Guide

New semiconductor fabs in 2026: the global construction queue, and where India's timeline sits inside it

Published 25 August 2026. Facts verified to 25 August 2026 unless dated otherwise.

A fab under construction is a calendar, not a press release, so this page tracks the world's active builds by the dates their owners and the trade record actually put on them.

What this page counts, and what it refuses to count

Fab coverage runs on two numbers that are easy to publish and hard to verify: the announced cost of a plant and its eventual capacity. Both are usually company guidance restated by outlets that did not check it, both change without correction, and both are frequently attached to the wrong entity when a project has several phases. This page therefore prints statuses and dates, which are observable, and avoids per-plant capacity and capital figures except where a named source carries them and this desk has read that source. Where the circulating cost figures for a project disagree with each other, the page says so and prints neither.

The page also has a narrow remit. It is not an account of national chip programmes and their politics, which sits at the global chips race and where India fits, nor a comparison of subsidy regimes, which sits at chips acts compared. This is the construction queue: who is pouring concrete, moving tools and qualifying lines during 2026, on what stated dates, and what that queue does to the cost and availability of the things India's own build needs.

Read that way, the global buildout is not background context. It is the market in which the Dholera project buys its lithography tools, hires its contractors and competes for process engineers, and it is the comparison set against which any Indian slip should be scored.

The one global number worth printing

The most reliable aggregate view of the buildout comes from SEMI, the industry association that maintains the fab tracking datasets the equipment industry itself plans against. Its 300mm fab outlook projects global 300mm front-end equipment spending reaching a record US$133 billion in 2026, an increase of 18 percent on 2025, followed by US$151 billion in 2027, which would be the first year above US$150 billion in the industry's history. SEMI projects further increases to about US$155 billion in 2028 and about US$172 billion in 2029, with total 300mm equipment spending across the three years to 2028 put at US$374 billion.

The same tracker gives a sense of the physical count rather than the money: SEMI reports coverage of 404 facilities and lines globally, with nine new fab and line projects added since the previous edition published in December 2025. On regional spending, SEMI places China as the largest spender across 2026 to 2028 at about US$94 billion, followed by Korea at about US$86 billion and Taiwan at about US$75 billion.

Three things follow for a reader of this record. Equipment demand is at a record and rising, which sets the queue position and the price for anyone ordering tools now. The count of tracked facilities is growing rather than consolidating, so the competition for skilled construction and process labour intensifies through the period in which Dholera must hire. And the spending is concentrated in East Asia, which means the supplier relationships being formed in 2026 are being formed largely elsewhere.

North America: four projects, two on schedule and two rescheduled

TSMC in Arizona. The largest single construction programme in the United States is at north Phoenix. Trade and wire reporting in April 2026, including coverage carried on Yahoo Finance recording the start of construction on a third Arizona plant, put the second fab as structurally complete and intended for 3nm-class production, with the third breaking ground at the same site in the same month and earmarked for 2nm and A16 process technologies REPORTED. Equipment move-in for the second fab has been reported for the third quarter of 2026 with high-volume production targeted in 2027 TARGET. TSMC's own published description of its Arizona site, checked on 25 August 2026, sets out a programme of six logic wafer fabs, two advanced packaging facilities and a research and development centre. Headline investment figures for the site circulate in at least two incompatible forms in secondary coverage, so this page prints none of them.

Samsung at Taylor, Texas. The Taylor site moved from construction into production preparation during 2026. The Korea Times reported on 13 July 2026 that the Taylor fab had entered a production phase for Tesla's AI5 processor, and TrendForce reported on the same date that the design had completed tape-out, clearing the way for a 2nm ramp at the site REPORTED. The schedule is genuinely disputed: some reporting through 2026 has full-scale mass production slipping into early 2027, while Samsung's own framing has been that start of production means completing preparations for mass production by the end of 2026. Both readings are in the public record and this page carries the disagreement rather than a single date.

Micron at Clay, New York. The clearest example of a rescheduled megaproject. Construction Dive and Engineering News-Record both describe the Clay programme as a roughly US$100 billion investment, and TrendForce reported on 10 November 2025 that the New York fabs faced a delay of two to three years, moving the first facility to 2030 from 2028 REPORTED. Construction Dive's account of the revision places the start of first-fab construction in the second quarter of 2026, second-fab construction from the fourth quarter of 2030 to the fourth quarter of 2033, and subsequent phases stretching towards 2041. The county executive for Onondaga, Ryan McMahon, attributed the delay to longer construction cycles and labour shortages, which is a cause worth noting because it is not a chip-market cause at all.

Intel in Ohio. The longest slip in the current cycle. CNBC reported on 28 February 2025 that Intel had delayed the opening of its Ohio plant to 2030, against an original plan to begin production in 2025, and later trade reporting, including Construction Dive's coverage of the revised programme, records the second factory finishing around 2031 with production following REPORTED. A project that was announced as the flagship of American fab revival will start producing roughly five years after its first stated date. That is the single most useful data point on this page for anyone scoring an Indian schedule.

Japan: Rapidus at Chitose

Japan's re-entry into leading-edge logic is concentrated in one project on Hokkaido. Rapidus, working with technology developed with IBM, brought a 2nm pilot line into operation at Chitose in Hokkaido in April 2025, per TechPowerUp's report of the company's confirmation, which also carried mass production set for 2027 TARGET. The same body of reporting records that Japan's first volume-production-ready extreme ultraviolet lithography system was installed in December 2024, that 2nm gate-all-around transistor operation was verified on 300mm wafers, and that a pre-release process design kit was issued to early-adopter customers. Rapidus's own newsroom records that Japan's New Energy and Industrial Technology Development Organization approved the company's plan and budget for the 2026 financial year covering 2nm integration and short-turnaround manufacturing technology and 2nm-generation chiplet and package development DURABLE.

Rapidus matters to this record for a specific reason rather than a general one. It is the clearest working example of the sequence a national programme actually has to run: pilot line first, process design kit to customers second, volume production third. India's own project sits at an earlier point on the same sequence with a different node ambition, and the intermediate milestone that Rapidus has publicly passed, putting a design kit in customers' hands, is precisely the milestone no Indian commercial fab has yet reached publicly.

Europe: ESMC at Dresden, the closest analogue to Dholera

The European project most worth an Indian reader's attention is the joint venture at Dresden between TSMC, Bosch, Infineon and NXP, operating as the European Semiconductor Manufacturing Company. TSMC's newsroom recorded the groundbreaking, and Digitimes described it in August 2024 as a fab of about EUR 10 billion. TrendForce reported on 20 November 2025 that the plant had wrapped its main structural build and was eyeing equipment move-in in the second half of 2026, with fab systems installation running through the first half of that year and production targeted for 2027 TARGET. The plant is described as Europe's first foundry able to support 28 and 22nm and 16 and 12nm process technologies.

The parallel with Dholera is close enough to be analytically useful. Both are mature-node foundry plants rather than leading-edge logic. Both are built by a consortium in which a Taiwanese foundry supplies the process and local partners supply capital, demand or both. Both are aimed at automotive and industrial customers rather than at consumer processors. And both are justified by supply security arguments made after the 2021 shortage rather than by a cost advantage over incumbent Asian capacity. Dresden broke ground in 2024 and targets production in 2027. Dholera broke ground in March 2024 and is guided to commercial production in mid-2028. The gap between those two schedules, roughly a year, is the honest measure of India's position on this particular ladder, and the reasons for it are ecosystem depth and construction sequencing rather than ambition.

The crunch this queue creates, and who feels it

A record equipment spend and a growing count of tracked facilities produce three pressures at once, and each one reaches India through a different channel.

• Tools. Lithography, deposition, etch and metrology tools are made by a small number of suppliers with fixed annual output. When spending hits a record, delivery slots lengthen, and a first-time buyer with no purchase history sits behind repeat customers. This is why the memorandum between ASML and Tata Electronics announced on 16 May 2026 by both companies, naming 28, 40, 55, 90 and 110nm nodes, is treated on this record as a schedule signal rather than a diplomatic one.
• Contractors. Cleanroom builders, process piping specialists and high-purity mechanical trades are a globally mobile and finite workforce. The Micron delay at Clay was attributed in part to labour shortage, in a country with a deep construction base. The mechanics of the build itself are set out at how a fab is built.
• People. Every project in this queue needs process integration engineers with fab-floor experience, and they mostly exist in Taiwan, Korea, Japan and the United States. Every new fab bids for the same pool, which is why secondment and training arrangements matter more to an Indian schedule than any incentive line.

Where India sits in the queue

Placing India's build inside this calendar is the point of the page, and it can be done without flattering anyone. The Dholera fab broke ground in March 2024, civil work stood at about 50 percent as of mid-2026 with cleanroom fit-out reported under way REPORTED, and commercial production is guided to mid-2028 per the Union minister's statement of 17 July 2026 TARGET. No supported first-silicon date exists, so any claim of first silicon in December 2026 is unsupported and is not carried here.

Against the queue, that sequence is unremarkable in the ordinary sense of the word, which is the finding. A plant that broke ground in early 2024 and targets commercial output in 2028 is running roughly a year behind a comparable European mature-node project that broke ground months later, and comfortably ahead of two American projects that have moved their first output to 2030. India is not an outlier on schedule. It is an outlier on inexperience, which is a different risk and shows up in different places: first-time procurement queues, a first-time contractor base, and a first-time regulatory path for large volumes of process chemicals.

The other honest placement is on node. Nothing in India's approved portfolio competes with what Arizona, Taylor or Chitose are building. The comparison set for Dholera is Dresden, not Phoenix, and that is a sensible place to be given that mature nodes carry the bulk of unit demand in automotive, industrial and power applications, an argument set out at why mature nodes matter.

What the slips teach a reader scoring Dholera

The two large delays in this queue are instructive precisely because neither was caused by anything exotic. Intel's Ohio project moved by roughly five years on a mix of company financial pressure and demand reassessment. Micron's New York project moved by two to three years on construction cycle length and labour availability. Neither was a technology failure, and neither followed a dramatic announcement of failure. Both were revised quietly, in stages, through updated schedules.

That is the pattern a reader should expect to look for at Dholera, and it is the reason this record scores against dated, attributable guidance rather than against enthusiasm. The scoreable target is mid-2028 commercial production, stated by a named minister on a named date. The intermediate markers that would reveal a slip earlier are cleanroom completion, tool move-in, and the first named customer, in that order, because each one has to happen before the one after it. The general anatomy of fab delay is set out at why fabs slip, and the dated Indian build record at the Dholera fab construction timeline.

What this record will watch through the rest of 2026

Five items on the global queue would change how India's position reads. Tool move-in completing at TSMC's second Arizona fab on the reported third-quarter schedule, which would confirm that the equipment supply chain is meeting record demand on time. Whether Samsung's Taylor ramp lands in 2026 or 2027, which is the clearest live test of how far a stated start of production can drift from mass production. Equipment move-in at Dresden in the second half of 2026, since that plant is the closest structural analogue to Dholera. Rapidus holding its 2027 mass production date after a 2025 pilot start. And SEMI's next spending revision, because a downward revision would ease the tool and labour queue that a first-time buyer sits in, while a further increase would tighten it.

On the Indian side the watch list is shorter and more consequential: a company-issued schedule to replace ministerial guidance, cleanroom completion and hookup, the first confirmed tool deliveries, and the first named customer. Until those arrive, the position stated here stands. India has one plant in the global construction queue, it is a mature-node plant with a credible analogue in Europe, its stated dates are not out of line with the queue, and nothing yet in the record supports either the triumphal reading or the dismissive one.