Microchips: The Bottleneck of the Digital Age

Last week we published an article about demand-induced stress in the supply chain for RAM and the computer components that rely on that kind of memory (and there are a lot of them). But the core of the conflict lies elsewhere: chips. This tension isn’t new. It has been building in waves—peaks and pullbacks—for years. You can invoke—rightly—the always swampy concept of geopolitics, but the hard fact is simpler to illustrate: maintaining the cutting edge in microprocessor manufacturing is a strategic advantage the West is not willing to lose to China.

What Is a Microprocessor?

A microchip (also called a chip, computer chip, integrated circuit, or IC) is a set of electronic circuits on a small, flat piece of silicon. On the chip, transistors act as miniature electrical switches that can turn current on or off. The pattern of these tiny switches is created on a silicon wafer by adding and removing materials to form a multilayer network of interconnected shapes.

Silicon is the material of choice in the chip industry. Unlike metals commonly used to conduct electric currents, silicon is a semiconductor, meaning its conductive properties can be increased by doping it with other materials such as phosphorus or boron. This makes it possible to activate or deactivate the electrical current.

A fingernail-sized microchip contains billions of transistors, so it’s easy to see how small a chip’s features must be—they’re measured in nanometers. A nanometer is one billionth of a meter, or one millionth of a millimeter. For comparison, a human red blood cell is about 7,000 nanometers in diameter, and the flu virus is around 100 nanometers. The most advanced microchips contain features as small as a few dozen nanometers.

The Lord of Nanometers

In chips, nanometers (nm) not only describe the physical size of the transistor, but a complete technological generation: density, energy consumption, performance and costs. That said, today the market can be organized into three large levels.

Avant-garde chips (leading edge) We are talking about 3 nm and 2 nm (TSMC, Samsung, Intel entering). They are the nodes used for high-performance AI, premium CPUs and GPUs. They offer maximum transistor density, lower consumption per operation and better performance per watt. They are very expensive, difficult to manufacture and require EUV lithography no matter what. Here are the chips that define the technological and military frontier.

Advanced chips They include 5 nm and 7 nm (and in some cases 4 nm, an optimization of 5). They remain extremely competitive and dominate a large part of the market for smartphones, data centers, and "non-top-of-the-range" AI accelerators. Consumption and performance are very good, but with somewhat more controllable costs. China today can produce in this range with advanced DUV and engineering tricks, although with lower efficiency than TSMC.

Standard or mature chips (mature nodes) They go from 14 nm, 28 nm and up (40, 65, 90 nm...). They are not "bad": they are the basis of cars, appliances, routers, sensors, industry, IoT. They are cheap, robust and easy to mass produce. They are not used to train large AI models, but they move the real economy. China largely dominates this segment.

Who Produces Microchips?

The advanced semiconductor industry (excluding China) is organized around three well-defined roles: foundries, IDMs, and fabless companies.

Foundries are pure-play manufacturers: they don’t design their own chips; they produce for third parties. The paradigmatic case is TSMC (Taiwan Semiconductor Manufacturing Company Limited), which manufactures the world’s most advanced chips for Apple, NVIDIA, AMD, Qualcomm, and much of today’s “digital” economy. Its strength is not only technological but industrial: scale, yields, and predictability. That’s why it sits at the heart of the system. Taiwanese people often refer to TSMC as the “divine protective mountain,” not only because of the income it has generated and because it placed Taiwan at the forefront of semiconductor manufacturing, but also because it functions as a deterrent against any attempt at invasion or war with mainland China. The destruction of TSMC would leave the planet without new microprocessors and would slow the development of computing for years. The damage would be incalculable. Morris Chang, its founder, established it in 1987 and it gained legendary status on the island.

On the next rung is Samsung, which manufactures chips both for third parties and for itself. Technically it competes with TSMC, but it carries a structural disadvantage: it is supplier and competitor at the same time. That inevitably creates friction with clients who may themselves be rivals. Based in South Korea, it is one of the most relevant technology companies on the planet, with major product lines in phones, home appliances, and RAM.

Finally, Intel—an iconic U.S. company—is a classic IDM (Integrated Device Manufacturer): it designs and manufactures its own processors. In recent years it has opened parts of its manufacturing capacity to third parties through Intel Foundry Services, although most of its capacity remains internal and its role as an external foundry is still being built. It has manufacturing operations in the United States (Oregon, Arizona, New Mexico), Ireland, Israel, and China, as well as assembly and testing facilities in Asia (including Malaysia and Vietnam). It is also building new sites in Germany and Ohio, expanding its global footprint.

At the other end of the spectrum are fabless companies (no factories, in plain terms) such as NVIDIA and AMD. These firms manufacture nothing themselves: they design chips and rely entirely on foundries to turn those designs into silicon.

Controlling the Flow of Data

Functionally, CPUs (microprocessors) are strong at multitasking and overall system control, while GPUs excel at highly parallel, repetitive workloads. That’s why AI training, graphics rendering, and gaming keep pushing GPU demand, while general-purpose computing still orbits around CPUs. Both worlds, however, crash into the same bottleneck: manufacturing capacity.

Key fabs are concentrated in Taiwan and South Korea, which already introduces an obvious geographic risk. But there is an even more critical point: only one company in the world can build the machines required to produce advanced chips. EUV lithography tools are made by ASML, a Dutch company that is—without exaggeration—the jewel of the Western technology crown. ASML is protected politically and strategically with almost the same zeal with which Taiwan protects TSMC. Without ASML, there are no modern chips. It’s that simple.

Semiconductor manufacturing is so complex that there is no quick way to ramp production. Any disruption creates bottlenecks. During the pandemic, the slowdown in global trade caused widespread shortages. Something similar happened during the Ethereum mining boom: explosive GPU demand coincided with the launch of NVIDIA’s RTX 30 series and produced a global shortage of graphics cards. On top of that are structural problems in sourcing critical raw materials—such as coltan and other rare earths—whose extraction has been directly linked to civil wars, chronic violence, and genocides in parts of Africa over the last two decades.

This landscape becomes even more tense when you factor in the dispute between China and the United States over control of 5G. The conflict between Huawei and its Western counterparts isn’t commercial; it’s strategic. From a U.S. national-security perspective, allowing a Chinese company to manage critical telecommunications infrastructure would amount to information-security suicide. Controlling networks means controlling the flow of data—and that is power in its purest form.

In that context, China set out to reach Western standards of chip manufacturing so it wouldn’t be locked out of a technology that is already omnipresent. Cars incorporate more and more computers. Planes simply don’t exist without them. Phones stopped being phones and became portable computers. Appliances are moving in the same direction, even if it isn’t always a good idea—though that’s another debate. The central point is this: falling behind in semiconductors today means falling behind in everything. And that’s why the fight doesn’t let up.

International Supply Chains

To this picture we have to add a deeper layer: the gradual decoupling between design, manufacturing, and technological sovereignty. For decades, digital capitalism carried an implicit assumption: design was the value, manufacturing was a commodity. That assumption is dead. Advanced manufacturing has once again become a scarce strategic resource, comparable to oil in the 20th century. It’s not enough to know how to design chips: if you don’t have guaranteed access to fabs and cutting-edge lithography, your advantage evaporates. The efficiency created by globalized free-trade flows is giving way to a push for production autonomy.

The United States realized this late, but reacted forcefully and, in August 2022, enacted the CHIPS Act to recover lost capabilities. Subsidies, tax incentives, and direct financing aim to rebuild manufacturing capacity on U.S. soil—not to win on cost, but to reduce geopolitical dependence. Intel is the main beneficiary, but TSMC and Samsung are also building fabs in Arizona and Texas. The goal is not economic efficiency; it’s autonomy.

Europe, for its part, is playing a different game. It doesn’t lead in advanced logic manufacturing, but it controls an even more delicate bottleneck: equipment. ASML is not only the only company capable of producing EUV machines; it is also the point where Western political alignment becomes material. Export restrictions to China are not decided by ASML as a private company: they are state decisions coordinated among the Netherlands, the United States, and the European Union. The chip value chain is, quite literally, legislated.

Japan is a quieter but essential player. Firms such as Tokyo Electron, Nikon, Canon, and multiple chemical and materials suppliers remain indispensable. Without Japan there are no modern semiconductors, even if it doesn’t manufacture 3 nm leading-edge chips itself. This extreme fragmentation helps explain why the industry is so hard to replicate: it’s not one factory, it’s an ecosystem distributed across allies. It’s striking how closely the microprocessor industry is linked to photography—because the whole process ultimately relies on printing patterns of light onto silicon. That is precisely what ASML’s lithography machines do.

Meanwhile, the rise of generative AI introduced an additional distortion. Never before had a single type of computational workload concentrated so much demand for such specific hardware. Training large models pushed GPU and high-bandwidth memory (HBM) demand to unprecedented levels. This shifted capacity away from other sectors and reshaped priorities: today, a data-center GPU can be worth more than dozens of industrial chips. The market allocates silicon by computational profitability.

That helps explain why scarcity isn’t felt uniformly. It’s not that chips in general are missing; it’s that advanced chips are missing. Cars, appliances, and industrial devices often use older nodes, but they still compete indirectly for inputs, talent, and fab capacity. The shortage is systemic, not localized. Each new AI expansion pushes pressure up the entire chain.

China watches all of this from an uncomfortable position. It has market size, capital, and talent, but it does not control the critical nodes. U.S. sanctions aren’t designed to “stop” China in the abstract—they are designed to buy time: time for the West to consolidate its advantage, and time so China doesn’t reach the technological frontier before the global balance is reshaped.

In this scenario, semiconductors stop being a technical input and become civilizational infrastructure. Every autonomous car, every smart grid, every modern weapons system, and every AI platform depends on them. There is no quick substitute and no shortcut. Scarcity isn’t a bug in the system: it is a direct consequence of pushing digitization all the way down into the material layer.

Put differently: digital capitalism discovered its own physical limits. And chips are now the exact point where those limits become visible.

Chinese Democracy

Although we’ve already mentioned China, we saved the best for last. In recent weeks, there have been major developments regarding China’s microprocessor manufacturing ambitions. China doesn’t just want to participate; it wants to match the West’s strategic advantage—and that requires emulating ASML’s technical feat. Reuters reported recently on a possible government plan to organize a highly secret state effort that it explicitly described as a chip “Manhattan Project.”

The goal is clear: break the West’s technological leverage over advanced semiconductors for AI. The core of the plan is the development of a domestic EUV lithography machine—the key technology for manufacturing the world’s most advanced chips—which is effectively monopolized by ASML today under export controls promoted by the United States and its allies.

The central claim is that China has already built a functional EUV prototype capable of generating extreme ultraviolet light, and that it has been in testing since 2025. It doesn’t produce commercial chips yet, but the milestone would be enormous: EUV is among the most complex industrial challenges in existence. The effort reportedly relies on reverse engineering, reuse of older Western equipment, and recruiting engineers with previous experience in ASML’s ecosystem.

Coordination would be led by the Chinese state, with Huawei as an industrial hub, integrating laboratories, universities, and local suppliers. The official target is usable production by 2028, although sources cited by Reuters acknowledge that a more realistic scenario could be closer to 2030, given the immense challenges of precision, stability, and industrial yield.

Conclusions

The world’s industrial capacity to build computers is enormous, but it is not infinite. What we’re seeing—driven both by new demand (AI) and by new political rivalries (China vs. the United States)—points toward a broad expansion of manufacturing capacity. That won’t be an easy path. Conflicts may intensify and the struggle to maintain the frontier may escalate, but the end result could be a major multiplication of computing power.

Consider what happens when China’s industrial and organizational capacity is pushed into microchip manufacturing. Something similar could happen with RAM and other memory chips. These scenarios aren’t fully mapped yet, but we will inevitably see them materialize (or fail) in the coming years. It’s plausible that new generations of gamers won’t just weigh Intel versus AMD, but will start testing how a Zhaoxin—or a Huawei-designed chip—actually performs.