90% of the World’s Most Powerful Chips Depend on One Vulnerable Island

Why should you read this story?

Open your phone. Order a cab on Uber. Ask a question on ChatGPT. Pay with Google Pay. Stream a show on Netflix. Every single one of these actions runs on a tiny piece of technology smaller than your fingernail: a semiconductor chip.

A semiconductor chip, often just called a “chip,” is a thin sliver of silicon with billions of microscopic circuits carved into it. Think of it as the brain inside every electronic device you own. Your phone has several. Your car has dozens. An AI data centre that powers ChatGPT has thousands. Without chips, there are no smartphones, no internet, no digital payments, no modern cars, no defence systems, no AI. The entire digital economy sits on top of them.

Now here is the part that should concern every investor. Despite chips being the foundation of a multi-trillion-dollar global economy, the entire supply chain that designs, builds, and equips chip factories depends on a shockingly small number of companies, most of them concentrated in a single geography that sits at the centre of a geopolitical flashpoint.

India just took its first real step into this story. Understanding what that step means, and what it does not mean, requires understanding the four layers that hold up the global chip supply chain.

Here is a simplified map of who controls what:

Four layers. Essentially, four bottlenecks. Now, let us walk through each one.

Layer one: Who designs the chips?

When we talk about “AI chips,” we mostly mean a specific kind of processor called a GPU (Graphics Processing Unit), originally designed for rendering video game graphics but now repurposed for the kind of heavy parallel computation that AI models need.

NVIDIA, a US company, dominates this space. Depending on the estimate, Nvidia controls between 80 and 92% of the data centre AI accelerator market by revenue (Silicon Analysts, May 2026 market share report). Its Q3 calendar 2025 data centre revenue alone was around $51 billion (Nvidia FY2026 10-Q filing), which is more than the entire annual revenue of AMD, its nearest competitor.

But here is the crucial detail: Nvidia does not make a single chip. It only designs them. The actual physical manufacturing is outsourced to a foundry partner. Nvidia keeps the high-margin design and software business. The foundry keeps the capital-intensive, risky manufacturing side.

This “fabless” model (meaning the company has no fabrication plant, or “fab”) is common in the chip industry. Apple, Qualcomm, AMD, and Google all design chips but do not manufacture them. They all send their designs to the same place.

NVIDIA’s dominance is real, but it is not permanent. AMD’s MI300/MI350 chips have captured roughly 5 to 7% of the market (Presenc AI, May 2026). More importantly, big tech companies are designing their own custom chips: Google has its TPU (Tensor Processing Unit), Amazon has Trainium, Microsoft has Maia, and Meta has MTIA. TrendForce (May 2026, via Tom’s Hardware) projects that these custom chips will reach 27.8% of AI server shipments in 2026, growing at 44.6% year-on-year versus only 16.1% for standard GPUs like Nvidia’s. In simple terms: Nvidia is king today, but the court is getting crowded.

Layer two: Who makes the chips?

Nearly every advanced chip designed by Nvidia, Apple, AMD, and Google ends up being manufactured by one company: Taiwan Semiconductor Manufacturing Company, or TSMC, based in Hsinchu, Taiwan.

To understand TSMC’s dominance, you need to understand how chips are made. Manufacturing a chip means taking a flat disc of pure silicon (called a “wafer”) and carving billions of microscopic circuits onto it using light, chemicals, and extreme precision. The smaller you can make those circuits, the more powerful and energy-efficient the chip becomes.

The size of those circuits is measured in “nanometres” (nm). For context, a human hair is about 80,000 nm wide. The most advanced chips today are made at 3nm and 2nm, meaning the individual transistors (the tiny on-off switches that do the computing) are just 2 to 3 nanometres across. Getting to these sizes requires physics, chemistry, and engineering operating at the absolute edge of what is physically possible.

TSMC makes over 90% of the world’s most advanced chips, the kind that power AI systems, flagship smartphones, and cutting-edge laptops (SemiWiki, TSMC 2025 Update; TSMC 2024 Annual Report). If you zoom out to include all types of chips, not just the most advanced ones, TSMC still commands 60 to 70% of the global foundry market. In Q4 2025, more than 3/4th  of TSMC’s revenue came from chips that only TSMC can reliably make at scale (TSMC Form 6-K, FY2025 SEC filing). No other company comes close.

Samsung Foundry in Korea and Intel Foundry in the US are the only real alternatives, but both are behind TSMC at the leading edge. Samsung’s latest yields (the percentage of chips that come out working) have improved but still trail TSMC’s. Intel’s newest process, called Intel 18A, is real but has attracted few outside customers so far, with Microsoft being the highest-profile win.

The concentration risk here is hard to overstate. If something disrupts TSMC’s operations in Taiwan, whether from a natural disaster, a pandemic, or a geopolitical crisis, the entire global technology industry stops. Not slows down. Stops. Apple cannot ship iPhones. Nvidia cannot ship AI chips. AMD, Google, Amazon, and Microsoft cannot get their processors made. There is no backup at the leading edge.

Layer three: Who makes the machines that make the chips?

This is the layer most people have never heard of, and it might be the most extraordinary monopoly in industrial history.

To carve those nanometre-scale circuits onto silicon, you need a special machine called a lithography system. It works, in simplified terms, like a projector: it shines a specific kind of light through a “mask” (a stencil of the circuit pattern) and projects that pattern onto the silicon wafer coated with a light-sensitive chemical.

For the most advanced chips (anything below 7nm), the required light is called Extreme Ultraviolet, or EUV. EUV light has a wavelength of just 13.5 nanometres, so short that it gets absorbed by almost everything, including air. An EUV machine has to operate in a near-perfect vacuum, bounce the light off ultra-precise mirrors (not lenses, because even glass absorbs EUV), and hit accuracy targets measured in fractions of a nanometre.

There is exactly one company on Earth that can build these machines: ASML, headquartered in Veldhoven, the Netherlands.

ASML’s CEO Christophe Fouquet, in a May 2026 interview with TechCrunch, put it simply: “No one is coming for us.” The company’s 2025 revenue was 32.66 billion euros, up 15% year-on-year, and net bookings doubled between Q3 and Q4 of 2025 (ASML Q4 2025 earnings release; Yahoo Finance/Reuters, January 2026).

Each standard EUV machine costs roughly $150 to $200 million. The newest generation, called High-NA EUV (where “NA” stands for Numerical Aperture, essentially the precision of the lens/mirror system), costs over $380 million per machine (Motley Fool, March 2026). High-NA EUV allows even finer circuit patterns, enabling the next generation of 2nm and smaller chips. Think of it as upgrading from a regular projector to a 16K projector.

ASML’s supply chain is itself a wonder of concentration. The precision mirrors come from Carl Zeiss in Germany. The high-power lasers come from Trumpf, also German. The customer base is exactly three companies: TSMC, Samsung, and Intel. Japan’s Nikon and Canon, which once dominated lithography, today only make older-generation machines for less advanced chips.

China unveiled its first EUV prototype in December 2024, but most analysts do not expect Chinese EUV in commercial production before 2028 to 2030 (Motley Fool, March 2026). The monopoly is not just real; it is widening.

Layer four: Who supplies the raw materials?

This is the part most commonly misunderstood in popular commentary.

You may have seen the claim that “China controls 80% of the world’s silicon.” That is partially true and mostly misleading, because “silicon” means different things at different stages of the supply chain.

Raw silicon metal (the kind refined from quartz rock, used in steel, aluminium alloys, and as a starting material for further refining): China produces about 80% of the world’s supply (USGS Mineral Commodity Summaries 2025). This is true.

Polysilicon (a more refined form, primarily used for solar panels): China produces about 93.5% of global output (Bernreuter Research, November 2025; corroborated by TaiyangNews and pv magazine). This is even higher than 80%, but it is mostly solar-grade, not the ultra-pure grade needed for chips.

Semiconductor-grade silicon wafers (the actual thin discs of ultra-pure silicon that chips are built on): Here, Japan dominates, not China. Shin-Etsu Chemical controls roughly 28% of global 300mm wafer supply, and SUMCO controls another 23 to 25% (Semiconductor Insight, 2025 Silicon Wafer Market report). Together with GlobalWafers (Taiwan), Siltronic (Germany), and SK Siltron (Korea), the top five producers account for 82 to 85% of revenue. China’s share is in the single digits to low teens.

There is an even quieter chokepoint: photoresist, the light-sensitive chemical that coats wafers during the lithography process. According to Nikkei (November 6, 2025, cited by TrendForce), Japanese companies including Tokyo Ohka Kogyo and JSR together account for about 91% of the global photoresist market.

To summarise: China holds the upstream (raw silicon metal, solar polysilicon). Japan holds the downstream (chip-grade wafers, photoresist). For chipmaking specifically, the critical material dependency is on Japan, not China.

Where is the concentration risk? A simple summary.

Here is the clearest way to see the fragility:

The most dangerous concentration is in manufacturing (TSMC) and lithography (ASML). Design has alternatives. Materials have partial alternatives. But there is no second TSMC at the leading edge, and there is no second ASML at all.

Why does Taiwan matter geopolitically?

To understand why governments are spending hundreds of billions of dollars building chip factories outside Taiwan, you need a brief piece of geopolitical context.

China considers Taiwan a part of its territory. Taiwan has been self-governing since 1949 and operates as an independent democracy, but China has never renounced the possibility of reunification by force. The United States has historically maintained a position of “strategic ambiguity,” not formally recognising Taiwanese independence but providing defensive support.

In 2021, the top US military commander responsible for the Asia-Pacific region told the US Congress that China could be ready to make a move on Taiwan within six years, pointing to 2027 as the key date (USNI News, March 9, 2021). This warning became known in policy circles as the “Davidson Window,” named after the commander who made it.

But 2027 is widely misunderstood. What US intelligence agencies have actually said is that China wants to be capable of taking Taiwan by 2027, not that it has decided to do so (CIA Director’s public remarks, February 2023). There is a big difference between building the ability to do something and choosing to do it. When researchers surveyed 64 leading China experts, only 2 thought 2027 was a real deadline (CSIS ChinaPower Project, 2024). Most pointed to much later dates, or no fixed timeline at all.

On the ground, the signals are confusing. Chinese military aircraft were flying near Taiwan at record levels in 2024. By early 2026, that activity had dropped to multi-year lows (AEI China & Taiwan Update, March 2026). Nobody can say with confidence what either trend means.

The bottom line for investors: nobody knows when or whether China would attempt to take Taiwan. But the mere possibility that it could happen, combined with the fact that over 90% of the world’s most advanced chips are made on that island, has been enough to trigger the largest industrial reshoring effort in modern history.

The Global Response: Building chip factories everywhere

The response from governments has been massive. The US, Japan, and Germany are all paying TSMC to build chip factories on their soil, so that the world is not entirely dependent on one island.

United States. What started as a single $12 billion factory announcement in Arizona in 2020 has ballooned into a $165 billion, three-factory campus (TSMC March 2025 press release; SEC filing). The first factory started producing chips in late 2024, and early results have been strong: the Arizona facility is actually producing a higher percentage of working chips than comparable plants back in Taiwan (Bloomberg, October 24, 2024). The US government has backed this push with over $33 billion in direct funding and catalysed over $110 billion in total investment through a law called the CHIPS Act (NIST/Commerce data; Manufacturing Dive tracker).

Japan. TSMC’s factory in Kumamoto entered mass production in late 2024 and turned its first profit in Q1 2026 (Taipei Times, May 18, 2026). A second, more advanced factory is now planned with a $17 billion investment (Technetbook, February 2026).

Germany. TSMC broke ground on a factory near Dresden in August 2024, targeting production in 2027. This one is focused on chips for European carmakers (TrendForce, November 2025).

The pattern is clear: the world’s most important chipmaker is being pulled in every direction by governments willing to spend tens of billions to have a factory on their own territory. That is the context in which India’s move should be understood.

India’s move: The what, how and where of it

On May 16, 2026, Tata Electronics signed a Memorandum of Understanding with ASML in The Hague, witnessed by Prime Minister Narendra Modi and Dutch PM Rob Jetten. Under the agreement, ASML will supply its lithography tools and solutions to India’s first front-end semiconductor fabrication plant at Dholera, Gujarat (Al Jazeera, May 18, 2026, reporting by Usaid Siddiqui; ASML and Tata Electronics joint statement).

“India’s rapidly expanding semiconductor sector represents many compelling opportunities, and we are committed to establishing long-term partnerships in the region,” ASML CEO Christophe Fouquet said.

The plant is an $11 billion investment by Tata Electronics, targeting production of 300mm wafers. Let us explain what that means.

What is a 300mm wafer? A wafer is a thin, perfectly circular disc of ultra-pure silicon, the starting canvas on which chips are built. The “300mm” refers to its diameter (about 12 inches across). This is the current global standard for modern chip manufacturing. A larger wafer means more chips can be carved from each disc in a single production run, which lowers cost per chip. The previous generation used 200mm wafers; the industry moved to 300mm in the early 2000s, and essentially all advanced and mid-range chips today are built on 300mm.

What process nodes will the Dholera fab target? Tata is partnering with Taiwan’s Powerchip Semiconductor Manufacturing Corporation (PSMC) for technology transfer across 28nm, 40nm, 55nm, 90nm, and 110nm process nodes, with the initial target at 12nm-class production. Union Minister Ashwini Vaishnaw has indicated the plant should be ready by 2028 (Moneycontrol, cited by Al Jazeera).

Where does 12nm sit in the global picture? To put this in perspective: TSMC is currently shipping 2nm chips. The Dholera fab will start at 12nm, which is roughly five generations behind the leading edge. That sounds like a huge gap, and it is, but it misses the point. The vast majority of chips used in the world are not bleeding-edge AI processors. Cars, washing machines, industrial equipment, medical devices, telecom infrastructure, IoT sensors, and basic consumer electronics all run on chips made at 12nm to 90nm and above. This is a massive, growing, and undersupplied market. India is not entering the race to out-compete TSMC. India is entering the race to serve the enormous middle layer of global chip demand.

Where does India sit in the overall ecosystem? Here is a simple way to visualise where India is starting and where it already has strength:

So India’s strength is at both ends: design and software. The middle (manufacturing, equipment, materials) is where the gap exists. The Dholera fab is India’s first serious attempt to bridge that gap.

How significant is the Tata-ASML deal? CSIS senior fellow Sujai Shivakumar told Al Jazeera: “India is seeking to build out its semiconductor industry by building 12nm chips. ASML can supply the equipment needed to produce them.” Harsh V Pant of the Observer Research Foundation in New Delhi called it one of “the most important semiconductor developments India has seen in recent years,” signaling a shift “from mainly software services and AI talent toward owning part of the physical infrastructure behind AI itself.”

This is the first time ASML has publicly committed to enabling a major fab in a fourth geography beyond the existing TSMC, Samsung, and Intel triad. That matters for ASML (customer diversification) and for India (credibility in the global supply chain conversation).

What are the risks? There is a history of setbacks. Foxconn pulled out of a $19.5 billion Vedanta semiconductor joint venture in 2023 over incentive delays (Al Jazeera). The current announcement is an MoU, not a delivered factory. ASML’s standard process from MoU to first tool installation typically takes 18 to 24 months. India also faces infrastructure challenges: reliable high-quality power, water supply, and a deep pool of skilled technicians at all levels (Shivakumar, per Al Jazeera). India’s recent macro pressures add context: over $20 billion of equity outflow linked to the US-Israel-Iran conflict has weakened the rupee and pulled forex reserves from about $720 billion to $697 billion (Al Jazeera, citing market data).

The Dholera fab is not a guarantee. But it is the most credible step India has taken into chip manufacturing to date, complementing the country’s first indigenous chip, the Vikram-32 (a 32-bit microprocessor for space launches, unveiled October 2025), and the $1.07 billion India AI Mission announced in 2024.

What should a retail investor take away from this?

The supply chain is fragile, but not failing. The semiconductor supply chain has been this concentrated for years. What has changed is that governments and companies are now actively diversifying it at unprecedented scale. This is a five-to-ten year reshoring cycle, and Indian investors are now part of that story.

India’s opportunity is real but early. The Dholera fab is targeting first production in 2028. That is two years away, and the MoU still needs to translate into construction milestones, equipment delivery, and successful ramp-up. The Foxconn-Vedanta collapse in 2023 is a reminder that announcements and execution are different things. If Dholera succeeds, it is genuinely significant. If it slips, the narrative resets.

Where India already wins: design and software. Nearly 20% of the world’s chip design engineers are Indian. India’s IT services sector (TCS, Infosys, HCL, Wipro) and its growing AI ecosystem are already embedded in the global semiconductor value chain. The Dholera fab adds a manufacturing layer to an existing design and software foundation.

The biggest risks are not where you think. Most retail investors worry about which chip company to pick (Nvidia vs AMD vs Intel). The deeper risk is that the entire AI and tech supply chain runs through a single island (Taiwan), depends on a single company for manufacturing (TSMC), and depends on a single company for equipment (ASML). Concentration risk at the supply chain level is a bigger exposure than any individual stock pick.

The companies and themes named in this note are for educational and analytical purposes only and do not constitute investment advice. Investors should consult their own financial advisors before making any investment decisions.

Key sources referenced in this article: TSMC Form 6-K / SEC filings (FY2025); ASML Q4 2025 earnings release; Berkshire Hathaway Q1 2026 10-Q and 2025 Annual Shareholder Letter; USGS Mineral Commodity Summaries 2025; Bernreuter Research (November 2025); TrendForce (May 2026); Silicon Analysts (May 2026); Presenc AI (May 2026); Council on Foreign Relations (December 2025); CSIS ChinaPower Project; Defense News (May 2024); AEI China & Taiwan Update (March 2026); Foreign Affairs (2026); Al Jazeera (May 18, 2026); Bloomberg (October 2024); Tom’s Hardware (May 2026); Motley Fool (December 2025, March 2026, April 2026, May 2026); TechCrunch (May 2026); Nikkei/TrendForce (November 2025); AMRO Research (2025); Semiconductor Insight (2025); CNBC (February 2025); Manufacturing Dive CHIPS Act tracker; Taipei Times (May 2026).

_{Disclaimer: Investment in securities market are subject to market risks. Read all the related documents carefully before investing. Registration granted by SEBI, membership of a SEBI recognized supervisory body (if any) and certification from NISM in no way guarantee performance of the intermediary or provide any assurance of returns to investors.}

_{The content in these posts/articles is for informational and educational purposes only and should not be construed as professional financial advice and nor to be construed as an offer to buy /sell or the solicitation of an offer to buy/sell any security or financial products.Users must make their own investment decisions based on their specific investment objective and financial position and using such independent advisors as they believe necessary.}

_{Windmill Capital Team: Windmill Capital Private Limited is a SEBI registered research analyst (Regn. No. INH200007645) based in Bengaluru at No 51 Le Parc Richmonde, Richmond Road, Shanthala Nagar, Bangalore, Karnataka – 560025 creating Thematic & Quantamental curated stock/ETF portfolios. Data analysis is the heart and soul behind our portfolio construction & with 50+ offerings, we have something for everyone. CIN of the company is U74999KA2020PTC132398. For more information and disclosures, visit our disclosures page here.}