Two Taiwanese economists have now published on the same question this summer. They reach opposite conclusions.
The question: what does global supply chain reconfiguration actually do to Taiwan?
Chiu Ta-sheng, executive director of the Asia-Pacific Chamber of Commerce and Industry, argues in Commercial Times that this is Taiwan's largest opportunity in years.
His core claim, and the load-bearing one: when supply chains relocate, what moves is production bases — not core competitiveness.
What can't be moved, in his framing: R&D, process capability, supply chain management, and a complete industrial ecosystem. Taiwan, he argues, stopped being a contract manufacturer long ago. In semiconductors, AI chips, servers and high-end components it is now indispensable.
He extends this to the anxiety about Taiwanese firms investing in the US. The hollowing-out fear, he says, reads the situation backwards. Building advanced manufacturing in the largest end-demand market brings firms closer to customers, lowers policy risk, and deepens relationships with global tech leaders. It is evidence of strength, not leakage.
His prescription: keep core R&D, key technology, product design and supply chain management rooted in Taiwan. Let overseas sites take different manufacturing and service functions according to local conditions.
And on Southeast Asia — a point worth sitting with — he argues those economies don't primarily need equipment. They need smart manufacturing, automation, quality management, supply chain integration. Things Taiwan has spent decades accumulating.
So Taiwan's export should stop being only products, and start being technology, management experience, and whole solutions.
His conclusion: the basis of global competition has shifted from lowest cost to greatest resilience. Supply chain management capability is becoming a measure of national competitiveness. Taiwan can move from important participant to important integrator.
—
The read:
Set this next to Tsai Ming-fang's July column and you have Taiwan's actual internal debate, which almost never surfaces in English. Tsai looked at the same reconfiguration and saw dangerous concentration — a US surplus equal to 170% of the central government budget. Chiu looks at it and sees an upgrade path.
The disagreement traces to one assumption. Chiu's argument holds only if core competitiveness genuinely stays put when production moves. Some evidence is already testing that: TSMC's Arizona fab passed 4% of company revenue last quarter, and its first US packaging plant is under construction — packaging being the thing everyone said would stay in Taiwan.
Worth knowing where he's writing from. A chamber of commerce is an industry body, and this reads as advocacy as much as analysis. That doesn't make it wrong. It does mean the optimism is doing a job.
What would settle it: whether the US sites stay assembly-and-packaging, or start pulling process development with them.
Sources: Commercial Times, Aug 5 — opinion column by Chiu Ta-sheng. Comparison piece: Commercial Times / TechNews column by Tsai Ming-fang, Tamkang University, July 14.
Nvidia just named 16 Taiwanese companies as partners in its robotics push.
Read the list carefully and it tells you less about robots than about where the margin in robotics is going to sit.
Commercial Times reports the partners are exhibiting together at the Taipei International Automation Industry Show, which opened Wednesday. Named companies include Advantech (https://t.co/iqNvueuYyy), Delta Electronics (https://t.co/MuEj6jlYVj), Quanta (https://t.co/IU9YZHzE6D), Pegatron (https://t.co/SqEavTambG), Inventec (https://t.co/rMpTUYnyYR), Solomon Technology (https://t.co/aKXW9fLLuN), Leadtek (https://t.co/5WRDjiE0Ha), NEXCOM (https://t.co/5gLbsQA1nO), Kenmec (https://t.co/LNInpHXPwb) and Techman Robot.
What they're showing: physical AI, digital twins, autonomous mobile robots, collaborative arms, humanoids.
Now look at what's underneath every one of those demos.
Delta's robots do real-time 3D mapping, path planning and dynamic obstacle avoidance — running on Nvidia's Isaac ROS Nvblox and cuMotion.
NEXCOM's subsidiary, working with Taiwan's ITRI, is demonstrating a real-to-sim-to-real training loop — capture the physical environment, train in simulation, redeploy to hardware. Built on Nvidia's stack.
Solomon's humanoid and AMR systems integrate Isaac GR00T 1.7, Nvidia's open robot foundation model, plus its Cosmos world model and agent framework, computing on Jetson AGX Thor.
Sixteen companies. One stack.
Taiwanese industry analysts read this as Taiwan's role expanding — from servers, industrial PCs and contract manufacturing into robot bodies, edge compute, AI vision and digital twins.
That's true. It's also the smaller half of the story.
—
The read:
Most robotics forecasts size the hardware market. That's the wrong frame for what happened here.
Nvidia isn't selling chips into robotics. It's installing the simulation layer, the foundation model and the runtime — Omniverse, Isaac, Cosmos, GR00T — before the category has consolidated. Sixteen partners building on it is sixteen companies whose product roadmaps now depend on someone else's software release cycle.
Taiwan has been here before. In servers it moved up from components into full systems and still earns single-digit gross margins, because the stack owner captured the pricing power. The robot body is a more complex thing to build than a server. That doesn't make it a better place to stand.
What would change my mind: Taiwanese firms shipping robotics at volume on a non-Nvidia stack, or building their own foundation models. Neither is visible yet.
Sources: Commercial Times, Aug 19 — reporting by Chang Chia-jui. Partner list and technical details as published; the full roster runs to 16 companies.
#Robotics #Taiwan #AI
A contract manufacturer with a 4.6% gross margin just got upgraded to Buy, with its target price raised 57%.
That combination should look strange. It's the clearest signal yet of how Taiwan's supply chain is being repriced.
Commercial Times reports a major Taiwanese brokerage lifted Compal Electronics (https://t.co/9Qk4WLqGiX) from Hold to Buy this week, taking its target from NT$35 to NT$55 — roughly US$1.12 to US$1.76.
The catalyst: Compal is in position to win ASIC L11 assembly orders from a US cloud service provider, tied to the next generation of Google's TPU supply chain. Initial share is estimated around 10%. Confirmation may come within weeks.
The numbers behind the call:
Q2 revenue NT$238.3B (~US$7.7B), up 18% quarter over quarter and 32% year over year. EPS NT$0.72, ahead of expectations.
July revenue NT$81.0B (~US$2.6B) — down 15% from June as the quarter-end shipping peak passed, but up 38% year over year and a record for the month.
Server revenue is forecast at roughly NT$90B this year, up 108%. For 2027, the estimate is NT$288.8B — up 221%.
The brokerage raised its 2027 earnings estimate 59%, taking EPS from NT$1.93 to NT$3.66.
Now the part that complicates the story.
Compal's Q2 gross margin fell to 4.6%. Q3 is expected to fall further, to around 4.4%. The stated cause: rising memory and component prices squeezing PC and notebook profitability.
Capex has been raised to NT$18B (~US$580M), with new plants in the US and Taiwan due for completion by year-end.
—
The read:
This is the same cost shock I wrote about in the PCB materials chain, showing up one layer downstream. Memory and component inflation is compressing an assembler's margin at the same moment the market is re-rating it upward.
Which tells you the re-rating isn't about profitability. It's about mix. A 4.6% margin PC assembler and a 4.6% margin AI server assembler earn the same today, but they get valued on completely different multiples — and the entire thesis rests on an order that hasn't been confirmed yet.
The falsifier is unusually clean here: if the ASIC L11 allocation doesn't land in the next few weeks, the 2027 numbers don't hold.
Sources: Commercial Times / China Times News, Aug 19 — reporting by Wu Mei-kuan. Brokerage unnamed in the original. Figures as cited; NT$ conversions at roughly 31 to the US dollar.
#Semiconductors #SupplyChain #Taiwan
Four separate price hikes crossed Taiwan's business press this week. Power semiconductors. Legacy memory. Samsung's foundry quotes.
And now the layer underneath all of them.
Commercial Times reports PCB makers are preparing a second round of increases for the second half — on top of the 5% to 30% they've already pushed through this year.
The squeeze is in materials. Copper clad laminate and prepreg supply is tight. Major CCL makers raised 20-30% in the first half alone, with quarterly pressure expected to continue into 2027.
The bottleneck underneath the bottleneck: high-end glass fabric. The supply gap on second-generation Low-DK cloth is estimated at over 60%. The T-Glass gap is widening. Price increases there may run into the first half of 2027.
Here's the part worth sitting with — the same input shock is producing opposite outcomes.
Tripod Technology (https://t.co/zL1op5ylL6) has fully passed costs through. Q2 gross margin: 30.09%.
Dynamic Holding (https://t.co/7udlfGE3iC) has not. Q2 gross margin: 14.02%, down 2.1 points in a quarter, with raw materials named as a main cause.
Chin Poon (https://t.co/iaKVZkvxSN) sits in between — 70-80% passed through, but its auto customers negotiate slowly. Historically a 4-5 month lag between input cost and price adjustment. Even accelerated, 2-3 months.
On substrates, the rolling increases continue: Q3 and Q4 still carry double-digit room, with Unimicron (https://t.co/vUnShMD5Op), Kinsus (https://t.co/4CEq3wdkBy) and Nan Ya PCB (https://t.co/FERveBmflO) named as beneficiaries.
—
The read:
Most AI infrastructure models assume unit compute costs fall over time. The materials layer is currently moving the other way, and has been for three quarters.
What makes this hard to see from outside Taiwan is the lag. Pass-through runs 2-5 months depending on customer type, so the same cost shock lands in different companies' margins in different quarters. It reads as idiosyncratic execution when it's actually one shared input.
The spread between a 30% gross margin and a 14% one this quarter isn't a demand story. It's a pricing-power story.
What would change my mind: CCL capacity arriving on schedule in Q1 2027 and glass fabric gaps closing faster than the current 60%.
Sources: Commercial Times, Aug 19 — reporting by Yang Luo-hsuan. Gross margin figures from Q2 company disclosures as cited.
#Semiconductors #SupplyChain #Taiwan
Six weeks ago, a Taiwanese economics professor made a specific prediction about US chip policy. His argument was simple: no matter how high memory prices go, Washington will not loosen controls on China.
Everything since has proven him right — including Apple.
The setup, July 1. Memory prices were climbing. HBM, DRAM, NAND all tight. The industry hope was that cost pressure would eventually force Washington to ease restrictions on Chinese suppliers.
Tsai Ming-fang — professor of industrial economics at Tamkang University, and an advisor to Taiwan's Executive Yuan on trade and industry — argued that hope misreads how US policy actually works. He gave three reasons.
One: production cost has never been a factor. Since the 2018 China tariffs and the 2025 reciprocal tariffs, rising costs simply haven't entered US decision-making. Both Trump and Biden prioritized export controls that forced supply chain restructuring — which massively raised costs for American brands. They did it anyway. Twice. Under two parties.
Two: Congress is moving the other direction. A bipartisan bill introduced in April would make it harder for Chinese firms to produce AI semiconductors, and would push allies to cut equipment exports to China. Tsai flagged the exposure most people miss: ASML already complies on advanced-node tools, but mature-node equipment and servicing remain unrestricted. Tighten those, and equipment and service providers get caught too.
Three, and the sharpest one: for an American memory maker like Micron, tighter controls aren't a cost — they're protection from low-price Chinese competition. Economic security now outranks production cost. If Washington cared about cost, it would never have disrupted China's role as the world's factory in the first place.
His conclusion, written July 1: price increases aren't a policy failure. They're the opportunity cost of control. As long as the benefit exceeds that cost, there is no reason to loosen anything.
Then came the six weeks.
July 17 — bipartisan lawmakers urge the administration to ban US firms from buying Chinese memory entirely.
July 21 — Washington tightens defense supply chain rules on key materials from China and other non-allies.
July 31 — Apple confirms on its earnings call that it's still evaluating Chinese suppliers, including CXMT and YMTC. Tim Cook's own numbers: memory costs up in the March quarter, up significantly again in the June quarter, expected higher still into September.
Apple was the perfect test case. Maximum cost pressure, maximum leverage.
August 17 — Commerce Secretary Howard Lutnick tells Apple directly that the administration doesn't want it buying Chinese-made memory.
August 18 — reports of that pressure send US memory stocks higher.
The most cost-exposed company in the world asked. The answer was no.
Why this matters beyond Apple. Memory is the tightest link in the AI supply chain right now — DDR4 and SLC NAND are running up another 50% this quarter on legacy shortage. Every model that assumes pricing eventually normalizes through supply relief is implicitly assuming a policy change.
Tsai told you in July that assumption was wrong. Six weeks of evidence says he was right.
—
Sources: TechNews column by Tsai Ming-fang (Tamkang University), July 1. TechNews reporting, July 17 / July 21 / July 31 / Aug 17 / Aug 18. Apple commentary from its Q3 FY2026 earnings call. Lutnick reporting originally via The Wall Street Journal.
#Semiconductors #SupplyChain #Taiwan
The government hasn't broken down where that revenue came from, and that's the live fight. Tax receipts for Jan–Jul 2026 already hit NT$2.88T — 76.7% of the entire 2025 annual take, in seven months.
Opposition legislators say a large chunk is securities transaction tax from a red-hot TAIEX, and are demanding the government publish the breakdown. Their argument: at that collection rate, the payout should be NT$20,000.
Also relevant: local elections are this November, and the DPP opposed a nearly identical opposition-backed handout in 2025 on fiscal-discipline grounds.
Everyone's watching TSMC's price increases.
The more interesting number is happening one layer down, in components almost nobody outside Taiwan tracks.
Power semiconductors are about to get their third price hike of 2026.
#Semiconductors#AIInfrastructure#Taiwan
From October: 10–15% on non-contract products.
That's on top of two earlier waves this year. Cumulative increase versus last year is already running 15–20%.
Three hikes in twelve months, in a component category that spent the last decade in oversupply.
Why it's happening:
AI rack power keeps climbing. Data center architecture is moving to 800V DC.
Higher voltage means higher demands on the parts that control and convert that power — breakdown voltage, efficiency, reliability all have to go up.
The components got harder to make and more valuable at the same time.
Then the supply side tightened from three directions at once:
Wafer, leadframe and packaging material costs all rose. International majors ran short — Infineon ($IFNNY) raised some prices in July. And lead times on certain power devices have stretched to 52 weeks.
One Taiwanese industry columnist, Lin Hsiu-min, put it plainly: IGBTs, diodes, essentially anything a power module needs — all scarce right now. He doesn't expect it resolved before the middle of next year.
Here's the part with real strategic weight.
Following the Nexperia situation and broader shifts in China's supply chain, European and American customers have been actively hunting for non-China sources of these parts.
Meanwhile, very little new 6-inch and 8-inch mature-node capacity has been built in recent years.
Which leaves Taiwan's power device makers as one of the few available alternatives.
You can already see it in the numbers.
Pan Jit (https://t.co/3X8RIbC9av) posted July revenue of NT$1.45B — roughly US$47M — up 31% year over year and a record high.
Utilization near full. Some order visibility extending into mid-2027. Capacity expansion of 20–30% underway, coming online Q4 through Q1.
AI-related products were 11% of its Q2 mix. Target is 15% by year-end.
The read: this isn't a cyclical restocking bounce.
Taiwan's industry treats a third consecutive hike as the signal that power semis have shifted from destocking into genuine supply tightness — and they expect that to run into 2027.
Everyone's modeling what AI does to chip demand.
Fewer people are modeling what it does to the unglamorous parts that keep the racks powered — where the pricing power just moved, and where "not made in China" has become a specification rather than a preference.
Sources:
TechNews, Aug 18 — Taiwan power semiconductor makers plan October price increases
USTV Non-Fan News, Aug 18 — analyst commentary from columnist Lin Hsiu-min
Company revenue figures: Pan Jit July monthly sales disclosure
Taiwan just posted the best export numbers in its modern history.
Its own economists are calling the structure behind them dangerous.
Here's what the boom is actually sitting on:
First half 2026 (all figures USD):
Exports: $416.66B, +47.1% YoY
Imports: $319.25B, +40.3%
Surplus: $97.41B — up $41.69B in a year
Growth forecasts got revised up twice. To 9.64%. Then 9.5% by the ADB.
Now look where it came from.
Exports to the US: $133.31B in six months. Up 69%.
That's 32% of everything Taiwan sells abroad.
China and Hong Kong combined: $99.14B. 23.8%.
Eight years ago those positions were reversed.
2018: US took 14% of Taiwan's exports. China took 41%.
Today: US ~31%. China ~24%.
That's a national economy repointing itself in under a decade.
Former finance minister Christina Liu put a number on the risk.
Taiwan's total 2025 surplus: $157.1B.
Its surplus with the US alone: $150.1B.
Her verdict on that concentration: a very dangerous structure.
But here's the part almost nobody outside Taiwan reports.
Taiwan didn't leave China. It added America on top.
China and Hong Kong are still Taiwan's largest import source.
China is still Taiwan's second-largest source of surplus.
Why that's strange: most of Taiwan's East and Southeast Asian neighbors run trade deficits with China.
Taiwan runs a surplus both ways.
So the headline everyone wrote — "Taiwan pivots from China to the US" — has it wrong.
Taiwan is now maximally exposed to both.
A $150B surplus with a country actively repricing its trade relationships.
And a deepening reliance on Chinese imports underneath it.
That's the foundation under the best numbers Taiwan has ever printed.
#Taiwan #Trade #Macro
🚨🇺🇸 Dr. Tanya Haj, U.S. volunteer in Gaza:
“I held a lifeless child in my arms, there was no equipment to save him. This is not a war, it is a genocide of the innocent.”
Look at how she’s crying. She saw it all.