Making Sand Think: Intel & Silicon Sovereignty
Cathedrals of sand, venture nationalism, & the magic of semiconductors
Almost everything I've seen opining about Trump and the US government's new position in Intel is woefully misinformed about the nature of semiconductors and how critical they are for advanced economies and national sovereignty. Few understand how semiconductors work or how vital they are to modern national capabilities. And no wonder, it’s a weird niche topic. And you’re in luck! Because I’m a weird niche guy, I happen to know a fair amount about them.
Consider you might not know much about this industry. Consider what goes into creating these microscopic miracles exists outside your circle of competence, and before remarking on matters related to chips you should expand that circle a bit. This essay will do that. You will leave here with your circle enlarged.
I took a special interest in semiconductors about six years ago as I learned how essential and sophisticated their supply chains are. Credit and gratitude to Gavin Baker of Atreides Management as well as Brinton Jones and Jon Bathgate of NZS Capital for all the insight, podcasts, writing, and thoughts they’ve put out over the years on semis.
I listened to this music while writing.
The Intel takes are predictably tribal and negative in their own horseshoe-theory kinda way: “This is socialism!” from the Right, “Aha, fascism!” from the Left, and both miss the point by only perceiving actions through an ideological filter. This has nothing to do with economics or partisan narratives, but whether America wants to become technologically independent or remain a digital vassal of East Asia.
Don’t let the simple heuristic of ‘if the government did it, it’s bad’ shepherd all your thinking. While often an accurate guiding light, to always believe the same thing regardless of circumstances is a religious conviction, not a thinking one. It’s a reaction, not reflection. When China's military can hold a gun to your tech stack, it's time to incorporate some nuance.
The Invisible Hand of markets and the Iron Fist of geopolitics operate on different game theory; sometimes you must read the latter’s grammar to be literate.
This essay will explain why Intel isn’t just any business, why semis aren’t just any industry, and why US actions stem not from financial motives but military rationale in service of national autonomy. Think of this INTC position not as an investment but as fortifying digital borders.
The Intel of China is known as Taiwan Semiconductor (TSMC). There’s an element of warfare in what took place this past week if you look beyond the boilerplate political optics. Come, let’s look beyond them together.
You will also learn a great deal about the intricate magic that goes into semiconductors, here’s an outline of what’s to come:
I. Silicon Realpolitik
Semiconductors: the most important industry by far for technological capacity
II. Geopolitical Motivations, Not Financial
Venture Nationalism, not Venture Capitalism
III. Intel's Fall from Grace
Spreadsheet brains, missed transitions, and strategic myopia
IV. Terms
Semiconductor definitions you should know
V. TSMC Supremacy
The raw dominance of the Taiwanese giant
VI. What if TSMC Vanished?
How the world’s most powerful companies rely on one island
VII. Cathedrals of Sand
The unparalleled engineering feat of lithography: teaching sand how to think
VIII. The Mechanical Beauty of ASML
The company that builds machines that conduct laser symphonies
IX. Concluding: US Military Reliance on TSMC
Where advanced chips show up in warfare and national defense
Disclosure: long TSM, ASML, AMD, AVGO, LRCX, NVDA, AMAT (basically all the equities I own are part of the semiconductor supply chain)
Silicon Realpolitik
Imagine if, at the height of the British Empire, all of England’s coal had to be imported from a small island off the coast of France that Napoleon claimed as rightfully his. That's essentially America's position with semiconductors today. Except semis are vastly more critical than coal ever was to the Industrial Revolution.
The reason for this Intel stake by the US government is not financial. The motivation is very straightforward and apparent once you strip away political feelings: state-of-the-art chipmaking is a linchpin for global power.
Semiconductors are the most important industry, by far, for technological capacity. Not oil, not finance, not software. Silicon.
On these tiny wafers of sand, atoms are arranged into transistors smaller than viruses. Switching billions of times per second, forming the substrate on which modern power and force rest.
Every F-35, trading algorithm, and classified communication runs on this atomic choreography. And we’ve outsourced the show to an island Beijing considers a rogue province. Sometimes the ol’ invisible hand can give you a limp wrist if you worship it too much!
“If Intel falls further behind and leading-edge semiconductor manufacturing becomes concentrated in Taiwan then Taiwan will become geopolitically important in a way that the Middle East never was.”
“Intel is one of the most important national assets for the United States.”
-Gavin Baker
The company that invented the microprocessor, powered the PC revolution, and should have owned mobile and AI, instead spent years optimizing for quarterly earnings while TSMC optimized for long-term dominance. Because Intel was run by Spreadsheet Brains who couldn’t see past a couple quarters, the US is now structurally reliant on a tiny island 100 miles from mainland China for the chips that power its AI, weapons, and infrastructure.
A superpower cannot allow another country to control its ability to think. It cannot outsource the backbone of its technology. In the 21st century, these chips are the neural substrate of national cognition.
This is not trivial; it transcends supply chain hiccups or quarterly earnings reports and impacts the neural architecture of national power. Because of Intel's failures, we are beholden to TSMC for cutting-edge chips: exposed to China’s mercy for our best tech. Every Nvidia H100 powering LLMs, the processors in our missile defense systems, all emanating from one Chinese island. Do you see how this is not a normal business? Can you feel the intense vulnerability here?
All businesses and government activity are not created equally. Trump didn’t exactly take a stake in General Mills because he thinks Frosted Flakes should be rebranded to Tremendous Flakes (“many are eating this!”). We’re talking big-potatoes, existential stuff here.
Trump’s right to consider it a matter of national security that the US sucks at chip manufacturing. When Nvidia is handcuffed to TSMC for their finest GPUs, when our AI and cloud compute hinges on factories within artillery range of the People’s Liberation Army, we're not discussing market competition anymore. We're deciding whether America maintains technological independence.
Milton Friedman is cool and all, but if you cite comparative advantage or “free markets” in this context, you have respectfully disqualified yourself from the conversation. This is not a game of gross margins. It is not one of “whoever makes it cheapest wins!”. That’s perceiving the world only through Invisible Hand thinking, and the Geopolitical Fist carries a different philosophy and objective. Markets are vital circulatory systems for complex societies, but don’t worship them regardless of the competitive dynamic.
Truly sovereign nations cannot outsource the pivotal elements of semi supply chains: a harsh reality people are incrementally realizing.
“Currently, however, none of the world’s most advanced chips are produced in the United States; Taiwanese companies dominate the market. U.S. policymakers, worried that the United States would lose access to semiconductors during a conflict with China over Taiwan, have turned to industrial policy to address this issue. Such an approach, however, comes with the inherent trade-off of maximizing economic efficiency versus bolstering national security and resilience.”
-Council On Foreign Relations
Geopolitical Motivations, Not Financial
To illustrate why this is not financially motivated, let’s concretely understand the fundamental nature of government and companies under its jurisdiction: the US is functionally already a stakeholder in every company operating in its borders.
Why? It collects mandatory dividends from all of them, something of a national dividend. This yield requires no board seats, no corporate governance, heck they don’t even have to hold any shares!
Whereas dividends are completely voluntary for shareholders (you do not have a ‘right’ to a dividend if you're a shareholder, it's purely a board decision paid at their discretion), they’re not voluntary for the US government.
The government already extracts more reliable returns from Intel than most shareholders ever will, yet critics worry about 'picking winners and losers' as if the tax code isn't the ultimate expression of state preference.
A regular dividend is paid out of net profits (bottom-line earnings, not top-line revenue), and the same is true for taxation. The US government takes roughly 20% of net income from its companies, every year; they are the only ‘shareholder’ with a true ‘right’ to profits, whether the board agrees or not! And they don't need to hold a single share. As long as the company exists, the government receives a perpetual dividend that correlates with the growth and success of the company. Taxation is a de facto dividend with different branding.
A sovereign government’s finances are not like a lemonade stand, its balance sheet isn’t like yours or mine. You must see the decision through the eyes of the entity that made it, not by projecting Finance 101 concepts about why you would make such a move. Violence monopolists do not abide by rules, they create them - their calculus is completely different than anybody else’s.
When the US government takes an equity stake, we can readily infer it has nothing to do with financial ROI, as the tax code already handles that. What’s going on is more substantive. The “government shouldn’t pick winners and losers” critique doesn’t apply here. Trump & Co. aren’t picking Intel because they want it to economically win, but because Intel losing means America loses self-directed computational capacity.
It’s venture nationalism, not venture capitalism: when the state invests not for return on capital but return on sovereignty.
Chip capacity is genuinely of consequence for geopolitical power: it is the oil of the digital age. And the US is completely reliant on Taiwan (TSMC) and the Netherlands (ASML) for its best technology, and only one operates in hostile territory. If you outsource your most indispensable infrastructure to an adversary, you are a fool; thwarting this is not socialism, it’s not fascism, it’s avoiding fool-ism.
A 10% equity stake provides a seat at the table for company strategy: possible board representation (or observer status), soft veto power, the ability to whisper sweet nothings about national security into management’s ears. That's why Trump wants this. The government says it’s a “passive position”, and if you believe that you deserve a kiss on the forehead and a tummy rub.
The US couldn’t care less about capital gains. It doesn’t need Intel shares when the IRS extracts its pound of flesh anyway. This position’s only purpose is strategic influence in what’s legitimately a national interest.
This move is aptly understood as a low-intensity warfare countermeasure against Chinese dependence. It's the industrial policy equivalent of trying to get your pawn to the other side of the chessboard so it can become the queen you know it deserves to be. If China can simply declare reunification with Taiwan and dictate leading-edge chip procurement, a naval blockade could bring Western tech to its knees. Imagine if all advanced technology was subject to CCP approval; Intel acts as a defense against this.
>"The government can't have a stake in a national champion! Because then it will be run badly!"
I wish every company could be as incompetently run as TSMC is. What an Orwellian dream that would be.
Why does Intel need a babysitter? We didn’t reach this unfortunate situation by accident; an overnight success is a decade in the making, and so is an overnight fuckup.
Intel’s fall is a case study in how financial engineering can undermine actual engineering. While TSMC solved fab yield problems, Intel solved for quarterly earnings. While TSMC shrank node size, Intel shrank share count. Now one controls the future, and the other is in government daycare.
After reviewing Intel’s demise, we’ll learn just how remarkable and beautiful these byzantine little sand chips are!
Terms
Before we review the landscape and lore of semiconductors, here are some definitions you should know:
A semiconductor is a material with electrical conductivity between a conductor (like copper) and an insulator (like glass). It conducts electricity under certain conditions, making it ideal for controlling electrical signals. This property is the backbone of modern technology.
Silicon is the semiconductor material used as the substrate for most chips. It’s purified and sliced into thin disks called wafers, which serve as the base on which microchips are built. These wafers undergo numerous processing steps to become functional integrated circuits (ICs).
A fab (short for fabrication plant) is a sophisticated manufacturing facility where silicon wafers are processed into finished semiconductor chips. It involves a highly complex supply chain consisting of:
Photolithography
Ion implantation
Etching
Deposition of materials
Packaging and testing
Building and maintaining a fab costs tens of billions and requires cleanroom environments hundreds of times cleaner than hospital surgery rooms.
A foundry (like TSMC) produces chips designed by others. E.g. Apple, Nvidia, and AMD design chips; TSMC manufactures them.
Photolithography is a technique using light to etch microscopic patterns onto silicon wafers.
Node Size, in nanometers (e.g. 5 nm, 3 nm) refers to the smallest features that can be etched onto the chip. Smaller nodes mean faster, better chips, and require tools like Extreme Ultraviolet lithography.
EUV (Extreme Ultraviolet) Lithography is a cutting-edge chipmaking process using 13.5 nm wavelength light to etch ultra-small transistor features.
Only one company, ASML in the Netherlands, makes EUV machines
They cost between ~$150M-350M+ each, weigh ~180 tons, and are shipped via special cargo aircraft, freight, and multiple trucks (!!)
Semiconductors are the closest thing to magic that exists. Continue reading and perhaps you’ll believe in magic too.
Intel's Fall from Grace
Intel famously passed on fabricating chips for the iPhone in 2007, dismissing mobile as unprofitable; a decision that redirected billions in revenue to TSMC and established the foundry model's dominance. This was a harbinger of the incompetence to come.
Intel's struggles have implications beyond profits. For decades, American technological and economic supremacy rested partly on Intel’s manufacturing superiority. If Intel cannot recover, Taiwan could become the most geopolitically important country at any single point of time in history: imagine if one country was basically the only source of oil.
For nearly half a century, Intel reigned supreme in semiconductors and created a mini empire. Their chips were consistently faster, more efficient, and more profitable than competitors’. Today, that empire lies in ruins.
This reality has belatedly awakened US policymakers, though their response has been confused. Current policy restricts chip sales to China while allowing sales of equipment China needs to build its own… “We refuse to sell you airplanes but will help you build your own.”
Intel has not only lost its edge to TSMC but finds itself in the humbling position of partnering with second-tier foundries just to stay relevant. This is the sordid tale of how strategic missteps, cultural arrogance, and stagnation took Intel from kingmaker to supplicant.
The Lost Decade
The seeds of Intel's downfall were planted during Brian Krzanich's tenure as CEO from 2013 to 2018. When Krzanich took over, Intel faced a crucial inflection point. The cost of building cutting-edge fabs was skyrocketing, and the industry was shifting toward a model where companies designed chips but outsourced manufacturing to specialized foundries. Intel had a prime opportunity to transform itself into a foundry serving external customers.
Instead, Krzanich chose the path of least resistance, prioritizing quarterly profits over next-generation fabs and EUV. This pleased Wall Street, but the myopia planted seeds of Intel’s demise. By neglecting EUV lithography, Intel fell behind for the first time in its history. By 2018, TSMC was producing 7nm chips while Intel struggled with delays in its equivalent 10nm output.
Throughout Intel’s existence, it had maintained a 6-15 month technology lead, translating to roughly 20% better performance, power efficiency, and cost. No more.
Intel's old rival, AMD (which had nearly collapsed a decade earlier) was resurgent under CEO Lisa Su. But unlike in 2006, when AMD's chips were hamstrung by limited manufacturing, AMD had gone ‘fabless’, outsourcing all production to TSMC. This created a perfect storm: AMD now had both better chip designs and better manufacturing with TSMC.
Insane fact: Lisa Su and Jensen Huang, the CEO of Nvidia, are first cousins once removed. Jensen’s mother is the sister of Su’s grandfather, making Huang the older cousin relative to Su.
They didn’t grow up together. Despite the family tie, Su says they met later in their careers, first encountering each other at an industry event, not a family gathering!
In the past, when AMD had superior products, Intel could strong-arm customers by threatening price increases if they switched suppliers, knowing AMD's limited capacity couldn't meet full demand. Now, with TSMC's formidable bandwidth bolstering AMD, Intel lost its leverage. The bully became the bullied.
“Intel guessed correctly for the last forty years, and then they guessed wrongly. They probably inserted extreme ultraviolet lithography too late. That resulted in them losing process leadership to Taiwan Semi.
This is one of the highest stakes games being played in the world today: In the logic space between Taiwan Semi, Intel and to some extent Samsung. Even in the processor design space Intel, AMD and Nvidia are making giant two, three, four billion-dollar bets on designing a processor, making it work on a manufacturing process. They have to continuously guess right.” — Gavin Baker
A Culture of Arrogance
Beyond strategic mistakes, Intel suffered from deeper cultural rot. For decades they operated with a mindset where their engineers called the shots: client design teams had to work around manufacturing's limitations, use Intel’s proprietary tools, and accept whatever chips they delivered. Designers had to use increasingly archaic in-house software and accept production quirks. This worked while Intel was dominant.
The thing about foundries that Intel missed is they're also customer service organizations. TSMC doesn't dictate to customers, they ask "how can we help you build this?", while Intel behaved like a semiconductor tyrant.
“…the more fundamental issue is that a foundry is, as I wrote, a customer service organization: an entity like TSMC adapts to customers’ designs, not the other way around. They use industry standard design software. They have extensive libraries of IP that make designing a chip more akin to assembling a collection of Lego blocks. They ship when they say they will ship, and they run the fab for which a chip was designed forever.
Intel did none of these things, and had a mentality and culture that ran in the exact opposite direction: in a foundry, manufacturing is not king but a servant; customer sales is not about “take-it-or-leave-it” but “let us help you solve your problem.” I was — and frankly, remain — dubious about Intel’s ability to create that sort of culture internally, which is why I advocated for an acquisition, first of Global Foundries, and then of Tower. Thanks to its decade delay Intel didn’t have time to learn how to serve customers: it had rapidly obsoleting fabs that needed to be filled as soon as possible, if the company ever had hope of making enough cash to fund its push back to the leading edge.
Unfortunately China blocked the acquisition of Tower, in what I suspect was retaliation for U.S. restrictions on China. Worse, from what I have heard Intel responded by starting to sell a lot of old equipment at rock-bottom prices, which usually ended up in China.”
— Ben Thompson, Stratechery
Serving customers with this mentality is like opening a restaurant where the chef decides what you're eating regardless of your order, "You want salmon? Too bad, we're making beef today, and you'll use our proprietary fork to eat it. You’re welcome!". Meanwhile, TSMC built a culture of radical accommodation, redesigning manufacturing processes to facilitate unique client designs. The difference between a restaurant run by a temperamental cook and one where the waiter comes to your table, asks how you'd like your atoms arranged today and gives you a back massage.
Intel's story is one of hubris and missed transitions: assuming its dominance was permanent, its integrated model superior, and it could assert its will over the industry forever. They spent decades winning by default until these maladaptive practices came home to roost. When that advantage evaporated, they were exposed as an emperor with no clothes, no backup plan, and no clue how to compete on equal terms. By the time leadership recognized these assumptions were fatal, competitors had pounced.
None of this was inevitable. At any point in the last 15-20 years Intel could have recognized the writing on the wall. They could have embraced the foundry model or invested in EUV. Instead, they chose they chose the cash-flow statement and the status quo.
Semiconductor cycles span decades, not quarters. Intel needs substantial time to rebuild its manufacturing, transform its culture, and regain customer trust; Wall Street's transitory thinking doesn’t comport with this geopolitical reality. Intel’s renaissance necessitates breathing room and long-term shareholders who won’t dump it if revenue growth or margins take a hit over the next few years.
This is why the United States government holds a position in INTC; it cares not about quarterly financials, but telescopic improvements and regaining what was lost.
If Intel fails to recover its semi leadership, America loses its last foothold in pioneering chip production, ceding control of the 21st century’s most critical technology to Choyna.
Despite its failures, Intel remains the only American firm capable of achieving leading-edge production. It’s a wounded giant with the raw tools to rebuild if given time and support.
Until then, it’s TSMC’s world and everyone else is playing catch-up.
TSMC Supremacy
TSMC isn’t so much a chip producer as it is a place where chipmakers co-design the future. Their tight loop with customers becomes a recursive advantage no one else currently matches. Some fun facts:
The country of Taiwan produces ~90% of the world’s most-advanced chips (<10nm)
Packaging moat: For AI-class chips, advanced packaging is the bottleneck, and TSMC’s capacity has been effectively sold out while it expands. Competitors can’t ship systems at the same pace without this packaging throughput.
For total manufacturing share (leading-edge chips plus older models) TSMC has ~67% vs. Samsung’s ~8–9%; Intel’s foundry share is small (mid-single digits).
TSMC's R&D spending in 2024 alone ($8.5 billion) exceeds Intel's entire foundry revenue
Samsung, despite massive investment, has struggled with yields on advanced nodes, losing Apple as a customer and falling further behind TSMC.
What if TSMC Vanished?
To learn how important something is, picture what the world would look like without it.
Mega-cap tech companies depend on TSMC either directly for their own chips or indirectly through Nvidia, AMD, or Broadcom. If TSMC disappeared, the modern compute stack collapses.
Here are the top-ten tech companies by market cap and how each leans on TSMC. All are American save for two, I’d say this counts as a topic of national interest.
Nvidia (NVDA): With more than 70% market share, Nvidia dominates AI chip sales. It designs the GPUs (H100/H200, Blackwell B200/GB200) that power today’s AI boom; these chips are constructed at TSMC. If TSMC vanished, NVDA’s flagship supply would seize and cloud/AI buildouts would stall.
Microsoft (MSFT): Azure runs heavily on TSMC-made GPUs, and Microsoft’s own chips (Maia AI accelerator and Cobalt CPU) are fabricated on TSMC 5nm and advanced packaging. Without TSMC, both third-party and in-house silicon pipelines are hit. Source.
Apple (AAPL): All A-series (iPhone) and M-series (Mac/iPad) performance comes from TSMC’s 3nm family. Losing TSMC would cripple Apple’s product cadence and performance edge. Source.
Google (GOOGL): Google’s TPU roadmap is produced at TSMC, and Pixel’s Tensor G5 moved to TSMC’s 3nm. No TSMC = curtailed TPU capacity and smartphone silicon plans. Source.
Amazon (AMZN): AWS relies on TSMC both indirectly (Nvidia GPUs) and directly (in-house Graviton/Inferentia/Trainium chips). A TSMC outage would choke AWS. Source.
TSMC (TSM): They’re the 6th-biggest tech firm, and while this market cap represents the estimated sum of all cashflows TSMC will generate discounted back to today, it woefully understates its indispensable relationship to the rest of the technology we rely on.
Meta (META): Meta is deploying its own AI accelerator (MTIA) with TSMC and buys huge numbers of TSMC-made Nvidia GPUs. Without TSMC, Meta’s AI infra build would slow dramatically. Source.
Broadcom (AVGO): A fabless giant (ASICs, custom AI silicon for hyperscalers, Apple parts). Its 10K flags acute reliance on the Taiwanese giant. Source.
Tencent (TCEHY): Their cloud/AI depends on Nvidia GPU clusters (TSMC-built).
Oracle (ORCL): OCI’s AI is built around TSMC-backed Nvidia GPUs, and Oracle also runs Arm servers from Ampere (fabricated at TSMC 7nm)
Without TSMC, there’d be immediate global shock at state-of-the-art nodes (<~10 nm): Samsung cannot backfill the 90% advanced market share TSMC supports. Shipments of cutting-edge CPUs and GPUs would collapse for years, not months.
Global GDP Impact: The Institute for Economics and Peace (IEP) estimates a full-scale conflict over Taiwan could cause $10 trillion in global economic losses; a blockade scenario causing damages of ~$2.7 trillion.
NOT LISTED: every private AI-focused firm (OpenAI, Perplexity, Anthropic, xAI, etc.) that lives on TSMC-made chips.
Cathedrals of Sand
EUV lithography
Chipmakers “photograph” patterns onto a light-sensitive coating (photoresist) on silicon. To etch ever-smaller features, you need shorter light wavelengths. EUV uses 13.5 nm light generated by blasting tin microdroplets with lasers inside a high-vacuum system. It leverages ultra-precise multilayer mirrors, each with 100+ atomic-scale layers, to reflect and focus beams onto silicon.
Read each part of this carefully:
To generate EUV light, machines fire a CO₂ laser at a stream of tin droplets, vaporizing them into plasma at ~400,000 °F: that’s about 40x hotter than the surface of the sun.
The laser (made by a company called Trumpf) that hits these tin droplets fires at 100,000 pulses per second and requires thousands of watts of power: more than some small factories.
This plasma emits 13.5 nm light, but EUV light is absorbed by almost everything, including air and lenses. So the whole system must operate in an extreme vacuum.
EUV light is reflected off ultra-polished, multilayered mirrors made by a company called Zeiss, each with over 100 layers of molybdenum and silicon aligned to atomic precision.
The Zeiss mirrors used in EUV reflect only ~70% of the light, with a dozen reflections required.
After multiple reflections (typically 10-12 mirrors in the optical path), only ~1-2% of the original light reaches the wafer.
The entire process is fraught with challenges called "stochastic phenomena" (random variations at the atomic level) that can cause defects in chip patterns. This requires advanced metrology and control systems.
Imagine machine-gunning molten metal with light pulses 40x hotter than the surface of the Sun, one-hundred-thousand times every second, to create a controlled plasma that emits the specific light wavelength needed to etch patterns the size of DNA onto silicon.
Inside your phone resides a miracle.
Silicon comes from sand. We are taking pieces of sand and etching into them cosmically minute hieroglyphs, the symbols of technology known as transistors, the language of bits, sticking them into machines, and MAKING SAND THINK. SAND WAS NOT SUPPOSED TO THINK. This is either an affront to God or we’re honoring him in some strange way.
I’ve been to Rome, gazed upon the Sistine Chapel. Wandered the Vatican. I have seen the Eiffel Tower. Been to the Sagrada Familia in Barcelona. The Grand Canyon in Arizona. Walked the Parthenon in Athens. Explored countless cathedrals and museums across God knows how many European cities. I’ve lived in NYC and witnessed the most majestic towers that man has ever built.
While I lack the visual gifts to gaze upon something this intricately small, the awe I felt learning what we’ve accomplished with semiconductor engineering surpassed them all. And what’s more, we have turned this spectacle into banality. It’s omnipresent, in every device you touch, including the one you’re reading this essay on right now…. there’s a Sagrada Familia only tens of atoms in size, etched into sand, operating in front of you.
This is a symphony of quantum optics, thermal stabilization, and electromagnetic wizardry. From an engineering perspective, I genuinely think it is the most impressive human feat of all time.
It’s thought there are about 300 people on Earth who truly comprehend how to make EUV work at production scale. Not 300 companies. Not 300 teams. 300 human beings. This mythical scarcity adds to the grandeur of it. How many architects alive know how to build a skyscraper or cathedral?
Our technological civilization rests on a type of divine-secret knowledge: a small group of atomic priests, that could fit into a lecture hall, handle and summon this chimera. There are probably, I don’t know, several thousand or so people who have ever lived who comprehensively understand how to execute this.
Only five semiconductor manufacturers operate EUV at scale: TSMC, Samsung, Intel, SK hynix, and Micron. These companies collectively produce all of high-end logic and memory devices. And TSMC has a 90% market share here.
These 300 atomic priests steward knowledge past civilizations would consider mystical. Orchestrating microscopic matter that seems like science fiction. And the temples where this high clergy of techno-civilization produces its scrolls resides right next to the People's Liberation Army. And you think Intel is just another company…
A bird's eye of Barcelona and the Sagrada Familia:
Here are semiconductors up close. Nature's blueprints at every scale. Beautiful.
Nanometers in Context
An important note: “3 nm” is a node name and an industry marketing designation, not a literal measurement of every feature: dimensions vary by layer. For TSMC’s N3 process (3nm), the actual gate pitch is reported to be 45nm.
When you hear massive numbers like hundreds of trillions, the human mind doesn’t adequately appreciate how big those figures are. Similarly, we don’t really grok how incomprehensibly microscopic and excruciatingly precise this process is. How small is a nanometer? Here’s some help contextualizing that.
Visible light’s wavelength: ~400–700 nm.
EUV light: 13.5 nm
DNA double helix: ~2 nm across
COVID: ~70–100 nm across
Human hair diameter: 70,000 nm on average.
Human hair daily growth rate: 350,000 nm/day on average
A human red blood cell is about 7,500 nm in diameter
A single atom is ~0.1–0.5 nm in diameter
EUV steers 13.5 nm light off nanometer-smooth mirrors in giga-vacuum to print features smaller than a red blood cell by over a factor of 10: competently executed over billions of transistors per chip. Magnificent little sand cathedrals you cannot see yet feel their presence as they perform constantly in all that surrounds you.
This is why it’s as easy as saying “build them here”; this expertise requires decades to cultivate. The US must play the long game, and Intel needs shareholders who let it. Cutting-edge semiconductors are the most sophisticated engineering feats on Earth.
The Mechanical Beauty of ASML
ASML is the sole supplier of EUV lithography systems. This monopoly means that the core knowledge of integrating the entire system resides within one company and its closest partners: ASML manufacturers complete EUV machines, Zeiss produces the mirrors, and Trumpf provides the lasers.
The lore of EUV:
EUV technology was developed through US government–funded research, conducted at national labs such as Sandia and Lawrence Berkeley in the 1990s.
The Department of Energy granted ASML licensing rights and access.
It took over 20 years for EUV lithography to become commercially viable. ASML spent ~13 years developing prototypes and another decade to reach high-volume production.
Due to US government IP ownership, export controls apply to ASML's sales and the US controls where these machines ship, including restricting sales to China
What goes into these machines? Costs?
Price: Low-NA EUV scanners have been quoted around $150–$200M each; ASML’s new High-NA EUV costs ~$370M each.
Units sold per year: ASML recognized revenue on only 44 EUV systems in 2024, and 53 of them in 2023. Source.
Each EUV machine is about the size of a bus, composed of ~100,000 parts, and weighs around 180 tons! It’s shipped in pieces that include ~40 freight containers, ~20 trucks, and ~3 cargo planes (often 747 freighters), then reassembled onsite.
Really let those numbers wash over you. Imagine the extraordinary intricacy, coordination, and absolutely staggering degree of complexity that we have made prosaic, scalable, and dropped into every cell phone and computer on the planet.
I don’t believe there’s anything else man has ever created that’s this astoundingly complicated, delicate, and labyrinthine. It’s one of the more holistically beautiful and formidable things we as a species have ever accomplished. Cathedrals etched in silicon.
Concluding: US Military Reliance on TSMC
Leading-edge chips impact US defense. I’m sure I’m overlooking their applications in some ways, so consider this a conservative list:
AI for surveillance and targeting: Systems like Project Maven rely on bleeding-edge chips to rapidly scan images, videos, and signals, identifying threats, vehicles, or targets in real time. These AI models demand serious compute horsepower. The same GPUs running ChatGPT are used here. TSMC sends its regards.
Battlefield coordination and real-time data sharing: Modern warfare leans on software-driven communication networks. Think encrypted radios, satellite links, and drone control: all depending on high-speed chips to relay sensor data and make split-second decisions.
Electronic warfare and radar systems: Jamming enemy signals, detecting stealth aircraft, tracking missiles, all hinge on chips that process electromagnetic data at rapid speed. The more powerful the chip, the smarter and faster these systems get.
Missile-tracking satellites. Space warfare, basically.
America maintains its nuclear arsenal through simulation, not live testing. The El Capitan supercomputer, which models nuclear behavior, runs on cutting-edge TSMC chips.
What the US loses without TSMC access:
AI tempo and scale: Fewer state-of-the-art GPUs/NPUs → slower model training/fielding, decreased battlefield speed. The BIS imposed controls because these chips impact military preparedness and capabilities.
Edge performance: Drones, jammers, and missile seekers would have to use older, less efficient chips. This means bulkier gear, lower battery life, slower response times. Heavier, hotter, and less-capable processors = slower and less lethal.
Space latency: On-orbit processing for hypersonic/missile tracking would lag. More data would need ground processing, increasing latency. Speed kills, and lack of speed kills too… just not the way you’d prefer if you’re the slower one.
Key national-security computing projects and F-35-style weaponry would be stymied.
State-of-the-art chips are now a rate-limiter for core military functions and a vital tactical asset. Determining how fast you can see, think, move, and strike, they are like neurons on the battlefield. If America loses access, it loses time, scale, and edge: a shift in the balance of power if others have them and we don’t.
The war for semiconductors isn’t about stock prices, but whether America can compute its own future. Intel is a handicapped goliath yet remains the only US company with a prayer of challenging TSMC. The government's stake is a recognition that some capabilities are too vital to outsource, and some dependencies too precarious to accept.
We’ve taught sand how to think, and these microscopic miracles are crucial for economic and military might. Intel isn’t a matter of profit margins or capital gains, but of technological supremacy and the game-of-inches that is global competition.
In the 20th century, we fought over oil; in the 21st, we’ll fight over beautiful pieces of sand.
Subscribes and shares are very much appreciated. If you enjoyed this essay, please give it a like and share.
I’m building something interesting, visit Salutary.io for a wealth of financial history.
You can show your appreciation by becoming a paid subscriber, or donating here: 0x9C828E8EeCe7a339bBe90A44bB096b20a4F1BE2B
«Only five semiconductor manufacturers operate EUV in high-volume production: TSMC, Samsung, Intel, SK hynix, and Micron.»
They are mostly asian and that is because in "the west" it is far more profitable to trade in property and stocks and debt than to make chips, and anyhow manufacturing is reckoned to be much more easily infected by labor unions; asian investors are less reluctant to accept lower returns. However the economy of China-Taiwan has been dominated by trading in property for some decades which has been far more profitable so TSMC is an exception that may not last that long.
Great essay. I find myself increasingly annoyed by these CEOs who lack vision and choose short term financial wins over long term strategic wins.
It’s a little reassuring that the government seems to have properly diagnosed the problem. Hopefully they can steer the ship in the right direction with enough speed to save us from catastrophe.
Are there any books you’d recommend on this topic?