90% of the World's Advanced Chips Come From One Island. Everyone Knows. Nobody Cares.
Here’s a fun thought experiment: imagine 90% of the world’s electricity came from a single power plant on an island that two nuclear-armed countries are actively fighting over. You’d call that insane. You’d fix it immediately.
Welcome to semiconductors in 2026.
Taiwan makes 90% of the world’s high-end chips. Not a plurality. Not a majority. Ninety percent. Treasury Secretary Scott Bessent put it at 97% at Davos last month (the actual number is slightly lower, but honestly, at this scale of concentrated risk, arguing over rounding errors is missing the point). Every iPhone. Most of the GPUs running your AI models. A third of the chips in cars, pacemakers, and cell towers. They all flow through TSMC’s fabs on one island that China considers a “breakaway province” and has been running live-fire military drills around.
The New York Times dropped a detailed investigation on this yesterday, and the real news isn’t the situation — it’s how thoroughly and completely Silicon Valley has been warned, for years, by everyone, and responded by doing approximately nothing.
The Warning They Received (And Ignored)
In 2021, Admiral Philip Davidson told the Senate Armed Services Committee — on the record, under oath — that China wanted to be ready to take Taiwan by 2027. That’s not a think tank speculation. That’s the commander of US Indo-Pacific Command saying “the threat is manifest during this decade.”
Biden’s national security team ranked the Taiwan chip dependency as one of America’s greatest national vulnerabilities. They held secret briefings with Apple, AMD, and Qualcomm executives. They explained that a Chinese blockade could “bring the U.S. tech industry to its knees.”
The Semiconductor Industry Association quietly commissioned a confidential 2022 report that spelled out exactly what happens if Taiwan goes dark: U.S. economic output drops 11% — twice the severity of the 2008 recession. China, ironically, cratered even harder at 16% (mutually assured economic destruction being its own kind of deterrence).
The companies responded to all of this by… maintaining their supply chains exactly as-is and optimizing for quarterly earnings. Legendary stuff.
The Catch-22 Nobody Wants to Solve
To be fair to the industry — briefly and against my better judgment — the economics are genuinely brutal. Chips made in US fabs cost 25% more than chips made in Taiwan. Higher material costs, higher labor costs, more expensive permitting processes. Nobody wants to pay that premium voluntarily, and no company wants to be the one that blinks and absorbs the margin hit while their competitors keep buying cheap Taiwanese silicon.
This is the collective action problem in its purest form. As McKinsey’s semiconductor practice head put it: executives think “if we’re screwed, everyone else is screwed” — so they do nothing. Classic tragedy of the commons, played out with the world’s most critical technology supply chain.
The US has committed to spending $200 billion on semiconductor fabs through 2030 — the CHIPS Act money, plus tariff-coerced corporate commitments like Nvidia promising to buy from TSMC’s new Arizona plants. That’s a lot of money.
The result? By 2030, the US will account for 10% of global semiconductor production. Which is, coincidentally, roughly what it was in 2020 when all the alarm bells started going off. Decade of effort, billions of dollars, barely moved the needle.
The Nvidia Plot Twist
Here’s where it gets interesting: Jensen Huang was literally in the Oval Office recently when Trump told him he was planning to put tariffs on semiconductors because making them in Taiwan was “risky.” Trump, of all people, connecting the geopolitical dots.
Nvidia — the most valuable company on Earth, the company whose GPU shipments are practically treated as economic indicators — then committed to buying chips from TSMC’s Arizona fabs. That’s actually meaningful. When the company running the AI gold rush says it’ll pay more for domestic silicon, other companies pay attention.
But TSMC’s Arizona fabs are still… TSMC fabs. It’s a Taiwanese company’s workers and processes, on American soil. Better than nothing. Not the same as building independent American semiconductor capacity.
What a Blockade Actually Looks Like
The confidential SIA report gave companies something concrete: if Taiwan goes offline, most of the biggest US tech companies have several months of chip inventory before operations start breaking down. A few months is enough time to realize the scope of the disaster. It is not enough time to build new fabs, train new workers, or replicate five decades of Taiwanese manufacturing expertise.
China’s military planners presumably know this. The pressure campaign doesn’t even have to be a full invasion — a blockade, extended military exercises that disrupt shipping, or even credible threat of disruption could crater semiconductor prices and confidence. The “silicon shield” doctrine says Taiwan’s chip manufacturing is its protection because the world can’t afford to let it fall. But that only works if the calculation holds.
The calculation might not hold forever.
The Bottom Line
I’m not predicting a Chinese invasion of Taiwan. Plenty of serious people think the risks are overstated, that Xi is rational, that economic interdependence creates deterrence. They might be right.
But here’s what’s undeniably true: we built the most critical technology supply chain in human history on the geographic equivalent of a single data center with no redundancy. We got briefed on the risk repeatedly, by military officials and economists, in classified rooms and public hearings. We looked at the catastrophic scenarios, looked at the 25% cost premium to fix it, and chose the profit margin.
The entropy is real. The single point of failure is real. Whether or not it fails, designing a system this way is a choice we made consciously, with full information, and we should at least be honest that “efficiency” is just another word for “we’ll deal with it when it breaks.”
At urandom.io, we like to say entropy is not chaos — it is potential. But even entropy respects the difference between having one path and having many.
Sources: NYT investigation by Tripp Mickle (Feb 24, 2026); Stimson Center; SEMI industry data; Semiconductor Industry Association confidential 2022 report (as reviewed by NYT)