Transcription
No country on Earth can build an advanced AI chip alone. Or can it?
Why does Taiwan get mentioned in every potential geopolitical crisis? Why is this specific island the heart of the semiconductor industry? Why can't China, with all of its money and manufacturing power, build its own advanced chips? And why does Taiwan need the Netherlands, not Germany, not Switzerland, not anyone else in Europe, to keep the world dependent on it?
This is the new episode in the business of AI, the business of chips. And by the end of it, you will understand the entire semiconductor business, every country, every dependency, every choke point to really understand who controls this business and why one broken link can bring down the electronics market across the globe. Let's dive in.
The entire system runs on eight critical links. Each one dependent on the next.
Dependency one, electronic design automation software or EDA. Before a single atom of silicon is touched, the entire chip must be designed. This is done entirely in software. Two American companies, Synapsis and Cadence, each hold 35% of the global EDA market. The third, Seammen's EDA, American-based by owned by Germany, holds around 15%. Every chip designed on earth, every Apple A series chip, every custom ASIC, every Nvidia GPU is designed in this software.
But why in a world overflowing with SAS companies is EDA controlled by only these three? Because EDA is not ordinary software. It took 40 years to build. The hardest thing about it is proving mathematically that a chip with billions of transistors will work properly before it gets manufactured because the cost of error is 10 to $50 million per design.
Dependency two, the silicon. The wafer is a flat disc of pure silicon on which every transistor is built. A 300 millimeter wafer used for advanced chips must be made from silicon with purity of 99.99999. Just for comparison, 100% purity does not exist in nature. So this is as close as it can humanly get. If you misplace a single contaminated atom in the crystal structure, it'll propagate as a defect through every layer built on top of it. Growing a crystal of this purity requires exactly the right temperature, rotation speed, gas atmosphere, and slicing and polishing it to a flatness deviation of less than one nanometer across the entire surface.
There are only two companies in the world that manufacture silicon of this purity at scale and it's Shinu Chemical and Sunko. Both of them control 72% of the global silicon wafer market. Both of them are Japanese. Now, why are they Japanese? Because Japan has a history of precision materials manufacturing that started back in the 70s. The reason is Japan and not any other place in Asia is because they have proximity of chemical and equipment suppliers in the same industrial region and a manufacturing culture that is known for precision. Their key selling point is lotto lot consistency over decades.
Dependency number three, photoresist. Photoresist is the chemical ink coated onto the surface of the silicon wafer. Without this ink, lithography machines cannot transfer any pattern onto a chip. But why is this ink hard to produce? Extreme ultraviolet photoresist must have a purity of 99.99999. The same 999s as the wafer but in a liquid chemical compound. Getting to this precision of purity is decades of research in chemical synthesis. There's only a handful of countries on the entire planet that specialize in this type of chemical engineering. The Japanese hold a 100% monopoly for extreme ultraviolet photoresist. The key suppliers are JSR to Shin etsu chemical. Japan knows that it's a weapon and they're already using it as a geopolitical lever. Without photoresist, a semiconductor plant physically cannot run lithography. Full stop. In 2019, they restricted exports to South Korea, which immediately threatened Samsung and SKH Highix, one of the three companies on the planet that produced high bandwidth memory. And in late 2024, Japan deployed expert controls to China.
Like many of you, I have been thinking about starting a company and I keep coming back to the same idea. I want it to be a physical product. I've gone through a bunch of directions, but what sticks is something modular, customizable, high quality, and design forward. Something that people would actually use every day. So, I thought perhaps a furniture brand, a premium furniture brand designed specifically for people who work from home. But if you've ever tried to launch something like this, you realize pretty quickly that the product isn't the hardest part. Because before you spend a dollar on manufacturing, you need to know if people are actually going to buy it. And the fastest way to test it is to make the idea feel real. The brand, the visuals, creative direction, ads, content, distribution plan. Basically, a smallcale marketing campaign. And that's what I'm going to use Hicksfield Supercomputer for. I will connect it directly to Claude and use it as three different roles. First as a marketer because I want to pressure test my idea. Who actually buys premium work from home furniture and what it does is it narrows it down to highincome professionals optimizing for a focus aesthetics and status. The second as a designer in my brand I want taste, I want modern. I want sleek. That informs the brand and creative direction. What should the brand feel like? What kind of visuals signal premium without saying it? And in return, I get the visual direction, ad concepts, and product storytelling. And third, as a content strategist. It breaks down the hooks, the video structures, the formats that already work in this market for this type of product. And most importantly, it adapts them to my specific idea. What it does is it lets a solo founder operate as if they're a small company. You still need the judgment. You still need taste. and you still need to decide what's worth building, but you're not blocked by execution anymore. If you're trying to test an idea or build something, Hicksfield is a whole marketing and creative agent allin-one that can help bring any business to life before you make your first investment. And if you want a simpler version of it, you can use a standalone service from Hicksfield called Marketing Studio if you want to focus specifically on the content assets for your products. The link is in the description.
Process gases. Every chemical step in chip fabrication requires gases. Semiconductor grade gases must be pure, free of moisture, free of oxygen, free of hydrocarbons, and free of metal contamination. Neon gas used in lasers that power lithography machines was 70% sourced from, of all places, my native country of Ukraine before the 2022 Russian invasion. Why Ukraine? The thing is neon is present everywhere in air. The problem is that extracting and purifying it to semiconductor grade purity requires a specific industrial setup. You need massive equipment that processes enormous volumes of air to separate gases. And those are only economically viable when attached to something that already needs that infrastructure. That something is steel mills. Ukraine inherited hundreds of Soviet steel mills that were built at enormous scale during the Cold War to supply the Soviet military-industrial complex. The Soviets captured neon and other gases for experimental weapons and missiles and satellites. But after the Soviets collapsed, two Ukrainian companies mastered the purification process to semiconductor grade purity and spent decades supplying about 50% of the world's semiconductor grade neon. Why couldn't other countries just do the same? Because you need both the steel mill scale and the purification expertise to produce purified neon at scale. And most Western countries had either dismantled or never built the Soviet scale steel capacity. When the plans shut down in March 2024, neon prices spiked 600% and the global chip industry scrambled to find alternatives. China stepped in as the primary replacement and South Korea's POSCO announced expansion of industrial gas capacity to reduce dependence on Ukraine as of April 2026.
Dependency number five are finally getting to the Netherlands. The Dutch giant ASML, the gatekeeper behind every chip that Taiwan produces. Lithography is the process of printing circuit patterns onto the silicon wafer using light. The machine that generates and focuses this light is the extreme ultraviolet lithography system. There is only one company in the entire world that makes these machines. ASML advanced semiconductor materials lithography. Headquartered in Idenhovven, Netherlands. It holds 100% of the global EUV market. Why the Netherlands? Because ASML was spun out of Phillips, the Dutch electronics giant, it inherited Philips's research and development labs that specialize in precision engineering as well as a legacy of supplier relationship. The supplier relationships are massively important because ASML's machines contain over 100,000 components and ASML itself manufactures only 15% of them. The rest come from a supplier network of over 5,000 companies. The two most critical are the German Carl Zeiss SMT, the only company in the world that manufactures mirrors inside an EUV machine. Zeiss and ASML have been co-developing these optics for over 30 years. And these EUV mirrors do not exist anywhere else. And the American Syimer, the only company that makes the laser that generates EUV light inside the machine. Syimer was acquired by ASML in 2013, but it is a US origin technology and this is the key as to how the US can legally reach into ASML's business. Under the foreign direct product rule, the US can regulate exports of foreign-made products if they contain or are made using the critical US technology. Because ASML's EUV machines depend on Simer's US origin light source. Those entire scanners are treated as foreign direct products of the US which gives Washington the right to say you cannot sell your machines to China without our permission. And of course ASML does not have that permission. This is the critical reason why China cannot build its own chip fabrication plants. They do not have access to these machines. But why can't anyone replicate ASML? Because you need Zeiss optics that are under German expert controls. You need cyber lasers under US expert controls. And 30ome years of relationship, give or take, between optics, laser, and vacuum systems. But why the Netherlands? Why not France or Switzerland? They also border Germany. Because bordering Germany has nothing to do with it. The reason is Phillips. And Phillips being an Idenhovven specifically when lithography technology emerged in the 60s Philillips was already doing precision optics and photoiththography for its own consumer electronics neither France nor Germany or Switzerland had an equivalent dependency.
Six between each step in the lithography machine the chip needs other processes carving away unwanted material adding new layers on top adding heat to lock everything in place each of those actions requires its own specialized equipment and that equipment is where the second layer of the American and Japanese controls sit. Etching is the process of carving away the material that was exposed during lithography. The global market for etch equipment is dominated by lamb research an American company with 50% global market share. Applied materials also American holds most of the rest. Deposition is the reverse. It's adding ultra thin layers of new material onto the wafer surface. The American Applied Materials owns this niche as well. Thermal processing, heating the wafer to repair crystal damage and lock structures in place, is owned by the Japanese Tokyo Electron. Tokyo Electron also holds a near 90% global monopoly of these machines. And finally, inspection or quality control that runs through every layer. After every major transformation step within the wafer, the wafer gets scanned for defects. Kla, another American company, holds 80% of the global process control and inspection market. Without KLA, a fabrication plant, no matter where it sits, be it Taiwan or any other country on Earth, cannot produce a single reliable chip. Because without it, you cannot tell whether the process is working until the chip gets to the end of the line. By which point, hundreds of wafers can already be ruined. As you can see, this dependency is controlled primarily by the United States and the United States government in April 2026 forbid its companies from shipping equipment to China. Yet another reason why China cannot build its own chip.
Dependency 7, the photo mask and pelle. The photo mask is the template from which circuit patterns are printed. The pelle is a thin membrane stretched over the mask to prevent dust particles from landing on it. Who controls this step? For pelleles, the suppliers are primarily Canon and Shinetsu both Japanese. Japan holds approximately 30% of global supply. But for photo mask inspection equipment, the tool that inspects the photo mask for defects, Japan and specifically laser tech holds 100% global monopoly. This is important because you can in theory manufacture an EUV lithography machine, you can coat your wafers with photoresist. But if your photo mask has a defect and you cannot detect it because only laser tech makes the inspection tool. Every chip that you print from that mask can be defective. Laser tech monopoly is small in revenue terms but absolute in functional terms.
And finally, dependency 8, where we get to the beasts of semiconductor manufacturing. Everything I've covered, the wafers, the ink, the gases, the machines, the photo masks, all of it flows into a fabrication plant. And there are only a few massive players in the foundry market. The Taiwanese TSMC, the Taiwan semiconductor manufacturing company that just happens to hold 70% of the advanced chip market, and Korean Samsung with 7.2%. When Morris Chang founded TSMC in 1987, he made a bet that most semiconductor companies did not want to own factories. They wanted to focus on the design. And he was right. Apple designs chips. Nvidia designs chips, but they don't make them. And for nearly 40 years, every major chip company in the world sent their designs to TSMC to be physically manufactured in Taiwan. TSMC has been running production of cuttingedge chips longer than any other company on Earth. Within a 60-mi radius of TSMC's headquarters, there are hundreds of specialized semiconductor suppliers, chemical providers, precision equipment maintenance companies, specialized material suppliers, photomass makers that have been co-evolving with TSMC for 40 years straight and they can respond to production issues within hours. Samsung is the only company in the world that can even attempt leading edge production alongside TSMC. It has the equipment. It has the engineers, the EUV machines, the equipment, but it's struggling to make chips that come out functional. The current yield ratio is about 40%. At current rates, manufacturing costs for working chip are so high that major customers like Nvidia or Apple or Meta. Stay with TSMC. Samsung did grow 36% in late last year. But it happened because it was manufacturing its own chips for internal products like Galaxy processors and HBM memory in their own foundry. But it wasn't because they were able to sell to external customers. There's also the third player, Semiconductor Manufacturing International Corporation, who's China's national champion, and SMIC is genuinely capable, but they're heavily blocked by export controls. They are indeed producing Huawei chips, but they can't get top-of-the-line ASML machines from the Netherlands. SMIC is a serious foundaries for China's domestic market, but at the moment, it's irrelevant to the global AI chip race.
Now, back to the question, why no country can build a chip alone. This business is impossible to replicate quickly because it is heavily interconnected. Building the physical facility costs 15 to20 billion. That alone puts it out of reach for most nations. But the capital cost is not even the hardest part. One World Trade Center, the tallest building in the Western Hemisphere, rebuilt from the ashes of 9/11 on the most expensive piece of real estate in the world, cost $5.1 billion. You can build three World Trade centers and still have money left before reaching the cost of a single chip fabrication plant. But the real moat of TSMC is that it's been accumulating data from millions of chips. Every defect, every adjusted parameter, every tuneup is irreplaceable institutional knowledge. Samsung has spent tens of billions trying to match TSMC's capacity and is still significantly behind. Intel, with the world's largest semiconductor R&D budget, entered high volume manufacturing in late 2025. But even now, despite the budget, their yields are materially below TSMC's.
Now, if you zoom out and look at this map, you see the countries that repeat from dependency to dependency. So, a logical question would be, why isn't Japan building its own TSMC? Why don't the Netherlands manufacture chips directly and own 100% of the market? The curious thing is that Japan actually did have its own chip manufacturing and they were dominant in the 80s. But the economics of cutting edge fabrication became so capital inensive that the Japanese companies couldn't keep up. They were not able to fund both the chip design and manufacturing simultaneously. On top of it, Japan's chip companies were designed and manufactured for their own products and their internal market. And this model works when you're making chips for your own TVs, but it doesn't work when you need to manufacture chips for Apple or Nvidia or Meta, all with different architectures and requirements. And lastly, Japan made a deliberate choice to dominate the inputs rather than the fabrication itself. For them, it was more defensible and more profitable to own the materials that every fabrication plant on Earth depends on.
As for the Netherlands, ASML sells machines from two to $400 million each, and it has effectively zero competition and 100% margin leverage on the most critical tool in the supply chain. If ASML tries to become a chip manufacturer, it would need to build fabrication plants, compete with its own customers like TSMC and Samsung and Intel, hire tens of thousands of engineers, and develop yield expertise that took TSMC 40 years to accumulate. And in exchange, it would give up its position as the neutral supplier that everybody depends on. Right now, TSMC and Samsung both need ASML equally. The moment ASML starts manufacturing chips, it becomes a competitor to its own customers and those customers have every incentive to build their own supplier. ASML's monopoly is more valuable exactly because it stays in its lane.
So going back to the question, why is Taiwan the undeniable mecca of semiconductors? It's because every dependency above it converged there. The American design tools, the Japanese materials, Dutch machines, German optics, Eastern European gases, all ending at TSMC.
And finally, why no country can build a chip alone? Because chip manufacturing is a lock with eight keys spread across a handful of countries and assembled over half a century. No single nation controls all of it. And no nation can break away from it without triggering a geopolitical crisis and a collapse in a system that was built over decades.
We hope this was helpful and we'll see you in the next episode of the business of AI. Till next time.