Your Chip Supply Chain Is a Geopolitical Bet

Most companies think their biggest chip risk is price. It isn’t. It is a hundred-mile stretch of water off the coast of China.

Every board that signs off on a technology roadmap is making a geopolitical wager whether it realizes it or not. The question is not whether you are betting. It is whether you know which way.

The concentration problem

Roughly nine in ten of the world’s most advanced logic chips are produced by one company, on one island — Taiwan — situated about a hundred miles from a nation that has never renounced taking it by force.

That is the figure everyone quotes. The one they skip sits on the other side of the same strait. China has spent a decade building out mature-node capacity: the unglamorous 28- to 65-nanometer silicon inside power supplies, motor controllers, industrial sensors, automotive modules, and most of what your facilities actually run on. Its share of that segment rose from roughly 18 percent in 2020 to more than 31 percent by 2023, and the direction of travel has not changed.

So the exposure is not one-sided. Your most advanced components come from one side of the Taiwan Strait, and a large and growing share of your ordinary ones come from the other.

That is not a supply chain. It is a fault line.

Stop betting on the date

The temptation here is to argue about timing, and the timing argument is a trap.

For six years the planning conversation has anchored on 2027 — the “Davidson window,” named for the admiral who told Congress the People’s Liberation Army could be capable of taking Taiwan by then. Capable. That word did a great deal of work, and most people dropped it.

You do not need the worst case. You only need the strait to stop working normally.

In March 2026, the Office of the Director of National Intelligence published its Annual Threat Assessment and stated plainly that Chinese leaders do not currently plan to execute an invasion of Taiwan in 2027, nor do they have a fixed timeline for achieving unification.

Read that as relief if you like. I read it the other way.

A fixed date is a planning problem; you can schedule against it. No fixed timeline means the decision stays available, indefinitely, to be taken whenever conditions favor it. And the assessed emphasis has shifted toward coercive pressure short of invasion — which happens to be the scenario your supply chain is least prepared for.

Because the threshold that should worry you was never invasion. It is closure. A quarantine, a customs inspection regime, an insurance market that freezes, a shipping lane commercial carriers decline to enter. No shots fired at your suppliers, no fabs destroyed. You do not need the worst case. You only need the strait to stop working normally.

And this is what makes the date question beside the point: the chips you standardize on this year will still be in your products and facilities ten to twenty years from now. Your sourcing decision is not a bet on a quarter or an administration. It is a bet placed across the entire window in which no one — including the intelligence community — claims to know what happens.

The reinvestment cliff

If the strait stops working, you do not reorder from somewhere else. You re-qualify a supply base: new vendors, new components, new firmware, new certification cycles, new validation for anything safety-rated or regulated.

People consistently underestimate what that costs. So look at what it takes when the conditions are good.

The FCC’s “rip and replace” program exists to pull Huawei and ZTE equipment out of American telecom networks. It is federally funded — roughly $5 billion between the 2021 appropriation and the $3.08 billion authorized in December 2024. It has clear legal authority, defined scope, willing participants, and no crisis. Reimbursement checks began going out more than four years ago.

As of June 2026: 126 funded projects, 53 complete. Forty-two percent, after six years.

The stated reasons are worth reading, because none of them are exotic. Thirty-five percent of participants cite supply chain delays, double the share six months earlier. The bottleneck is not radios — it is power supplies, cable connectors, and cable lugs, because rural tower upgrades are competing for the same parts as fiber builds and AI data center construction. Fifteen percent cite labor shortages. Fifteen percent cite weather.

Rip and replace — federally funded, six years in

42% complete
53 of 126 funded projects

Source: FCC program status, June 2026

That is the easy version. One vendor category, one industry, government money, no emergency.

Now price the version where it is your entire silicon base, on no notice, while every competitor is re-qualifying the same alternates from the same short list of suppliers at the same moment, at whatever price clears.

The time to build optionality is before you need it. Not during.

Availability is only half of it

Where a component was made, and by whom, is a security question, not merely a procurement one.

This is already written into U.S. law and regulation. Section 889 of the FY2019 National Defense Authorization Act bars federal agencies from procuring certain Chinese-made telecommunications and video surveillance equipment, and from contracting with entities that use it. The FCC maintains a Covered List identifying equipment and services — Huawei and ZTE among them — as unacceptable risks to national security. The rip-and-replace program is what enforcing that judgment actually costs.

Beijing applies the same logic in the opposite direction, which is the most instructive part of the story. In May 2023, the Cyberspace Administration of China ruled that Micron had failed a network security review and barred operators of critical information infrastructure — telecommunications, transport, finance — from purchasing its products, citing “serious network security risks” without publishing what they were.

Two governments, opposite conclusions, identical reasoning: for infrastructure that matters, provenance is part of the attack surface.

That framing carries a consequence most procurement organizations have not absorbed. What is inside your systems, and who could reach it, is a question you will eventually answer. Better on your own timeline than on someone else’s.

What mapping your exposure actually means

“Understand your silicon exposure” is easy to say. Here is what it requires, roughly in order of difficulty.

Wafer origin, not box origin. Your supplier’s country of incorporation tells you nothing. Final assembly location tells you almost nothing. The question is where the wafer was fabricated and where the die was packaged and tested — two different sites, often in two different countries, and frequently not the ones printed on the label.

Tier two and tier three. You probably know your direct suppliers. The concentration risk lives one and two levels below them, where a dozen of your vendors turn out to depend on the same foundry for the same controller. Aggregate exposure is invisible from the tier-one view, which is precisely why it surprises people.

Contractual disclosure. Most supply agreements do not require fab-level disclosure or advance notice of a sourcing change. That is a missing clause, not an unknowable fact. Add it at the next renewal and you convert a research project into a standing reporting obligation.

Re-qualification time, system by system. Sort your platforms by how long it would actually take to move to an alternate part: things you could re-source within a quarter; things that need a year of validation; and things carrying safety certification, regulatory approval, or an installed base that realistically cannot be moved at all. That last category is your real exposure. The rest is inventory management.

Then decide deliberately. Some of this risk is worth carrying. The objective is not to eliminate it. It is to stop holding it by accident.

The alternative is being built, slowly

Reshoring is no longer theoretical, and the commitments are large.

TSMC has raised its Arizona investment repeatedly — $12 billion in 2020, $65 billion in 2024, $165 billion in March 2025, and $265 billion announced in July 2026, covering as many as ten leading-edge fabs, two advanced packaging plants, and an R&D center, with 2-nanometer-and-below production in the plan. Micron announced approximately $200 billion in June 2025 — about $150 billion in manufacturing and $50 billion in research and development — across Idaho, New York, and Virginia, projecting roughly 90,000 direct and indirect jobs, with DRAM output from its first Idaho fab scheduled to begin in 2027. Samsung, Intel, and others are running their own programs.

One detail deserves attention from any resilience-minded buyer: China barred Micron from its critical infrastructure market in 2023. A supplier already pushed out of that market carries one less dependency into a crisis.

Now the caveat, which you should hear from someone who wants these projects to succeed. TSMC has declined to commit to construction timelines and has tied its build pace to market demand. That makes $265 billion a ceiling of intent, not a schedule. Announced capacity is not installed capacity, and installed capacity is not qualified capacity for your specific part.

Treat domestic supply as an option you have to actively buy into, not a rescue that arrives on its own.

The honest counterargument

The other side deserves its due, because it is real.

Domestic chips can cost more. China’s scale, cost structure, and market access are genuine advantages, not propaganda. U.S. fabs still have to prove they can deliver leading-edge volume at yield, not merely pour concrete. Concentration at this level has persisted for years without a rupture, and buyers who priced in disruption in 2021 have paid a premium every year since and collected nothing. And a great deal of what you build does not need advanced silicon in the first place.

All true. But resilience has always carried a premium — that is exactly what insurance is. You do not buy it because you expect the fire. You buy it because you cannot survive the one you did not plan for.

The asymmetry is the whole argument. Overpaying for optionality costs you margin. Getting caught without it costs you the product line.

The bottom line

This is not a political position. It is a risk position.

Map your silicon exposure to the wafer, not the label. Sort your systems by how fast you could genuinely move them. Put disclosure language into your contracts. And weight your long-term commitments toward suppliers you can still reach, and still trust, on the worst day.

You are already making this bet. The only open question is whether you have priced it.