{"path":"/world-in-transition/ai-and-robotics/global-semiconductor-race","content":{"title":"The Global Semiconductor Race: Supply, Capacity and Competing Ecosystems","summary":"Demand for cutting-edge chips is surging. An assessment of new fab capacity, supply-chain bottlenecks and the US, Chinese and other semiconductor ecosystems through 2030.","body":"*Assessment as of 12 November 2026 | Outlook through 2030*\n\n**The world is building semiconductor manufacturing capacity at an extraordinary rate, but it is not yet keeping up with demand for the most advanced AI chips.**\n\nThe interesting part is that this isn't simply a race between American and Chinese chip factories. There are two competing supply-chain ecosystems emerging, but neither is entirely self-sufficient, and several of the most important players are outside both countries.\n\nMy assessment is:\n\n- **The US-led ecosystem remains substantially ahead at the technological frontier**, particularly in chip design, design software and access to leading-edge manufacturing.\n- **China is building manufacturing capacity aggressively**, especially at mature process nodes, while making a concerted effort to establish a genuinely independent advanced-chip ecosystem.\n- **Taiwan remains the critical centre of advanced logic fabrication**, despite growing investment in the US and Japan.\n- **South Korea is indispensable for AI memory**, and its importance is increasing as memory becomes a bigger component of AI hardware costs and supply constraints.\n- **Europe and Japan control crucial manufacturing technologies and materials.** Neither is likely to dominate AI chip production, but both possess capabilities that the US and China would find extremely difficult to replace.\n\n## 1. How much new capacity is actually being built?\n\n**GLOBAL 300 MM FAB EQUIPMENT SPENDING**\n\n| Year | Forecast annual spending |\n|---|---:|\n| 2026 | **US$149 billion** (+31% year on year) |\n| 2029 | **US$227 billion** |\n\n*Source: [SEMI, 13 October 2026](https://www.semi.org/en/semi-press-release/global-300mm-fab-equipment-spending-to-reach-227-billion-in-2029-as-buildout-accelerates-semi-projects). Forecasts are equipment spending, not physical chip output.*\n\nThe October 2026 forecast from industry association SEMI is striking. It projects:\n\n- 20 new 300 mm fabs or production lines starting construction globally in 2026, and another 18 in 2027.\n- 13 new facilities or lines beginning operation or producing their first silicon in 2026, rising to 23 in 2027.\n- Global fab construction expenditure of approximately $47.5 billion in 2026 and $63.3 billion in 2027.\n\nThese are capital-spending and construction forecasts, not guarantees of productive capacity. Nevertheless, they establish that a major physical expansion is underway. [SEMI, October 2026](https://www.semi.org/en/semi-press-release/global-300mm-fab-equipment-spending-to-reach-227-billion-in-2029-as-buildout-accelerates-semi-projects).\n\nBut there's a critical distinction between *semiconductor capacity* and *cutting-edge semiconductor capacity*.\n\nA new factory producing 28 nm chips adds valuable supply for vehicles, industrial equipment and power management. It does not directly help Nvidia manufacture its next generation of AI accelerators.\n\nEven a new leading-edge fabrication plant isn't necessarily sufficient. To produce a finished AI accelerator, it needs high-bandwidth memory, advanced packaging, substrates, testing capacity and an enormous number of precision components.\n\n**The industry is scaling several interdependent production systems simultaneously.** That makes expansion harder than the headline investment numbers suggest.\n\n## 2. Where are the real supply bottlenecks?\n\nThe greatest constraints are concentrated in a surprisingly small number of activities.\n\n| Supply-chain component | Supply situation in late 2026 | Outlook |\n|---|---|---|\n| Leading-edge logic (2–3 nm) | **Very tight** | Significant expansion in 2027–28 |\n| HBM memory | **Severely constrained** | Tight through at least 2027 |\n| Advanced packaging | **Very tight** | Gradual easing in 2027–28 |\n| High-end substrates and materials | **Very tight** | Selected shortages into 2027–28 |\n| EUV lithography equipment | **Heavily booked** | More capacity, long lead times |\n| Mature-node chips | **Mixed** | Some shortages and some overcapacity |\n\nThese are qualitative assessments across different markets; there isn't a single global shortage percentage.\n\n### Leading-edge logic\n\nTSMC is adding capacity in Taiwan, the US and Japan, but customers continue to compete for its latest processes.\n\nIn the third quarter of 2026, TSMC's wafer revenue came from:\n\n**TSMC WAFER REVENUE MIX — Q3 2026**\n\n| Process node | Share of wafer revenue |\n|---|---:|\n| 2 nm | 8% |\n| 3 nm | 33% |\n| 5 nm | 28% |\n| 7 nm | 10% |\n| Other | 21% |\n\n*Revenue share, not wafer volume. Smaller nodes command higher prices. Source: [TSMC, 15 October 2026](https://pr.tsmc.com/english/news/3346).*\n\nTSMC has allocated US$60–64 billion to capital expenditure in 2026, and its October earnings call described additional 3 nm factories in Taiwan, Arizona and Japan entering production during 2027–28. [TSMC Q3 earnings call transcript](https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-10/eba4fac72873bf546b669bd2b5d973df5c93215d/TSMC%203Q26%20Transcript.pdf).\n\nIts challenge isn't a lack of willingness to invest. It's the time required to install, qualify and ramp equipment, alongside limitations in the supplier network.\n\n### High-bandwidth memory\n\nHBM is arguably the most important emerging bottleneck.\n\nAn AI accelerator needs enormous memory bandwidth. That generally means stacking multiple DRAM dies vertically, connecting them with very fine interconnects, and integrating them physically close to the processor.\n\nOnly three companies currently dominate this market: SK Hynix, Samsung Electronics and Micron Technology.\n\nSamsung reported in October that its planned HBM output for 2027 was already sold out. SK Hynix's CEO has warned that demand could exceed supply beyond 2030, although that is a supplier forecast rather than an independent certainty. [Samsung's October 2026 guidance](https://mbiz.heraldcorp.com/article/10918805).\n\nThere is a second-order effect: **HBM competes with ordinary DRAM for fabrication resources.** Making more AI memory can reduce available production capacity for ordinary server, desktop and laptop memory.\n\n### Advanced packaging\n\nManufacturing the processor die is only part of the task.\n\nAdvanced systems combine processor dies, memory stacks, interposers and high-density connections into packages. TSMC's CoWoS technology is particularly important here.\n\nResearch firm Counterpoint estimates that TSMC's CoWoS capacity is approximately 20% below demand in 2026, with the gap potentially narrowing to 10% in 2027. TrendForce expects packaging shortages to begin easing as investment and alternative suppliers expand. [Counterpoint Research, September 2026](https://counterpointresearch.com/en/insights/global-foundry-20-revenue-rises-25-yoy-to-a-record-966-billion-in-as-advanced).\n\nThis is a major reason the industry is investing in chiplets and advanced packaging technology, not simply smaller transistors.\n\n### The less glamorous components\n\nA particularly revealing example is **T-glass**, a specialised glass-fibre reinforcement material used in high-performance semiconductor substrates and circuit boards.\n\nJapan's Nittobo reportedly controls roughly 90% of the market for one critical T-glass category. TrendForce reported that additional capacity was not expected until at least mid-2027.\n\nOther bottlenecks include ABF substrate films, high-end package substrates, testing equipment, ultra-pure chemicals, precision optics and specialised manufacturing tools.\n\nAn accelerator cannot ship merely because its processor die is ready. Every necessary component must become available in the right quantities and pass qualification.\n\n---\n\n## 3. The American ecosystem: rapidly expanding and technologically powerful\n\nThe US's greatest advantage is that it controls an unusually broad set of high-value capabilities.\n\nIt leads in advanced chip design through Nvidia, AMD, Broadcom, Qualcomm, Apple and hyperscaler custom-chip programs. It is extraordinarily strong in electronic design automation and chip-manufacturing equipment. And it is now rebuilding manufacturing at scale.\n\nThe Semiconductor Industry Association reports that US-headquartered companies accounted for 53.4% of worldwide semiconductor sales in 2025. It also records more than $770 billion in announced private investment across 160 US projects since 2020. Importantly, announcements aren't the same as money already spent. [SIA's 2026 State of the Industry](https://www.semiconductors.org/2026-state-of-the-industry-report-historic-growth-amid-intensifying-global-competition/).\n\n### Major American manufacturing projects\n\n**TSMC — Arizona**\n\nThe biggest example. TSMC's US investment program has expanded to a stated $265 billion, with six planned logic fabs, two packaging facilities and R&D operations.\n\nN4 production is operating; N3 volume production is expected in Q3 2027. Later fabs will bring 2 nm and more advanced processes. [TSMC Arizona](https://www.tsmc.com/static/abouttsmcaz/index.htm); [TSMC earnings call](https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-10/eba4fac72873bf546b669bd2b5d973df5c93215d/TSMC%203Q26%20Transcript.pdf).\n\n**Intel — Arizona, Oregon and Ohio**\n\nIntel 18A entered production in 2025, with further development of 18A-P and 14A. Arizona manufacturing is expanding, while Ohio is a longer-term project.\n\nIntel offers the US a domestically headquartered leading-edge manufacturer, but its ability to win large external foundry customers remains a major commercial test.\n\n**Samsung — Taylor, Texas**\n\nAnother potential advanced-logic manufacturing source. Samsung's October results indicate operations beginning in 2027, with its foundry strategy increasingly focused on 2 nm and 4 nm customers.\n\nThis adds supplier diversity, but Samsung has not matched TSMC's leading-edge foundry scale or commercial position.\n\n**Micron — Idaho and New York**\n\nTwo major Idaho memory fabs are planned, with first wafer output scheduled for 2027 and 2028. Construction of a huge New York complex is underway, with first output planned for 2030.\n\nMicron's total planned US investment now exceeds $250 billion through 2035. Much of this will help supply advanced DRAM and AI memory. [Micron 2026 Form 10-K](https://www.sec.gov/Archives/edgar/data/723125/000072312526000023/mu-20260903.htm).\n\n**Amkor Technology — Arizona packaging**\n\nAmkor has expanded its planned advanced-packaging campus to approximately $12 billion, with agreements involving TSMC and Nvidia.\n\nThis is strategically important because a US-fabricated processor is much less useful for domestic supply resilience if it must travel overseas for essential packaging. [Amkor September 2026 expansion announcement](https://ir.amkor.com/news-releases/news-release-details/amkor-technology-announces-phase-2-arizona-advanced-packaging).\n\n### The American advantage goes beyond fabs\n\nThree US companies are particularly powerful:\n\n- **Applied Materials** — deposition, process equipment and materials engineering.\n- **Lam Research** — etching and deposition systems.\n- **KLA** — inspection, metrology and process control.\n\nIn chip design software, Synopsys and Cadence are dominant alongside Europe's Siemens EDA.\n\nThese are incredibly valuable positions because even manufacturers outside the US depend on their products.\n\nThe weakness is that the US ecosystem is still **multinational rather than domestically self-sufficient**. Taiwan supplies manufacturing expertise, Korea supplies much of the HBM, Japan supplies critical materials and Europe supplies lithography.\n\nThe American strategy increasingly aims to bring more of those activities onto US soil without losing access to the existing international network.\n\n---\n\n## 4. The Chinese ecosystem: catching up under constraints\n\nChina's position is fascinating because it is simultaneously very large and substantially behind at the frontier.\n\nThere are effectively two Chinese semiconductor strategies running at once.\n\n**First: dominate production at established nodes.**\n\nThis is already well underway. SEMI expects China to remain the world's largest installed semiconductor-capacity market in 2026 and 2027. That includes substantial mature-node and memory capacity, not leadership in 2 nm manufacturing.\n\nChina's major foundries include SMIC, Hua Hong and Nexchip.\n\nDemand from AI servers is also supporting mature-node production: a high-end server needs numerous power-management, interface and networking components, many of which do not require leading-edge processes.\n\n**Second: construct an independent advanced semiconductor ecosystem.**\n\nThis is the much harder project.\n\n**China's emerging domestic technology stack**\n\n| Area | Domestic players |\n|---|---|\n| AI processor design | Huawei Ascend, Cambricon, plus newer domestic accelerator designers |\n| Advanced logic manufacturing | SMIC and Huawei-linked manufacturing initiatives |\n| DRAM / HBM | CXMT |\n| NAND flash | YMTC |\n| Etching, deposition, cleaning | NAURA, AMEC, ACM Research Shanghai and others |\n| Lithography | Emerging domestic systems; substantial foreign-technology gap remains |\n\n### China is scaling advanced logic production\n\nIn February 2026, Nikkei reported plans involving SMIC, Hua Hong and Huawei-linked manufacturers to increase relatively advanced production from fewer than 20,000 wafers to approximately 100,000 over one to two years, with more ambitious targets toward 2030.\n\nThose figures are reported plans, not verified achieved production.\n\nHuawei is also demonstrating that China can produce commercially relevant advanced AI hardware. In September, Huawei said demand for its systems exceeded available supply and that it was restricting overseas sales to serve domestic customers.\n\nHowever, there are important limitations.\n\nAdvanced chips fabricated with older-generation deep ultraviolet lithography can require extensive multiple-patterning. This increases complexity, processing time and opportunities for defects.\n\nAs a result, producing something described as a 7 nm or 5 nm-class chip doesn't automatically mean matching TSMC's cost, power efficiency, yield or production throughput.\n\n### China's equipment program is becoming serious\n\nIn July, Reuters reported that China had begun manufacturing domestically developed immersion DUV lithography systems.\n\nInitial production was reportedly very small—about five systems in 2026 and 20 in 2027—and the machines still required substantial improvements in performance and reliability. Domestic EUV remained at a prototype stage. [Reuters, July 2026](https://finance.yahoo.com/technology/articles/china-begins-making-homegrown-duv-141307886.html).\n\nThe broader equipment push is significant. NAURA and AMEC are expanding in manufacturing steps such as etching and deposition, where China has more credible domestic alternatives than in EUV lithography or the most sophisticated inspection systems.\n\nThis is not complete self-sufficiency, but it's a meaningful change in the ecosystem.\n\n### The memory gap is especially important\n\nChina's CXMT has been developing HBM3E, but Reuters reported on 28 October that some production yields were as low as 20%.\n\nSamples have been supplied to domestic chip designers. Yet the report suggests the resulting chips may be more expensive and less capable than their Korean counterparts. [Reuters, 28 October 2026](https://www.reuters.com/world/china/chinas-cxmt-start-advanced-ai-memory-production-despite-poor-yields-sources-say-2026-10-28/).\n\nChina has also made substantial progress in conventional DRAM and NAND. Reuters reported on 6 November that YMTC wants domestic equipment to account for at least 60% of its planned fourth Wuhan factory, while still relying on overseas technology for certain critical processes. [Reuters, 6 November 2026](https://www.reuters.com/world/china/chinas-top-flash-memory-chipmaker-targets-more-homegrown-tools-new-factory-2026-11-06/).\n\nThe important conclusion is that **China is moving from importing a semiconductor ecosystem toward reproducing it domestically.**\n\nIt has large capital resources, a substantial industrial base, a sizeable domestic customer market and strong political incentives to succeed.\n\nThe main constraint is no longer simply whether China can make advanced chips at all. It's whether it can make enough of them, economically and reliably, while independently supplying the equipment, memory and packaging needed for the next generation.\n\n---\n\n## 5. The other players are indispensable\n\nThe US–China framing becomes misleading if it treats everyone else as supporting actors.\n\nIn several parts of the supply chain, other countries actually hold the strongest positions.\n\n**Taiwan — the manufacturing centre**\n\nTaiwan is the most important location for leading-edge contract chip manufacturing. TSMC dominates advanced logic, while ASE and other Taiwanese firms are important packaging suppliers.\n\nTSMC is adding new 3 nm capacity in Taiwan in 2027–28 and developing still more advanced processes.\n\nThe vulnerability is obvious: geographical concentration, especially given the Taiwan Strait's geopolitical risks.\n\n**South Korea — memory superpower**\n\nSK Hynix and Samsung are two of the three central HBM suppliers. They are constructing enormous new memory facilities.\n\nSK Hynix announced 54 trillion won of investment in new Yongin and Cheongju fabs in August 2026. Samsung's latest investment plans include an HBM packaging facility in Asan, with operations scheduled for 2028. [SK Hynix newsroom, August 2026](https://news.skhynix.com/en/fab-facility-investment-2026/).\n\nKorea is also a credible source of advanced logic through Samsung Foundry.\n\n**Netherlands and Germany — manufacturing technology**\n\nASML has a monopoly on commercial EUV lithography equipment. Germany provides essential optical and laser technologies through ZEISS and TRUMPF.\n\nASML's EUV equipment was reportedly largely sold out through 2027 as of September 2026. [Reuters, September 2026](https://www.reuters.com/world/asia-pacific/asml-extends-chipmaking-dominance-customers-embrace-high-na-2026-09-14/).\n\nEurope also has specialist companies such as Infineon, STMicroelectronics and NXP. Infineon opened a €5 billion power semiconductor factory in Dresden in July.\n\n**Japan — materials, equipment and an attempted foundry comeback**\n\nJapan has Tokyo Electron, SCREEN, Advantest, DISCO, Shin-Etsu, SUMCO and other essential suppliers.\n\nIt is also building fabrication capacity. TSMC's second Kumamoto fab is targeting 3 nm production in 2028.\n\nRapidus is attempting to establish domestic 2 nm production, although a Bloomberg report on 10 November said its schedule was slipping because of technical difficulties. [Bloomberg via Japan Times, 10 November 2026](https://www.japantimes.co.jp/business/2026/11/10/companies/rapidus-behind-schedule/).\n\nJapan's manufacturing-equipment and materials positions are much more firmly established than its new advanced-foundry ambitions.\n\n### Other significant participants\n\n**Singapore, Malaysia and Southeast Asia** are important in assembly, testing, packaging and specialist fabrication. This region can absorb some diversification away from China and Taiwan. Broadcom and Powertech, for example, are pursuing new packaging capacity in Singapore.\n\n**India** is becoming a more serious participant in assembly and testing. Micron opened a $2.75 billion memory assembly and testing facility in Gujarat in February 2026. That's meaningful industrial capacity, but not equivalent to operating a leading-edge processor fab.\n\n**The UK** has Arm, a foundational architecture and IP supplier, along with important specialised design and compound-semiconductor capabilities.\n\n**Israel** remains important in chip design, networking technology and semiconductor research, especially through the activities of global companies.\n\nThe Middle East is a large prospective customer and investor in AI infrastructure, but it is not currently a leading centre of advanced semiconductor production.\n\n---\n\n## 6. US versus China: a direct comparison\n\nHere I use *US-led* to mean the US plus its closely integrated international supply chains, not simply factories physically located in America.\n\n| Capability | US-led ecosystem | Chinese ecosystem |\n|---|---|---|\n| Leading-edge chip design | **Clear leader** | Strong and improving |\n| Advanced logic production | **Clear lead through TSMC and allies** | Behind; constrained by equipment and yields |\n| Overall manufacturing capacity | Extensive, geographically distributed | **Enormous domestic scale**, particularly mature nodes |\n| HBM memory | **Strong through Korean and US suppliers** | Emerging, substantial gap |\n| Advanced packaging | **Strong but constrained** | Developing, still behind at frontier |\n| EDA software | **Dominant** | Domestic alternatives developing |\n| Fabrication equipment | **Dominant multinational network** | Improving in selected categories |\n| EUV lithography | **Exclusive practical access through ASML** | No demonstrated mass-production equivalent |\n| Materials supply | Strong through Japan, Europe and US | Broad industrial base, selected dependencies |\n| Ability to produce independently | Not entirely self-sufficient | Not entirely self-sufficient |\n| Ability to scale next-generation AI chips today | **Substantial advantage** | Strong demand, constrained production |\n\nOne particularly important asymmetry is that the US-led ecosystem has access to a network of highly specialised companies, while China's domestic program is attempting to substitute for many of those companies simultaneously.\n\nHowever, the Chinese effort is also creating suppliers that may eventually become international competitors.\n\nChina doesn't have to overtake the frontier immediately to cause substantial economic disruption. If Chinese manufacturers become highly competitive in established processes, power chips, memory and industrial components, they could take market share across a very large part of the semiconductor industry.\n\nThe EU has separately recognised the problem of dependence and is revising its Chips Act strategy to address design, packaging, materials and industrial demand, rather than concentrating on fabs alone.\n\n## 7. How will supply develop through 2030?\n\nThis is my synthesis of the available investment plans and industry forecasts, rather than a claim that precise supply and demand can be predicted.\n\n**2026 — Tight**\n\n**Demand growth is outrunning usable supply.** Advanced logic, HBM and packaging remain constrained. New fab expenditure is enormous, but much of the benefit is still in the future.\n\n**2027 — Expanding**\n\n**New capacity starts to matter.** TSMC's additional 3 nm fabs, Micron's Idaho memory production and Samsung's Texas facility are important. Packaging capacity should increase, but demand may still absorb most of the improvement.\n\n**2028 — Transition**\n\n**A more substantial supply wave arrives.** Additional HBM, 2 nm-class logic and advanced packaging investments should contribute. This is when some bottlenecks may ease significantly, depending on demand growth and successful factory ramps.\n\n**2029–30 — Uncertain**\n\n**A different industry configuration emerges.** America's domestic fabrication and memory capacity should be much larger. China's domestic equipment supply should be more mature. The key question becomes whether AI demand continues to absorb expansion—or the industry enters a period of excess capacity.\n\n### Three uncertainties I would pay particular attention to\n\n**1. How long does the AI capital-spending boom continue?**\n\nNew fabs take years to construct, commission and qualify. Decisions made during the current demand surge will produce capacity toward the end of the decade.\n\nIf demand keeps growing rapidly, those fabs could be heavily utilised. If growth moderates, semiconductor manufacturing's familiar overcapacity cycle could return.\n\n**2. Does China begin to close the lithography gap?**\n\nA domestically competitive immersion DUV production system would materially improve China's manufacturing independence, even if it still lacked EUV.\n\nA commercially effective domestic EUV system would be much more consequential. The distinction between a prototype and sustained, high-yield industrial operation is critical.\n\n**3. Does the bottleneck move from chips to power?**\n\nEven with plentiful chips, building AI infrastructure requires electricity, transformers, cooling, high-speed networking and functioning data centres.\n\nThese constraints can limit how fast purchased chips become productive computing capacity.\n\nThe effective supply of AI compute is therefore not simply the number of chips manufactured. It's the number of chips that can be integrated into working, powered systems.\n\n---\n\n## 8. My overall interpretation\n\nThere are two very different races underway.\n\nThe first is the **race to produce the most powerful chips**. This is where the US-led multinational ecosystem has a commanding lead, and where TSMC, ASML, Nvidia and the HBM suppliers are central.\n\nThe second is the **race to control a complete industrial supply chain**. China is investing heavily in that objective, and the US is increasingly pursuing its own version of it.\n\nI suspect the second race will be at least as consequential over the longer term.\n\nHere's why.\n\nSemiconductors are unusual because the technological frontier is exceptionally valuable, yet there's also a vast amount of economic activity below that frontier. Mature-node chips don't stop being useful simply because a 2 nm process exists.\n\nAs AI spreads from centralised training clusters into local inference, appliances, factories, vehicles and robotics, the demand for different kinds of chips may diversify enormously.\n\nSome applications will need the highest possible performance per watt. Others may work perfectly well with much cheaper chips produced on established processes.\n\nThis suggests a potentially important outcome:\n\n**The US-led ecosystem could retain technological leadership while China becomes exceptionally strong in supplying the broad, lower-cost semiconductor base for the AI economy.**\n\nThose outcomes are not contradictory.\n\nAnd the single most important strategic fact is that the technologically leading ecosystem is still not exclusively American. It remains a complex network spanning Taiwan, South Korea, Japan, Europe and the United States.\n\n### What I would watch\n\nFor the next two years, the strongest indicators are TSMC's actual wafer output at 2–3 nm, HBM supply commitments and production yields, CoWoS and alternative packaging capacity, China's domestic lithography qualification, and the rate at which new fabs transition from construction announcements to commercially productive output.\n\nThe central question is not whether enough factories are being built. Clearly, they are.\n\nIt's **whether the global manufacturing system can expand all its critical components together quickly enough to match the extraordinary growth in demand—and whether that expansion ultimately creates two increasingly independent semiconductor ecosystems.**\n\nMy expectation is that capacity will increase substantially through 2030, but fully independent American and Chinese semiconductor ecosystems are both much further away.\n\n---\n\n## References\n\nThe sources below are the industry reports, company disclosures, research publications and reporting cited in this article. Announced investments and projected capacity are distinguished from completed manufacturing output in the analysis above.\n\n1. **SEMI** (13 October 2026). [Global 300mm Fab Equipment Spending to Reach $227 Billion in 2029 as Buildout Accelerates, SEMI Projects](https://www.semi.org/en/semi-press-release/global-300mm-fab-equipment-spending-to-reach-227-billion-in-2029-as-buildout-accelerates-semi-projects). Global fabrication-equipment spending and construction forecasts.\n\n2. **TSMC** (15 October 2026). [TSMC Reports Third Quarter EPS of NT$29.43](https://pr.tsmc.com/english/news/3346). Quarterly results and advanced-node revenue mix.\n\n3. **TSMC / LSEG StreetEvents** (15 October 2026). [Q3 2026 TSMC Earnings Call Transcript](https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-10/eba4fac72873bf546b669bd2b5d973df5c93215d/TSMC%203Q26%20Transcript.pdf). Capital spending, node roadmap and new-fab plans.\n\n4. **The Herald Business** (October 2026). [Samsung Electronics says next year's HBM supply already sold out, with more requests coming in](https://mbiz.heraldcorp.com/article/10918805). Samsung's HBM demand and supply commentary.\n\n5. **Counterpoint Research** (September 2026). [Global Foundry 2.0 Revenue Rises 25% YoY to a Record $96.6 Billion in Q2 2026 as Advanced Packaging Takes Center Stage](https://counterpointresearch.com/en/insights/global-foundry-20-revenue-rises-25-yoy-to-a-record-966-billion-in-as-advanced). Advanced-packaging industry estimates.\n\n6. **Semiconductor Industry Association** (27 July 2026). [2026 State of the Industry Report: Historic Growth Amid Intensifying Global Competition](https://www.semiconductors.org/2026-state-of-the-industry-report-historic-growth-amid-intensifying-global-competition/). US-headquartered market share, R&D and announced domestic investments.\n\n7. **TSMC** (2026). [TSMC Arizona](https://www.tsmc.com/static/abouttsmcaz/index.htm). Arizona manufacturing expansion and investment plans.\n\n8. **Micron Technology / US SEC** (2026). [Micron Technology 2026 Form 10-K](https://www.sec.gov/Archives/edgar/data/723125/000072312526000023/mu-20260903.htm). US memory-manufacturing capital expenditures and plans.\n\n9. **Amkor Technology** (September 2026). [Amkor Technology Announces Phase 2 Arizona Advanced Packaging Expansion](https://ir.amkor.com/news-releases/news-release-details/amkor-technology-announces-phase-2-arizona-advanced-packaging). US advanced-packaging construction and investment.\n\n10. **Reuters (via Yahoo Finance)** (July 2026). [China begins making homegrown DUV lithography machines](https://finance.yahoo.com/technology/articles/china-begins-making-homegrown-duv-141307886.html). Chinese domestic lithography progress and challenges.\n\n11. **Reuters** (28 October 2026). [China's CXMT to start advanced AI memory production despite poor yields, sources say](https://www.reuters.com/world/china/chinas-cxmt-start-advanced-ai-memory-production-despite-poor-yields-sources-say-2026-10-28/). Chinese HBM3E production and yield reports.\n\n12. **Reuters** (6 November 2026). [China's top flash memory chipmaker targets more homegrown tools for new factory](https://www.reuters.com/world/china/chinas-top-flash-memory-chipmaker-targets-more-homegrown-tools-new-factory-2026-11-06/). YMTC equipment localisation target.\n\n13. **SK hynix Newsroom** (August 2026). [SK hynix Invests 54 Trillion Won in Yongin Y2 and Cheongju M17 to Secure Mid-to-Long-Term Production Base for AI Memory Demand](https://news.skhynix.com/en/fab-facility-investment-2026/). Korean HBM production infrastructure expansion.\n\n14. **Reuters** (September 2026). [ASML extends chipmaking dominance as customers embrace High-NA](https://www.reuters.com/world/asia-pacific/asml-extends-chipmaking-dominance-customers-embrace-high-na-2026-09-14/). EUV lithography demand and ASML equipment position.\n\n15. **Bloomberg (via The Japan Times)** (10 November 2026). [Rapidus behind schedule](https://www.japantimes.co.jp/business/2026/11/10/companies/rapidus-behind-schedule/). Rapidus manufacturing timeline and technical difficulties.\n","author":"Emlyn O'Regan","hero":"01a1ce49-60ca-727e-964d-5a0c94bbded7","media":[{"asset_id":"01a1ce49-60ca-727e-964d-5a0c94bbded7","alt":"Editorial illustration of a multicoloured silicon wafer and advanced processor showing the global semiconductor supply race."}],"social_description":"Who can supply the next generation of AI chips? A global analysis of fabs, HBM, packaging and equipment—and the race between US-led and Chinese ecosystems."}}