik called “premature deindustrialisation,” where deindustrialisation in developing economies happens earlier than economically optimal for the country.

An example for premature deindustrialisation would be redirecting outsourcing from China to Vietnam or Bangladesh, as Chinese labor became more than 8x more expensive than the 2000 price level.

Traditional Western interventions are counterproductive

Historically, keeping manufacturing in one location led to the creation of interest groups that tend to focus on achieving a local maxima of success. Labor unions exercised power through social instruments such as codetermination, strikes, or protectionism. For instance, labor unions often limit automation potential or try to maximize the number of jobs filled by local workers, over foreign workers with more specialized knowledge that could help with knowledge transfer.

In 2022, the CHIPS Act was launched to revive semiconductor manufacturing in the US. Under the CHIPS Act, TSMC decided to construct a semiconductor factory in Phoenix, AZ. The company planned to bring in Taiwanese workers to assist in setting up the complex and costly machinery, but labor unions attempted to block the move, citing concerns over foreign workers undercutting domestic wages. They contended that Intel workers in the US possessed sufficient tacit knowledge. TSMC and labor unions reached a high-level agreement eventually, but TSMC had to postpone the launch due to a shortage of skilled workers.

A social motivation for labor unions exercising their power is ideology. Preserving the local maxima of labor-intensive manufacturing in a country often defaults to anti-tech sentiments. This creates cultural forces and memes that are harmful for keeping price-competitive manufacturing in the country in the long-term.

Germany is emblematic of these developments. Consider the EU’s degrowth ideology and a tendency towards natural fallacies on “organically” produced products and anti-nuclear sentiment.

As soon as a manufacturing-heavy economy stops investing in the automation of manufacturing, incumbents and conglomerates tend to build strong competitive advantages over more dynamic, but smaller firms.

Take Airbus versus Boeing as an example. Boeing tried to position their core competency as “an assembler” of aircraft components, rather than a producer of critical core components. However, Boeing’s engineers stated that engineering knowledge on the components is required to successfully assemble planes. An early example of outsourcing destroying tacit manufacturing knowledge in the Western world.

Automation is the only way to preserves tacit knowledge

If we accept the premise that manufacturing knowledge is crucial for Western Dynamism, we need to find ways to manufacture price-competitive in the US without losing tacit knowledge. If we also accept the premise that traditional manufacturing interventions are counterproductive, we face a need for new innovation to solve the critical issue of reviving Western Manufacturing.

The problem that is being ignored is the one all companies like Hadrian face: they are working against the ticking time bomb of mass retirement of a generation of machinists whose latent knowledge and processes will retire with them. Governments can capture the lost jobs with subsidies and protectionism but startups must capture the lost knowledge with technology.

Hadrian uses technology as an opportunity to lower the burden of the talent shortage by automating as many tasks as possible and making the skill acquisition for the remaining tasks as simple as possible.

In order to achieve the necessary levels of efficiency, Hadrian pairs a software engineer from an inhouse team with a machinist to develop software to automate the high-precision manufacturing process as much as possible. Chris Power, Founder of Hadrian, estimates that Hadrian’s internal tooling can automate around 60 to 80% of tasks. For the remaining tasks, Hadrian builds process-driven software to enable rapid training within 30 to 90 days for every hire – even ones with no prior manufacturing experience.

Manufacturing is pretty fast in adopting innovation

Technology innovation in manufacturing needs to be accompanied by adequate social innovation. For instance, automation fears among blue-collar labor can be addressed by changing their incentive structure.

For instance, Hadrian issues equity to all manufacturing workers to ensure that machinists benefit from sharing their knowledge. Moreover, some machinists report that their jobs become more interesting after the most repetitive components are automated away.

The rapid iteration in the manufacturing industry to enhance productivity is not new. Manufacturing productivity increases historically happened faster than innovation in a broader economic context.

For illustration, consider the invention of the steam engine or “scientific management” in manufacturing. Both innovations replaced old processes and had an immediate impact on manufacturing productivity, lowering costs. Furthermore, manufacturing has historically faced low burdens of adoption, such as regulation, leading to more rapid adoption as a result.

Moreover, manufacturing advances can happen without major technological breakthroughs, while “Deep Tech” needs actual technological progress. In a world where technological progress is scarce, innovation happens through process- and business model-innovation. Lean Manufacturing was developed from the 1940s onwards and got its name in the 1980s.

The fact that startups can feasibly change incentive structures, combined with the industry’s historical quick adaptation to technological changes, makes me optimistic about the feasibility of reviving American manufacturing.

Conclusions

First, the outsourcing waves of the 20th century were disastrous for engineering companies such as Boeing and national security interests – F35 jets were only mission-capable 55% of the time because of unreliable replacement component supply.

Second, technology now has the potential of saving American manufacturing and in turn providing stable prosperity to machinists that SWE jobs once did.

Third, I think American manufacturing matters for “Deep Tech” startups because startups like Hadrian will enable faster feedback loops and increase reliability for startups producing hardware. We can’t move fast and break things if there’s not enough to move or break.

Thanks to Zi C. (Sam) Huang for extensive collaboration and editing.