Onpode
Cover art for Why Taiwan is building America's AI chip future, not China's

Why Taiwan is building America's AI chip future, not China's

November 15, 2026 · 10 min

Eliza Ward & Brian Reed

Hey. Good to be back. Yeah, good — you've been looking at this Taiwan thing all morning? I have. And the top line is — January 2026, the US and Taiwan sign a Memorandum of Understanding. Taiwanese firms pledge at least $250 billion into US advanced semiconductors, energy, AI. Big number, big signing. And then…

In January 2026, the United States and Taiwan signed a Memorandum of Understanding under which Taiwanese semiconductor and technology companies pledged at least $250 billion in direct investment to expand advanced semiconductor, energy, and AI production in the United States.

0:0010:10
Get the next episode on World →

Follow it free — new episodes land in your feed.

Or make your own — any topic, in minutes

More Onpode episodes on World →

About this episode

In January 2026, the United States and Taiwan signed a Memorandum of Understanding under which Taiwanese semiconductor and technology companies pledged at least $250 billion in direct investment to expand advanced semiconductor, energy, and AI production in the United States.

Grounded in 7 sources
Taiwan's tech companies to invest $250B in US under new trade deal | AP News ↗ · apnews.com
Taiwan to invest $250B in US semiconductor manufacturing ↗ · finance.yahoo.com
US, Taiwan reach trade deal focused on semiconductors, US Commerce Department says ↗ · finance.yahoo.com
Top AI Chip Companies in the World | Full Industry Guide ↗ · ameztrix.com
Microsoft Maia 300: How Much Nvidia Business Is at Risk? ↗ · io-fund.com
lawfaremedia.org ↗ · lawfaremedia.org
What Is the AI Hardware Stack? Chips, Data Centers, and Power, Explained | MindStudio ↗ · mindstudio.ai
Read transcript

Eliza Ward: Hey. Good to be back.

Brian Reed: Yeah, good — you've been looking at this Taiwan thing all morning?

Eliza Ward: I have. And the top line is — January 2026, the US and Taiwan sign a Memorandum of Understanding. Taiwanese firms pledge at least $250 billion into US advanced semiconductors, energy, AI. Big number, big signing.

Brian Reed: And then February the ART — the Agreement on Reciprocal Trade — formalizes it. Tariff drops from 20 to 15 percent on Taiwanese goods.

Eliza Ward: Right. So that's the deal on paper. Then you get to September 2026 — nine months in — and the actual deployed number is $24.1 billion.

Brian Reed: Out of $250 billion.

Eliza Ward: Out of $250 billion. And wait — $20 billion of that $24.1 billion is TSMC alone. So basically one company.

Brian Reed: So the rest of the Taiwanese semiconductor sector — all those other firms in the pledge — collectively moved, what, $4 billion? That's the part I don't get. Is that just slow paperwork, or is the gap bigger than a timing problem?

Eliza Ward: It might be both, but here's what cuts through the timing argument — Peter Harrell, who actually ran economic security at the Biden White House, said the pledge levels 'never seemed plausible.' Not 'disappointing in execution.' Never plausible.

Brian Reed: Wait — the people who built this deal said that?

Eliza Ward: While it was being signed, basically. And he still acknowledged $37.5 billion as the meaningful number. So think of it like — a contractor bids $250,000 on your renovation, everyone in the room knows the real budget is $37,000, and the contract gets signed anyway. The big number is a signal, not a schedule.

Brian Reed: Right — so the $250 billion is doing political work, not financial work. It's the number that gets announced at the signing. But then — wait, actually there's something weirder underneath that. Because I'm not even sure the $250 billion is one number. There's $250 billion in direct investment and $250 billion in Taiwanese government credit guarantees, and it's genuinely unclear whether those are additive or whether the guarantees are just backing the investment. That's not a small ambiguity.

Eliza Ward: Hold on — so we might be looking at $250 billion total, not $500 billion?

Brian Reed: That's exactly the unconfirmed part. The MOU from January 2026 established both commitments, the ART in February formalized them, and nobody has officially stated whether they stack. So the headline could be overstating the pledge by a factor of two before we even get to the deployment gap.

Eliza Ward: And the tariff — the ART dropped it from 20 to 15 percent. Five points. That's not relief, that's a rounding error. Which tells you the tariff was never the real leverage here.

Brian Reed: No, the real driver is security and production geography, not tariff relief. So the question is whether $37.5 billion — Harrell's 'meaningful' number — actually moves enough semiconductor capacity to justify the architecture around it.

Eliza Ward: But here's what the '$37.5 billion is meaningful' framing misses — when you look at who moved that capital, it's almost entirely TSMC. $20 billion of $24.1 billion. So when people say 'Taiwan is reshoring its semiconductor industry,' that's not what's happening. That's TSMC making a bet.

Brian Reed: Is that a problem though? I mean — TSMC is the dominant advanced foundry on the planet. If you're trying to move critical capacity, isn't one very capable company better than ten smaller ones?

Eliza Ward: It's exactly the problem. Because the whole fabless model — Nvidia designs, TSMC builds — that entire AI hardware supply chain converges at one junction. And if that junction is TSMC, and TSMC is the strategy, then you haven't diversified the risk. You've named it.

Brian Reed: Wait — so Nvidia's billion-dollar supplier deals, the six packaging and substrate firms — those all flow back through TSMC fabrication anyway?

Eliza Ward: Right. Nvidia is fabless. It doesn't own a fab. So every AI accelerator it makes runs through TSMC. The supply chain financing deals are real, but they sit downstream of the foundry layer. That layer is still TSMC.

Brian Reed: So the scenario is — a Taiwanese chip designer calls TSMC to book a wafer run on their latest AI accelerator. TSMC says yes. And that transaction is now, what, a geopolitical event?

Eliza Ward: Logged, reviewed, subject to rules — and here's the part that actually unsettles me — those rules were written after the last time it went wrong. The Foundry Due Diligence Rule, January 2025, came into existence because TSMC-made chips had already reached Huawei. That's not a preventive architecture. That's a reactive patch.

Brian Reed: So the rule exists because the diversion already happened. The barn door and the horse.

Eliza Ward: And we'll get into whether that rule is even still standing — because what John Moolenaar had to ask BIS in August 2026 is genuinely alarming and it changes what any of this architecture is actually worth.

Brian Reed: And that's the thing that stops me — a committee chair had to *ask* BIS whether their own rule was still in force. John Moolenaar, August 2026, pressing the Bureau of Industry and Security to confirm the Foundry Due Diligence Rule still exists. That's not oversight. That's a congressional fire alarm.

Eliza Ward: The rule that came from Huawei. The diversion that already happened.

Brian Reed: Right — and BIS is simultaneously administering the controls and absorbing political pressure from House Republicans who, I mean — we don't know exactly what they want, but Moolenaar asking that question in public suggests the rule's continuation is not settled. That's not a stable enforcement floor.

Eliza Ward: And then June 2026 — Taiwan signals it wants to restrict AI chip sales to Chinese customers directly. Extend the control beyond US jurisdiction. That's the signal. But it's not law. No enforcement mechanism confirmed.

Brian Reed: So wait — the US baseline regime goes back to 2022, the Foundry Due Diligence Rule patches a failure in 2025, and Taiwan's June signal would extend beyond US jurisdiction — but that extension isn't finalized. We're describing a stack of controls where each layer was built after the previous one failed.

Eliza Ward: That's — yeah, that's exactly the architecture problem. Reactive, not preventive.

Brian Reed: And Nvidia just announced billion-dollar financing deals with six suppliers — substrates, packaging, testing, optical engines — to lock in AI accelerator capacity. That's a private company hedging the same geographic risk the governments are supposedly managing. If the architecture were solid, Nvidia wouldn't need to do that.

Eliza Ward: Or they'd do it anyway, because six supply chain financing deals is just good procurement at that scale. I'd want to know if this is — actually no, the timing matters. These are announced against the same backdrop of enforcement uncertainty. That's not coincidence.

Brian Reed: So the two things to watch: does Taiwan's chip restriction become actual law with an enforcement mechanism, and does BIS hold the Foundry Due Diligence Rule under that congressional pressure. Those are the two questions. If either slips, the whole geographic containment logic starts moving on sand.

Eliza Ward: And those two questions don't resolve at the same time. Taiwan's June restriction — that's still unfinished policy. No enforcement mechanism confirmed as of now. That's the clearest near-term read on whether the containment logic actually travels beyond US jurisdiction. If it doesn't get finalized, the whole geographic control thesis — the idea that where chips are built determines leverage — it stays a thesis.

Brian Reed: And the deployment gap doesn't close in the meantime. $24.1 billion against $250 billion — I mean, even if you take Harrell's $37.5 billion as the real target, we're at $24.1 billion after nine months. That's not — wait, actually that's not obviously off-track for $37.5 billion. But it's nowhere near $250 billion, and that number is what Taiwan's credibility is pegged to publicly.

Eliza Ward: The credibility question is real. If that number keeps getting quietly revised — from $250 billion to $37.5 billion to whatever September 2027 looks like — the Trump administration noticed a big gesture. A revised number is a smaller gesture.

Brian Reed: And we genuinely don't know which direction the Foundry Due Diligence Rule goes under that congressional pressure. That's not a rhetorical open question. Moolenaar had to ask. BIS hasn't made the answer obvious.

Eliza Ward: So: the pledge gap, the Taiwan restriction, the BIS rule — none of those are closed. We'll watch them.