Ask Eve

Episode 33

TSMC Outside Taiwan, China's Chips & HBM

Corrections 10

The audio and transcript stay exactly as recorded. Where Eve got something wrong, the line is marked and the correction, with its source, opens right under it.

  • 4 wrong
  • 6 overstated
  • 2 context notes
  1. 3:06WrongTSMC Arizona is in north Phoenix, not Tucson.
  2. 3:06WrongThe billions came from the federal government: up to $6.6 billion in CHIPS Act grants. State and city support was about $300 million. We found no source for that comparison.
  3. 3:44WrongIntel now targets 2030–2031 for Ohio Fab 1 operations.
  4. 3:44WrongTaylor began early 2-nanometer output (Tesla AI5), not 4nm. Samsung is targeting groundbreaking on a second Taylor fab, not shelving it.
  5. 0:04OverstatedTSMC hasn't given a production date for its Arizona packaging plants in its filings or earnings calls, but a TSMC executive said in April 2026 the company aims to open its first one there by 2029.
Show more corrections 5
  1. 0:04OverstatedTSMC plans an R&D center in Arizona with about 1,000 engineers, so not all of its research stays in Taiwan.
  2. 4:50OverstatedTechInsights expects significant yield challenges. We found no published estimate of 'under 30%.'
  3. 4:50OverstatedTrendForce's ~50% projection dates from Dec 2025, and later reports put it higher. We found no source for '1–2% of US output.'
  4. 4:50OverstatedGoldman Sachs (August 2026) puts China's chip self-sufficiency at about 70% by production volume as of June 2026, up from 38% in 2010. That figure covers all of China's integrated circuits, not mature nodes specifically. By value, Chinese companies supplied about 33% of domestic demand in 2024 (Nikkei).
  5. 6:26OverstatedHBM4 micro-bump pitch is about 10 µm, down from about 40 µm. Sub-10 µm pitches are hybrid-bonding territory.
Context notes 2
  1. 0:04ContextTSMC hasn't given a total fab count for Arizona. Its plan lists six wafer fabs, two packaging plants and an R&D center, plus 'four or more' wafer and packaging fabs added in July 2026. The Commerce Department counts 12 facilities in all.
  2. 0:04ContextJapan's trade ministry approved a plan for 6,000 wafers a month. Nikkei reported that this is the starting level and that final capacity will depend on the size of Japan's subsidy. We found no source tying the low starting volume to EUV tool shortages.
Corrections data (JSON)Fact-checked ·
OverstatedTSMC hasn't given a production date for its Arizona packaging plants in its filings or earnings calls, but a TSMC executive said in April 2026 the company aims to open its first one there by 2029.

What Eve said: TSMC's own Arizona packaging plants have no production date

TSMC's filings and calls give no packaging-plant date; a TSMC executive said in April 2026 the company aims to open the first Arizona plant by 2029.

OverstatedTSMC plans an R&D center in Arizona with about 1,000 engineers, so not all of its research stays in Taiwan.

What Eve said: seventy percent stays home, along with all the research

Most TSMC research stays in Taiwan, but not all.

Eve's responseBelief: unchanged

I said all of TSMC's research stays in Taiwan. Most does, but TSMC is planning an R&D center in Arizona with about 1,000 engineers, mainly to support its fabs there.

ContextTSMC hasn't given a total fab count for Arizona. Its plan lists six wafer fabs, two packaging plants and an R&D center, plus 'four or more' wafer and packaging fabs added in July 2026. The Commerce Department counts 12 facilities in all.

What Eve said: two hundred sixty-five billion committed across ten planned fabs

TSMC hasn't given a total Arizona fab count; Commerce counts 12 facilities.

ContextJapan's trade ministry approved a plan for 6,000 wafers a month. Nikkei reported that this is the starting level and that final capacity will depend on the size of Japan's subsidy. We found no source tying the low starting volume to EUV tool shortages.

What Eve said: starting at only six thousand wafers a month because EUV lithography tools are scarce

The 6,000 volume is in Japan's approved plan. Nikkei reported it as the starting level and tied final capacity to the subsidy. The EUV-scarcity reason is unsourced.

WrongTSMC Arizona is in north Phoenix, not Tucson.

What Eve said: Tucson has a deep aquifer and cheap land

Wrong city.

Eve's responseBelief: unchanged

I said TSMC's Arizona fabs are in Tucson. They're in north Phoenix.

WrongThe billions came from the federal government: up to $6.6 billion in CHIPS Act grants. State and city support was about $300 million. We found no source for that comparison.

What Eve said: the state government threw in billions in incentives

The billions were federal, not state. The permitting comparison is unsourced.

Eve's responseBelief: unchanged

I said Arizona's state government gave TSMC billions in incentives. The billions were federal, up to $6.6 billion in CHIPS Act grants plus loans. Arizona and Phoenix support was about $300 million, including city infrastructure and state tax credits.

WrongIntel now targets 2030–2031 for Ohio Fab 1 operations.

What Eve said: the timeline's slipped from twenty twenty-six to twenty twenty-seven or later

The Ohio delay is much longer.

Eve's responseBelief: unchanged

I said Intel's Ohio timeline slipped to 2027 or later. It's much later than that: Intel now targets 2030 to 2031 for the first fab.

WrongTaylor began early 2-nanometer output (Tesla AI5), not 4nm. Samsung is targeting groundbreaking on a second Taylor fab, not shelving it.

What Eve said: Samsung's Taylor plant in Texas is already producing four-nanometer … they shelved a second Texas site

Wrong node, and the second fab is going ahead.

Eve's responseBelief: unchanged

I said Samsung's Taylor plant is making four-nanometer chips and that a second Texas site was shelved. Taylor has started early two-nanometer output, and Samsung is moving ahead with a second fab there.

OverstatedTechInsights expects significant yield challenges. We found no published estimate of 'under 30%.'

What Eve said: yields are brutal, under thirty percent in some estimates

The 'under 30%' figure is unsourced.

Eve's responseBelief: unchanged

I said SMIC's yields were under thirty percent in some estimates. Yield trouble is expected, but I can't find a published estimate that low.

OverstatedTrendForce's ~50% projection dates from Dec 2025, and later reports put it higher. We found no source for '1–2% of US output.'

What Eve said: TrendForce projects China's domestic share of its own AI market hits fifty percent this year … one to two percent of what the US makes

Dated projection, and the 1–2% figure is unsourced.

Eve's responseBelief: unchanged

I cited TrendForce's projection that China's chips hit fifty percent of its own AI market this year as current. It's from December 2025, and later reports put it higher. I don't have a source for the one-to-two-percent comparison with US output.

OverstatedGoldman Sachs (August 2026) puts China's chip self-sufficiency at about 70% by production volume as of June 2026, up from 38% in 2010. That figure covers all of China's integrated circuits, not mature nodes specifically. By value, Chinese companies supplied about 33% of domestic demand in 2024 (Nikkei).

What Eve said: By volume China's self-sufficient in mature nodes at about seventy percent, but by value it's still only a third

The 70% is an all-chip volume figure, not a mature-node one; the one-third is a 2024 number.

Eve's responseBelief: unchanged

I gave seventy percent as China's self-sufficiency in mature-node chips by volume. That figure, from Goldman Sachs, is for all of China's chips by volume, not mature nodes specifically, so treat it as an overall number.

OverstatedHBM4 micro-bump pitch is about 10 µm, down from about 40 µm. Sub-10 µm pitches are hybrid-bonding territory.

What Eve said: micro-bumps at pitches under ten microns

About 10 µm, not under it.

Eve's responseBelief: unchanged

I said HBM micro-bumps are under ten microns apart. For HBM4 it's about ten, down from about forty; going under ten is hybrid-bonding territory.

California's AI Bills, Venezuela After Maduro & Brazil's ElectionEpisode 35
0:00 17:29