We're Close to Disaster and Nobody Knows: Why Taiwan Chip Risk Forces Terafab

China has army, navy, and air units staged around Taiwan. About 90% of the world's most advanced chips — the ones in AI systems, phones, cars, and data centers — come from that one island. Prediction markets put a China-Taiwan military clash by end of 2027 somewhere around 11 to 18%. Economic models put year-one global GDP damage near $10 trillion. That is worse than COVID and the 2008 crash stacked together. The market is still acting like this is a niche foreign-policy story. It is not.

The single point of failure

Taiwan Semiconductor Manufacturing Company makes chips for Apple, Nvidia, AMD, Qualcomm, Tesla — basically everyone who matters in tech. TSMC produces roughly 90% of sub-7-nanometer chips and holds something like 64 to 71% of global foundry revenue. More than half of all chip manufacturing money on Earth runs through one company on one island sitting in the hottest geopolitical neighborhood on the map.

Chips are not "like oil." They are more important than oil. Oil moves cars and heat. Chips run phones, cars, email servers, AI models, and the defense systems countries depend on. If that supply snaps, the global economy does not slow down politely. It freefalls.

The Davidson window is next year

Xi Jinping — effectively president for life — has ordered the People's Liberation Army to be invasion-ready for Taiwan by 2027. That target is what people call the Davidson window, after Admiral Philip Davidson, former U.S. Indo-Pacific Command chief, who told Congress the PLA would have the capability by then. At the time of this recording, that is next year.

On Polymarket, real money sits behind those 11–18% clash odds. Call it roughly 15%. A 15% chance your house burns down next year is not "unlikely enough to ignore." You buy insurance. A 15% shot at a $10 trillion hit is the same math at civilization scale.

The drills are already practice, not theater. In December 2025, the PLA ran Justice Mission 2025 and simulated a full blockade of Taiwan. A blockade can choke chip output without looking like a classic invasion that forces the same automatic international response. PLA units now routinely cross the Taiwan Strait median line. Daily air-defense-zone incursions through 2026 are part of the pattern. New nuclear subs are being built to keep the U.S. Pacific Fleet at bay. Gray-zone pressure: not quite war, constantly testing how far they can push.

Why Beijing wants the island

Politics is the public story — Taiwan as a "breakaway province." The strategic story is chips. China still cannot make the most advanced ones. Hundreds of billions later, SMIC is stuck around 7nm with multi-patterning workarounds because it cannot buy ASML's extreme ultraviolet lithography tools. The U.S. and allies blocked those sales. SMIC's localization push lifted profits, but it remains roughly a decade behind TSMC on the leading edge. In chips, a decade is forever. You do not train frontier AI without the best silicon.

China's March 2026 15th five-year plan quietly dropped the word "chip." The prior plan had hard domestic chip targets. The new one swaps that for digital-economy value-add at 12.5% of GDP by 2030 — measuring AI penetration instead of chip output. That reads like a quiet admission: we cannot win the fab race soon, so we will try to win the AI-usage race. They still need the chips. The best ones still come from Taiwan, about 100 miles offshore. If you are Xi and chips are the defining resource of the century, the calculus is ugly and obvious.

Iran stretches the map

The Middle East fight is not a separate show. The U.S. and Israel campaign against Iran, Hormuz pressure from the IRGC, oil above $100 (up ~40% at recording) with Iran talking $200 — all of it pulls U.S. attention and munitions west. Hudson Institute analysis has framed Iran as a structural Chinese military asset: every dollar spent on Red Sea lanes and Middle East contingencies is a dollar not spent on Pacific basing, subs, or Taiwan planning. China wants America thin across two theaters.

The counter-read from other defense shops: Iran strikes signal that America will use overwhelming force when interests are hit, which could deter Beijing. Taiwan's own leaders reportedly want the Iran mission finished fast so Pacific resources come back. The $10 trillion question is which story is true — resources freed, or magazines emptied.

What a Taiwan hit actually does

Insurance Journal and independent modeling put first-year global GDP loss near $10 trillion. U.S. GDP is about $30 trillion. That is on the order of wiping out a third of America's annual output from the world economy in twelve months. Smartphones stall. AI data-center expansion freezes. Every modern car has a chip problem. Advanced shortages last three to five years minimum because fabs take years and billions; you cannot spin one up on a crisis timeline. TSMC's 3nm wafers cost about $19,000 each. Defense systems ride the same supply. Tech stocks that assumed uninterrupted TSMC — Apple, Nvidia, AMD, Qualcomm, the whole AI boom — reprice hard.

Taiwan is not cooperating with a full offshoring fantasy. Vice Premier Cheng Li-chiung rejected a U.S. ask to move 40% of capacity stateside, saying the semiconductor ecosystem "cannot be relocated." It is fabs plus engineers, materials, tools, and decades of know-how. TSMC has pledged over $100 billion for Arizona — and Arizona yields have even beaten Taiwan by about four points — but advanced packaging still stays home, Fab 2's 3nm timeline slips into 2027–2028, and Europe's €43 billion Chips Act is a slow climb from 10% to 20% share by 2030. Diversification is real. Diversification is not done before the Davidson window closes.

Terafab: Musk's $20 billion insurance policy

In 2026 Elon launched Terafab — Tesla's own chip manufacturing push targeting 2nm, past TSMC's current 3nm leading edge. Initial aim: 100,000 wafer starts per month, ambition toward 1 million — on the order of 70% of TSMC's current output from one effort. Target volume: 100 to 200 billion custom AI and memory chips a year. First product in line is likely Tesla's AI5, feeding Full Self-Driving, Cybercab, Optimus, Dojo, xAI's Grok, and eventually SpaceX-linked compute. Same playbook as Gigafactory: when the world said batteries would take forever, he built scale that bent the cost curve. Terafab is that move applied to silicon.

xAI's Memphis Colossus is racing toward multi-gigawatt power — city-scale electricity — with hundreds of thousands of Nvidia GPUs and a Mississippi plant plus dozens of gas turbines to chase a million-GPU footprint. Elon has said the goal is more AI compute than everyone else combined. All of that still sits on Nvidia, and Nvidia sits on TSMC. One island. Double-digit conflict odds. So he builds the fab himself.

These are not generic Nvidia clones. They are custom silicon for Tesla and xAI workloads — driving inference, robot control, specific training architectures — the same logic as Google TPUs or Amazon Trainium, but across a wider stack: cars collecting data, models training on that data, chips made in-house, robots that could eventually help make the chips, energy from Tesla solar and Megapacks. Closed loop. Vertical integration taken to the end of the sentence.

Sovereign AI and the asymmetry

The U.S. Chips Act helped trigger hundreds of billions in private projects. TSMC's U.S. investment plans have swollen toward $165 billion. America is on track to triple chip manufacturing capacity by 2032. Saudi Arabia's Humain push — terawatt-scale data-center ambition with Nvidia and xAI partners — shows even oil states are treating AI infrastructure as national strategy. Designing your own chip while still manufacturing in Taiwan is not the same as owning the fab. Google, Meta, and Amazon still lean on TSMC for production. Terafab on U.S. soil is a different insurance layer.

Maybe Xi never moves. Maybe deterrence holds and the window passes. Then Tesla still owns custom chips tuned for its workloads. The downside of the "wrong" call is a highly efficient in-house fab. The downside of the right call without preparation is Bloomberg-scale trillion-dollar damage and a cratered portfolio. Even a 10% disruption chance over five years makes a ~$20 billion fab look like rational insurance — the point Musk floated around Tesla's Q4 2025 earnings discussion. You do not buy fire insurance because you are sure the house will burn. You buy it because uninsured catastrophe is unacceptable.

If you own Nvidia, Apple, AMD, Qualcomm, or the broader AI stack, you are long Taiwan whether your broker says so or not. Track exposure. Watch who is actually building manufacturing independence, not just slide decks. Terafab, Colossus, Optimus, Megapack — connect those dots and you see a bet designed for exactly this shock. The clock on 2027 is not abstract.

Check the video here.