The marriage of technology and manufacturing could be the salvation for which our nation is searching. The prospect of 4%+ GDP growth does not come from SaaS, banking, real estate, investing, or AI. The opportunity to achieve sustainable GDP expansion comes by way of improved production. Agnostic of debt cycles, we are an industrial society; the idea of a post-industrial economy is a dream cooked up by nihilists. In the pursuit of profits, we have disincentivized trade skills, hollowed out industry, become addicted to services, and enriched our rivals to magnitudes that now threaten our way of life.

The most valuable thing to pursue is also the hardest; rebuilding a modernized US industrial base will take trillions of dollars, but will also extend our prosperity and abundance into the next millennium.

I’ve spent most of my career in technology. I worked at a FAANG company, where I touched IT/ sysops and mechanical engineering (energy and self-driving cars), a smaller tech company doing IT/ sysops, and co-founded (CTO) an AI SaaS startup in market research with two friends. Today, I’m leading Atomic Industries, a company I founded to exascale the American manufacturing sector. With the bottom falling out of tech recently, I feel at ease working on real world challenges, because there’s tremendous value to unlock. The next wave of trillion dollar companies in the coming decades will be in hard tech and/or manufacturing adjacent.

Just like Elon and other founders first built in SaaS, then turned their attention to the world of atoms, techno-industrialists are finding their moment. The last 40 years in tech have created massive last-mover advantage on top of which to build. This technical margin has paved the way for us to transform manufacturing and heavy industry forever.

How We Got Here

I first want to explore an idea that's obvious at face value, but has been so overlooked and ignored that the nation is now facing an existential crisis: knowing how to make things is the ultimate advantage (see: Monozukuri).

Without question, WW2 was the pinnacle of the Industrial Revolution. The United States’ tour de force in spirit and steel overwhelmed our enemies and won us the war.

In just under four years, from 1940 to 1944, the United States transmuted nearly 100 million tons of steel and other metals into a vast array of war materiel. Under the direction of the War Production Board, the U.S. produced over 300,000 aircraft, 86,000 tanks, 15 million guns, 6,500 naval vessels, including aircraft carriers, battleships, and numerous other classes, and over 270,000 artillery pieces. Programs like Training Within Industry were pivotal in training millions of workers to facilitate this production boom. We developed early forms of Just-In-Time inventory management techniques, enhancing efficiency. The financial investment was immense, with the government spending today’s equivalent of $3.22 trillion.

The ramp up and down of pure industrial power, from raw materials and factory mobilization, to machine tools, to the presses stamping out airplanes, tanks, and ships 1,000s of times daily, all across the nation, was truly a miracle. The amount of trade skill Americans developed during this time has never been matched in terms of speed and depth.

Outproducing your rivals is the key to victory, both on and off the battlefield. But after the war ended, we shuttered our factories and pushed globalization and free trade to the point of failure. The pandemic elegantly revealed a litany of sensitivities in the global supply chain, and today we are scrambling to define and close gaps in our industrial base as we move into a more decentralized global economy.

As is now painfully obvious, shedding our industrial base for a post-industrial, higher profit and margin service-based economy was a terrible oversight. Doing so enriched our ideological opponents in both skill and GDP, giving them real and perceived advantages that allow them to comfortably sit atop the throne of industrial power.

Of course, our shift away from manufacturing was not without its gains. The information technology sector exploded over the last 50 years, providing us countless innovations across computing, communications, software, and robotics. But we can't escape our need to build in the physical world. We're just starting to pay for how much we've neglected the world of atoms, it will only get worse.

There is a way to right the ship, to overcome the seemingly inexorable challenge of restoring our hollowed out industrial base. We can and will build our way back. We will do this with the successful marriage of technology and manufacturing; we will accelerate out of industrial stagnation. The post-industrial society is a lie, and we need to forget it right now.

The Industrial Base: A Model

In my conception, there are three primary components to the US industrial base (Figure 1):

(1) Capacity refers to a fundamental layer of processing power in terms of efficiency, throughput, and raw materials. E.g., how many factories and machines do you have, what are their capabilities, and what do they process or produce?

(2) Execution refers to the layer above capacity required to operate and produce goods, AKA tacit/trade knowledge. E.g., high-skill trade knowledge in manufacturing, operations, engineering, or materials.

(3) Application refers to the most abstracted layer — what goods are actually produced. E.g. semiconductors, submarines, cars, apparel, food, drugs.

We are ignoring policy and energy — coupled but separate issues.

---

The Application Layer

Generally speaking, the application layer is the most visible part of manufacturing. You hear more stories about artillery, submarines, toilet paper, drugs, microchips, and drones, than you do about why exactly it is we can’t seem to make them anymore.

Everyone wants X, and they want it now. Anything in the application layer is abstracted away from the execution and capacity layers, and generally, more financially attractive when decoupled from those layers.

“Designed in California, Made in China.”

Outsourcing your production is a blessing and a curse. While you don’t have the added overhead of manufacturing your product, you are not connected to the skill required to actually manufacture it. More, application layer-level product designers are, categorically, not manufacturers, so classically, their designs require copious amounts of feedback from manufacturers to reach a state of manufacturability that is compatible with the factory making the product. This feedback loop is a nightmarish hellscape for everyone who makes physical products.

Generally I like to classify companies by what they bill themselves as when raising money. A drone, biopharma, or satellite company does not say “we are a manufacturing company,” they say “we specialize in drones, biopharma, or satellites.” This is because using “manufacturing” in your billing is negative optics and makes investors think heavy capex. Avoiding manufacturing also signals they are more design oriented.

The amount of time to design any physical product (especially new ones) compared to the time to manufacture it rounds down to zero. In this equation, design is overrated, and manufacturing is underrated.

Sidebar: SaaS Hate

Everyone loves SaaS, but the most valuable companies are all reliant on enormous amounts of capex, either directly or indirectly. Just take a look at Apple, Google, Tesla, Ford, GM, P&G, etc. SaaS is trendy, but companies with actual capex dominate.

You’ve probably never read the 10-K for Netflix, but if you did, you would discover how utterly dependent they are on AWS to survive (current spend estimate is ~$1.5b/year) (Figure 2):

In 2023 alone, Amazon itself spent roughly $20b on infra capex — money spent on acquiring or upgrading physical infrastructure assets such as data centers, servers, network equipment, etc. — to improve AWS.

Not surprisingly, markets and investors are addicted to capital-light, cash-flow reliant, high margin, quick-turn businesses so they can keep building and accelerating returns. But to what end? We don’t get to Mars and become a multi-planetary species by building SaaS.

SaaS has robbed us of so much gifted talent, only to use it to optimize ad clicks, make discovery algorithms more addictive, and create advanced ML models that apply filters to your face. It’s a major misallocation of capital. Too much talent chases software. Having 10 founders all in the same SaaS niche, when they could all be more successful in heavy industry, is a waste.

Further, it's a misconception that SaaS margins are phenomenal — they're actually bad. On average, the EBITDA margins of public SaaS companies are negative. High fixed costs associated with engineering, sales, and support scale significantly after startup phase. Mature SaaS companies just resemble cash flow businesses rather than super high-margin/profitable enterprises.

This is all to say that the market's addiction to SaaS and other capital-light business models is ultimately a malign influence on long-term American dominance. When we are unwilling to think about the capacity and execution layers of the industrial base until it’s too late, we are then forced to pour obscene amounts of blunt-force capital into them when our hand is forced, as happened with the CHIPS Act or the yet-to-be-passed IFCUS Act. Our short term thinking is tragic and costly.

This brings us to the execution layer.

The Execution Layer

The execution layer is the fundamental connective tissue between the capacity and application layers. The true value of this layer lies in tacit/trade knowledge, the stuff on which nations and empires are built. Why are the Germans so good at precision machining and mechanical engineering? How did the Japanese learn to make VCRs, electronics, and cars so well? Why did Taiwan end up as universal epicenter of semiconductor manufacturing?

Each of these questions has a simple answer: they taught, honed, and incentivized these skills more than everyone else. The skills in question are not easily replicated or repeatable without similar tribal understanding and application. Companies like LEGO, TSMC, ASML, and Toyota have ensured their dominance with applied trade knowledge at scale.

Without proper incentives, the labor pool of any given trade will diminish. This is critical because without skilled workers, a nation (for instance) loses the ability to produce critical goods and must outsource or purchase them instead. High-skill trades can take decades to truly master (and even in average trades such as with electricians, certification generally requires a five-year apprenticeship), creating a severe shock to any supply chain that's lost them.

We can see this happening today with semi-conductors. The CHIPS Act supplies capital to build fabs in the US, but we don’t have enough skilled workers to staff or build them, and we banned the Taiwanese — the only people who do know what they’re doing — from helping us build these new fabs (though rumor has it that this is being rectified in Washington right now).

Additionally, operating factories and all their equipment is non-trivial. In terms of full, true automation, we have yet to create an operations abstraction that is both effective and general enough in anything more than the simplest “lights out” facilities. And even those still require a significant amount of high touch labor at some point in the process.

Execution is the most important layer. Without the knowledge of how to make things, it doesn't matter how many applications you invent or how many empty factories you have. Remember: Monozukuri.

The Capacity Layer

The capacity layer represents the raw production power of factories, capital equipment, and raw materials. Manufacturing processes are generally inflexible and are dialed in to limited types of production. Processing, handling, and inspection equipment are configured into manufacturing lines to perform careful choreographies. If the factory is even remotely close to modern, there is typically a data layer as well, but this data is mostly used for operational efficiency gains, such as minimizing downtime, optimizing resource allocation, and streamlining workflows. Rarely is the data fed back into improving the manufacturing process or equipment themselves.

There aren’t many true advancements in production capability, similarly to how most power plants still just spin turbines in one way or another. You can agonize a great deal over how a factory is run, and we've seen this can yield incredible efficiency gains. In general, though, we rarely create more efficient methods to manipulate matter. Take Computer Numerical Control machines (CNCs). We still use metal cutters for CNCs to cut metal, while laser cutting and electron beam machining still aren’t even close to competing.

In terms of how accurately and quickly we can manipulate matter, we have machines like lathes that turn metal or wood, gun drills that bore super accurate holes deep into metal blocks, and Extreme Ultraviolet Lithography machines for etching silicon wafers. It’s been 75 years since we invented the CNC, 35 since the invention of metal 3D printers, and around 50 years since the development of the earliest Computer-Aided Manufacturing systems. We’ve been making small step improvements and efficiency gains over the last 100 years.

The faster we can manipulate or arrange matter, the faster all of manufacturing becomes. We need ways to multiply manufacturing speeds, not just make small step improvements — whether it's at a fundamental level, or process orchestration level.

---

Exascaling the American Industrial Base

The term“‘exascaling” comes from computing. It refers to computing systems capable of performing at least one exaFLOP, or a billion billion (10^¹⁸) calculations per second. I’m reappropriating it for the industrial base, which we need to exascale to the tune of 10^³⁰ atoms per second. The first goal should be surpassing the gap in our manufacturing’s share of GDP (Figure 3) we have with China. This means building technology to augment every industrial layer, but mainly the execution and capacity layers. This is not “industry 4.0” reloaded. Exascaling has a specific emphasis on scalability and technological augmentation at an unprecedented scale.

In our global economy, skills that take decades to learn, require extreme precision in practice, and interface with the physical world drive the most value. Yes, AI is encroaching on knowledge workers (ones who don’t adopt AI), but the most defensible and valuable areas to build will exist in the industrial base. Assume most SaaS functions in the world are dumped on AI and automated away.

If estimates are remotely accurate, the trends in the chart above continue. Many Western nations are suffering similarly, with aging workforces and degraded manufacturing output, and China will continue to eat their shares along with other low cost countries (LCCs) under China's wing. The holistic vision of closing and leapfrogging this gap in under 20 years will require a massive investment in technological breakthroughs.

It should also be stated that a never ending supply of manufacturing slave labor is not tenable. Eventually all LCCs improve, wages will go up, and basic cost basis comparisons of labor will be not be advantageous without that labor being augmented by significant technology, or completely replaced with automation (Figure 4).

Exascaling is basically the opposite of how the American industrial base has been built, since a significant amount of key innovations were functions of conflict and defense spending. My simple framing is that:

(a) if we’re not in conflict

(b) we want to complain about how far behind we are in artillery, submarines, pharma, drones, etc.

(c) the companies developing these applications can not vertically integrate due to cost or contracts, and

(d) they rely on a supplier base that’s evaporating

…it’s a lose-lose, because no one will exist to build our applications, and no one will fund the factories to build them. Commoditizing the means of production via tech is the only way out. Or get into another world war.

The Role of Technology

If you’re not familiar with manufacturing, you might be surprised to learn that the marriage of tech and manufacturing is like oil and water. Yes, industry folk are excited about new tech, but market forces still apply, and adoption of major customers is their key driver. You have massive incumbents that make Computer-Aided Design (CAD), machine tools, robots, components, raw materials, etc., and they’re quite entrenched with major customers.

That said, the market is enormous and nearly impossible to monopolize. The one nice thing about tech in the context of manufacturing is that tech companies are purpose-built vehicles to scale a solution to a single problem, then expand into other offerings as they grow. This approach can be ultra effective in manufacturing because outcomes are measured by consistent, high-quality results. Notice how Tesla approaches this (Figure 5). The factory is the product.

If everyone from OpenAI and Google left their jobs tomorrow and decided to exascale the industrial base, what problems would you sick them on first? Replacing CAD? Make J.A.R.V.I.S.? Have them re-engineer the CNC machine? Build nuclear reactors attached to factories?

We don’t have an Operation Warp Speed for manufacturing. It’s too big to tackle. This is why the amount of resources (capital, materials, and talent) needed and opportunities are staggering.

How does tech augment each industrial base layer? The discrete buckets I mentioned previously look like this (not a complete depiction) (Figure 6):

Every company that is founded, funded, and scaled in these areas will have an additive or multiplicative effect on the ramp to exacaling. Can or do we measure total manufacturing output, worker productivity, volume of work produced by sector, or total capacity of all machines? Yes — more below in the Stats section. Even better, we can individually measure existing manufacturing processes and measure changes in efficiency and output to gauge the application of tech beyond just gross margin expansion.

There’s a theoretical limit to the cost of any manufacturing process, which can be broken down this way:

• T: Level of technological advancement.

• E(T): Efficiency as a function of technology.

• L(T): Labor cost as a function of technology.

• Cᵣ: Raw material cost.

• Cₘ(T): Overall manufacturing cost as a function of technology.

All this means is, "As you apply technology to a manufacturing process, it becomes more efficient, requires less labor, and approaches the cost of raw materials."

Here, Cₘ(T) decreases as E(T) increases due to technological improvements, while Cᵣ remains relatively stable or decreases slightly due to better resource management and recycling technologies.

OK — but what exactly is T? Imagine the effort that Tesla undertook to reduce its gigafactory manufacturing footprint by 40%, and what all their spending must have been used for to drive COGS down by 50%. Every process in that factory was scrutinized heavily, and likely rebuilt from the ground up. A majority of it with off-the-shelf equipment from suppliers, but also with the gigapress.

New equipment innovations can happen. Many don’t understand where innovation can even happen, though. A transformational process in manufacturing looks exactly like the above. Find something horribly inefficient, and rebuild it using modern technology. Ensure consistent, reliable, quality results, and scale it.

Bold founders, engineers, investors, and market dynamics will decide what T is, in this paradigm. We can build this modern industrial base, or continue to idly let the chance and our position in the global economy slip through our fingers. This is a fight against stagnation in the world of atoms.

The Opportunity

This is all to say that the direction of the world for manufacturers is, dare I say, more “service-based platforms” — aka converting their capex lines into opex that is ultra flexible, capital efficient, and powered by advanced software and automation built by technologists and customized to the manufacturer’s needs.

If you don’t believe me, that’s fine. I can tell you that many of the largest manufacturers on the planet are discussing this idea right now. Not only are we at an amazing cultural inflection point to take on hard problems in tech, but the macro window is going to be focused on this space for quite some time. The demand even now at the beginning of the exascale transformation is monumental.

The above is a simplistic view of where progress stands today, but also what the investment opportunity looks like. This S-curve represents how far we are from true Exascale™. The green zone will likely last five to 10 more years, and the overall macro cycle 20 to 30 years.

The breakout signal here will be the first layers and services of the modern industrial base coming online at scale, with lower level companies tackling fundamental manufacturing like machining, welding, sheet metal forming, die casting, assembly, tool and die, riveting, forging, and stamping. Simultaneous evolution in computing, robotics, ML, AI, and materials will swing the pendulum even further. The ability to simulate and optimize complex mechanical systems and designs is imperative; even more important is to connect the design of these systems to the machines that will fabricate them (Automatic Design for Manufacturability/ no more CAD).

The primordial soup of true Von Neumann replication systems will be built within our lifetimes, by us.

Outlook

You can stare at economic reports, charts, and conflicting articles on reindustrialization all day, or you could just go visit the Rust Belt, or any factory for that matter. Unfortunately, you’ll see and hear a similar story — the outlook is not stellar. If Intel is having issues with hiring semiconductor labor — a modern, incentivized, high skill trade — imagine what’s going in other areas of manufacturing. We’ve neglected the trades to the point where the replacement rate is 0.4: for every five retirees, only two replace them (if that).

There are 600,000+ jobs open in manufacturing, which is estimated to grow to 2.1m by 2030. Will a tech-enabled industrial base help manufacturers earn more if they can output more with less actual labor? Will workers make as much as software engineers at Google? Will manufacturers be able to ignore the labor rates in China or other LCCs? Can the workers learn 20 years of skills in one or two years? Will their actual jobs be less tedious, dangerous, and require a mix of high-tech skill like programming or robotics?

Yes. These are a few of the likely benefits of techno-reindustrialization.

We may even get to the point where it only takes a few people to operate a factory, just like running a SaaS company, and instead of an over-saturation of apps, we'll have a glut of production capacity waiting to chew through the next job. That capacity could be shifted around, to and from different economies. There will almost certainly be an immense need for this type of manufacturing when the space economy explodes. I’m largely convinced that we need it to realize any kind of sci-fi future (think tech scope of Blade Runner). That work will not be manual/by hand/via one-off shops anymore. If we ever want to land drop ships on Titan, we need to get serious.

A tech-enabled industrial base will be the salvation of America. By developing scalable, modern, and flexible production processes, we can build a machine that allows us to dream up and produce anything. To manipulate the physical world more rapidly than ever thought possible — even at the same speed as software.

The jobs, stability, economic growth, prosperity, and virtue of building has the potential to do more than just save our GDP. It will be the antidote required to pull many ailing American cities back from the depths, to their former glory.

Wrapping Up

Our priorities must be viewed through the lens of urgency and time. Do we currently need a stop gap to develop more industrial infrastructure? Absolutely. Do we need longer-term technological innovation targeted at the means of production itself? Even more. Can policy and other soft targets be effective in some areas? Definitely. Can we compete with the nearly $1t that China has poured into its own industrial base? Also yes (and spoiler: the ROI on US government manufacturing dollars is at least 2x).

The idea of the industrial base as an analog to AWS is apt. Obfuscate away the complexity, vertically integrate. There’s no such thing as too much automation, just poorly implemented automation. Manufacturing and capex aren’t dirty words. Spend the time, money, and resources to make manufacturing and capex more efficient, affordable, and scalable. We need cheaper factories to enable more companies.

Want to think about defense? Think about living somewhere where our single most intimidating deterrent (aside from nukes) is our ability to outproduce everyone on Earth — cars, toys, drugs, rocketry, and whatever else your country wants: we'll have it to you at the drop of a hat.

The most talented people in tech enjoy deeply violent challenges, so here’s my challenge: leave your SaaS job and go build something you’ll be proud of on your deathbed. Join a manufacturing startup, or start one. Some people in tech think there’s more glory to building AGI, but in reality it will be heavy industry that becomes indistinguishable from magic.

So go bang your head against the wall trying to figure out unsolved, insane, real world problems. Change the world atom by atom, because you can, and the future is waiting to be built.

-Aaron Slodov

--

Stats, errata

--

--

--

--

--

--