The short version
Part 3 argued that entrenched positions in this industry rarely fall to ordinary competition, and that political intervention is now the most active force acting on them. This part is about that force.
A note on tone before starting. This subject attracts strong opinions and confident predictions. What follows sticks to what is documented, attributes contested claims, and says plainly where the evidence does not settle the question.
What export controls actually restrict
Beginning in October 2022 and tightening repeatedly since, the United States has restricted the sale of advanced computing hardware and manufacturing equipment to China. The controls target three layers.
The chips themselves, above defined thresholds of performance and interconnect bandwidth. Nvidia responded by designing reduced-capability parts specifically for the Chinese market, which were then restricted in turn as the thresholds moved.
The equipment, which is the more consequential restriction. Without EUV lithography machines, Chinese fabs cannot manufacture leading-edge chips efficiently. Because ASML is Dutch and much other equipment is Japanese or American, this required coordination between governments, which was obtained.
The organisations, through the Entity List. The US added dozens of Chinese entities across multiple rounds in 2025.
The picture in 2026 is less a wall than a moving and politically contested boundary. Nvidia was required to obtain licences for its H20 chip. Then, in August 2025, the Bureau of Industry and Security approved the H20 and AMD's MI308 for sale in China on terms under which, in the Congressional Research Service's words, "the U.S. government would receive 15% of proceeds." A government taking a cut of specific export sales is an unusual instrument, and it established the pattern that later approvals were measured against. Nvidia recorded USD 4.6 billion of H20 sales in one quarter of its 2026 financial year, and took a USD 4.5 billion charge when the rules moved against the product. As of February 2026 it was reported to be concerned that Chinese domestic alternatives could take the market before its approved products sold.
That last detail captures the policy's central tension. Restricting sales denies China capability. It also hands Chinese suppliers a protected domestic market.
Where China actually stands
Start with what is not knowable. You will see confident numbers for how far behind China is, often expressed as a percentage of US capacity or a multiple of US compute, and often attributed to an official-sounding source. Treat them carefully. The underlying production data is not published, the credible public estimates disagree with each other, and figures in this debate get repeated far from whatever document first carried them.
What can be said with sources behind it is qualitative, and it differs sharply by layer.
Logic manufacturing. SMIC produced a 7nm processor for Huawei in 2023 using older deep ultraviolet equipment, which surprised many observers. Doing so requires patterning the same layer in multiple passes, which reduces yield and raises cost considerably. In December 2025 a teardown found SMIC had refined that process further, into a variant used in Huawei's Kirin 9030, but independent teardown analysts still describe it as a scaled evolution of the same 7nm-class family rather than a true 5-nanometer-class node, which leaves Huawei working with increasingly dated technology relative to the leaders.
AI accelerators. Huawei's Ascend line is the main domestic alternative and is being produced at scale. Its constraint has reportedly been less the processor than the memory beside it, which is the same HBM bottleneck described in part 1, made worse by restricted supply.
Memory. This is where China has moved fastest, and it matters most to Korea. CXMT has been mass-producing HBM2 since the second half of 2024, and is estimated to be several years behind the global leaders. It has said it is targeting HBM3 production by the end of 2026 and HBM3E in 2027, though multiple 2026 reports describe that timeline as slipping. It is expected to finish 2026 with around 350,000 DRAM wafer starts per month, with a meaningful share of that dedicated to HBM3. It has reported its first annual profit and listed in Shanghai in July 2026. YMTC is running an "all-domestic" pilot line in Wuhan.
Three or four years behind sounds comfortable until you consider what CXMT is competing for. Much of the memory market is not the leading edge. If Chinese capacity floods the conventional DRAM segment, it compresses margins across the industry, including for the Korean firms currently earning record profits.
The lithography wall
The single hardest constraint on Chinese advancement is the one described in part 3. There is no substitute for EUV, and Dutch export-control rules, coordinated with the United States, bar ASML from selling EUV systems to China. Domestic Chinese lithography programmes exist and receive substantial state funding, but reproducing a machine that took a multinational consortium two decades to develop is not a problem money solves quickly.
This is why the equipment controls matter more than the chip controls. A restricted chip is one product generation. A restricted tool is every product generation that tool would have produced.
Everyone is de-risking, slowly
The concentration that makes the industry efficient also makes it fragile, and every major government has concluded the same thing at roughly the same time.
The United States has subsidised domestic fabs and secured commitments from foreign manufacturers. TSMC has substantially raised its Arizona investment commitment, pulled volume production of its third fab forward to 2027, and expects that when all announced facilities are complete, a substantial share of its 2nm and more advanced capacity will sit in Arizona. Samsung is building in Taylor, Texas, anchored by a large, multi-year Tesla contract.
Japan is rebuilding manufacturing capacity it lost decades ago, while retaining its commanding position in materials.
The European Union has its own subsidy programme, with ASML as the anchor asset it does not need to attract.
China is pursuing self-sufficiency across the entire chain, at a cost it has decided to accept.
Korea is doing all of the above simultaneously, which part 6 covers.
The common thread is that all of this is measured in decades. A fab takes years to build and more years to qualify. Nothing announced in 2026 changes the concentration in 2026.
The Taiwan question
Which leaves the risk nobody has solved.
TSMC manufactures the large majority of advanced chips, and its position at the leading edge is stronger still, though no public breakdown measures that separately. Those fabs are on an island whose political status is disputed by a neighbouring power. Every serious analysis of supply chain risk reaches the same place and stops, because the honest conclusion is uncomfortable: there is no substitution path on any short timescale. If leading-edge Taiwanese production were interrupted, no combination of money and urgency would replace it within a year or even several.
The "silicon shield" argument holds that this very centrality is protective, since disruption would damage every major economy including the aggressor's. It is a real argument and it may be right. It is also, notably, a hope rather than a plan, and the scale of Arizona, Kumamoto and Dresden investment suggests governments are not relying on it.
The 2019 lesson Korea already learned
In July 2019, Japan restricted exports to Korea of three chemicals essential to chipmaking, including the photoresists needed for advanced lithography. The dispute was political, arising from a historical grievance rather than anything to do with semiconductors, and it demonstrated something Korea has not forgotten: a close ally can convert a supply relationship into leverage very quickly.
Korea responded with a national localisation programme, and it worked faster than most observers expected. By the Trade Ministry's own account, hydrogen fluoride imports from Japan fell 66 percent by value in two years, from USD 36.3 million in 2019 to USD 12.5 million in 2021.
Even EUV photoresist, the hardest of the three to replace, moved: Korea's dependence on Japan for it fell below 50 percent by 2021 as Belgian supply came in.
The limit is a different one, and worth stating precisely because it is often muddled. Japanese firms still hold roughly 90 percent of the global photoresist market. That is a fact about world supply concentration, not about Korea's exposure to one government, and the two get conflated constantly. Korea reduced its political vulnerability. Nobody has reduced the industry's structural dependence on a handful of Japanese chemical firms.
The episode is the best single illustration of the position Korea occupies. Enormously important in some layers. Deeply dependent in others. Which is the subject of the final part.
Part 6 sets out where Korea and Korean companies actually stand, what has been committed, what is genuinely strong, what is genuinely weak, and what all of it means for people living here.