Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

TSMC's Texas plan hinges on a tax credit Congress may not extend

TSMC is weighing a Texas campus of fabs costing at least $20bn each. The plan depends on a 35% manufacturing tax credit that requires construction to start by the end of December, The Next Web reported on 3 November.

WorldAnalysisDr. Amara PatelPublished: 3 October 20266 min readSources 3
TSMC's Texas plan hinges on a tax credit Congress may not extend

The Next Web reported on 3 November that TSMC is considering a Texas campus of multiple fabs, each costing at least $20bn, to build more AI chips. The report cites people familiar with the plans. It says TSMC declined to comment on market rumours. According to that report, TSMC also weighed Singapore. Not Europe.

The timing is the story. The 35% manufacturing tax credit behind the maths requires construction to start by the end of December. No extension has a clear path this year. Senators Mike Crapo and Ron Wyden say they are committed to one, per The Next Web. Until that lands, a campus worth tens of billions sits on a legislative calendar, not a construction schedule. For chip buyers and the states courting fabs, the next round of US AI capacity is a political bet, not a signed order. The shortlist is Texas and Singapore. Neither company has said when a decision is due.

TSMC has already committed $265bn in Arizona: ten fabs, two packaging plants and a research centre. North America brought in more than 75% of its wafer revenue this year. A single Texas fab would cost about twice the whole Dresden project, which is budgeted above EUR 10bn.

Europe's position is the contrast. Brussels approved EUR 5bn in state aid for the Dresden joint venture in August 2024. That plant will make 40,000 wafers a month. Equipment moves in this half, production starts in 2027 and full capacity comes in 2029. It runs 28 and 22nm CMOS alongside 16 and 12nm FinFET. That is several generations behind the nodes Nvidia and Apple buy. No AI chips come out of Dresden.

What the export controls changed

The US controls on advanced chips bound for China are the reason capacity location matters so much. Restrictions have pushed buyers toward workarounds. Enforcement has turned into a running caseload. Ars Technica reported on 2 October that a US tech CEO was arrested and accused of smuggling $300m in Nvidia chips into China. The Register covered the same arrest the same day. It described a Californian accused of shipping the hardware without Washington's approval. Transport Topics, the Los Angeles Times, Quartz and Business Insider all carried versions of the charge on 2 October.

The numbers in those cases are small next to the buildout, but they set the tone. ProductNation reported on 3 October that the US says two Malaysia-based firms sent $176m in a chip-smuggling case. Free Malaysia Today reported the same day that a Penang-based company denies moving AI chips to China. Neither company has been convicted, and the allegations are just that. The pattern, though, is a control regime that keeps finding new routes.

Agents can also produce output much faster than humans, though the quality of that output varies.

Epoch AI's estimate, published on 3 November, puts the scale of the buildout in plain terms. AI chips shipped through 2027 could run tens to hundreds of millions of concurrent frontier-model agents. Running nonstop, those agents would supply as many weekly working hours as about 140 to 720 million full-time employees. HBM shipped during 2025 and 2026 could support 16 to 56 million concurrent agents once deployed. Adding shipments through 2027 raises that to roughly 30 to 170 million.

Then the arithmetic gets uncomfortable. Using 20% of Epoch's central capacity estimate would imply $2.6tn to $5.3tn a year in API-equivalent spending. That compares with roughly $1tn in developer revenue by the end of 2027 at fivefold annual growth. Epoch is explicit that demand could fall behind this potential supply, creating an overabundance of capacity. The key uncertainty it names is whether sustained, rapid growth in demand will justify the investment.

That is the same question hanging over the Texas fabs. A fab takes years and billions to stand up. Export controls shape who can buy the output. A tax credit shapes whether the fab gets built at all.

The sceptics and the memory squeeze

Ed Zitron's newsletter, published on 22 September, argues that Nvidia's revenue growth is almost entirely the result of speculative purchases. He also revisits a Guardian investigation he worked on. It found Microsoft had approximately 2.2 million GPUs, backed by around $50bn in chips and roughly 1.993GW of capacity. Bloomberg later reported, per Zitron, that only about 2GW of Microsoft's 12GW of capacity is centred on AI-specific chips.

Zitron's piece is opinionated and he says so. Treat the framing as his. The Microsoft figures are attributed to the Guardian investigation and to Bloomberg. The gap between 12GW total and 2GW AI-specific is the part that matters for anyone modelling demand.

Memory is where the controls bite hardest in practice. HBM is the component Epoch's whole estimate rests on. It is also the component smugglers reportedly target. Reports on 3 November describe smugglers pairing $4,000 AI memory with cheap server chips before stripping them apart in China. That is the enforcement problem in one sentence: the valuable part is small, dense and easy to move separately from the hardware it belongs in.

China's buying in older gear tells a parallel story. The South China Morning Post reported on 2 October that China's buying spree in ASML DUV equipment has prompted US calls for a total export ban. DUV tools are not the leading edge. But they are the machines that make the mature chips Dresden will also make. A total ban would widen the split between the two supply chains rather than narrow it.

Tencent, meanwhile, is leasing 100,000 AI chips from Oracle for $7bn, according to ET Datacenters on 3 October. BigGo Finance reported the same deal and said China's AI routes around restrictions via Southeast Asian data centers. Neither Oracle nor Tencent has said how the arrangement is structured. The deal is a reminder that controls move the market, they do not close it.

Europe's gap is structural

The EU wants 20% of the global chip market by 2030. It held 9.8% in 2022, and its own forecast puts it at 11.7%. The European Court of Auditors called the goal very unlikely in April last year. Industry has asked for a Chips Act 2.0, and the Commission is drafting one.

The money tells you why. The EUR 86bn the first act was meant to mobilise is mostly not EU money. The Commission funds 5%, with member states and private investors covering the rest. Europe's record is not empty. Infineon opened a EUR 5bn power chip fab in Dresden in July on EUR 1bn of EU support, the act's first win.

TSMC keeps its most advanced research in Taiwan whatever it builds abroad. So the question for Europe is not the leading edge. It is whether Europe is in the room when the next campus is picked. On current evidence, the answer is no: the shortlist was Texas and Singapore.

Taiwan has its own lever. It said in October it would curb AI chip exports to China as part of a US trade deal, per a 15 October report. That adds a second gatekeeper to the same flow the US is trying to control. It also puts Taipei in the enforcement business.

None of this settles the Texas question. TSMC has not confirmed the campus, and the tax credit is still a promise from two senators. What is clear is the direction of travel: capacity is being placed where policy allows, not where demand is closest. Buyers should plan for the shortage to outlast the announcement cycle.

Comments 0

Sources

3
  1. 01TSMC eyes a multibillion-dollar Texas campus for more AI chipsEN
  2. 02How many AI agents could run on the AI chips shipped through 2027?EN
  3. 03Where're All the AI Chips?EN

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Dr. Amara Patel

Dr. Amara Patel

Economy, business and world

Dr. Amara Patel covers business, world affairs and the economy for FLASH24, working from filings, central bank statements and trade data rather than press releases, and she does not let company spin stand in for numbers. She checks revenue recognition, debt covenants and currency effects line by line against audited reports and regulatory disclosures. Her week includes calls with analysts, logistics operators and trade lawyers, and she watches the calendar for rate decisions, earnings dates and port and freight updates, comparing each against prior quarters. Outside the desk she tracks tech-company accounts and rides cargo bikes, which keeps her close to both the balance sheets she reads and the supply chains she covers. She does not publish a figure she cannot trace to a primary document.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.