Chip Fab Construction: 18 New Plants in 2025 Feed 2026-27 Ramp

The forecast, and what it actually counted

In January 2025, SEMI's quarterly World Fab Forecast projected that chipmakers would break ground on 18 new fabrication plants during the year. The group defined the slate as three 200mm facilities and fifteen 300mm facilities, and it expected most of them to start operating in 2026 or 2027.11 The count covers volume production sites only. Research and development lines are excluded.11 Lead times explain why the forecast looked so far ahead. SEMI analyst Christian Dieseldorff has said it can take 1.5 to 2 years to go from groundbreaking to volume output, so a company that wants production in 2027 has to start building in 2025.14

Nearly two years on, that timing makes the 2025 figure worth another look. The 2025 groundbreakings are the plants that are supposed to be ramping now and next year. SEMI's later outlooks also show the investment cycle around them getting much bigger. The 18 starts were an early marker of a spending wave that later grew faster than SEMI itself first forecast.

Where the shovels went in

SEMI's regional breakdown was more evenly spread than many people expect from an industry with a reputation for concentration in East Asia. The Americas and Japan each had four projects, the most of any region. China and the Europe & Middle East region each had three. Taiwan had two, and Korea and Southeast Asia had one each.1317

Reporting on individual projects helps explain the numbers. The Semiecosystem newsletter, citing SEMI's list, named GlobalFoundries' Fab 8.2 in Malta, New York, and Intel's D1X module 4 in Hillsboro, Oregon, both 300mm fabs. In Japan it named Toshiba's 300mm power-semiconductor fab in Ishikawa Prefecture, phase 2 of TSMC's Fab 23 in Kumamoto, and a UMC foundry project in Mie. It also listed SK hynix's DRAM/HBM fab in Yongin, South Korea.14 The same reporting named Micron and SMIC among the companies expected to start projects that year.14

The coverage does not all line up here. A materials supplier summarizing the forecast said a significant portion of the 18 fabs would be in Asia. It also described geographic spread across Germany and India and put the weight of investment on 2nm and 3nm nodes.16 That account goes beyond what SEMI's release says. SEMI's own breakdown gave the Americas a share of the top spot, and its node analysis put much of the capacity growth in mainstream and mature technologies, not only the leading edge.1117 SEMI's primary figures are the more reliable source.

A comment thread on one hardware news site raised a different objection. Combined, China, Japan, Korea, Southeast Asia and Taiwan account for eleven of the eighteen projects, all in one geopolitically exposed stretch of the western Pacific.19 That is a fair point. The Americas and Japan topping the table is real progress for diversification, but the larger share of new groundbreakings was still in the Asia-Pacific region.

Flat construction counts, and why that matters less than it seems

The 2025 total was not a jump. SEMI recorded 19 construction starts in 2024, eight of them in China, so 2025 was flat to slightly down.14 The mix changed more than the count. China's share fell from eight to three, and the U.S. and Japan moved up.1314 That fits the subsidy programs in those countries. The CHIPS Act in the U.S., the European Chips Act, and similar programs in Japan, Korea and India have been widely credited with speeding up investment decisions.16

The construction count also hides a gap between how many fabs get built and how much money goes into them. SEMI's broader late-2024 data, as summarized at the time, projected fab construction investment slipping about 5% to $39 billion in 2024. Equipment spending, meanwhile, was expected to rise about 8% to roughly $111 billion.15 Most of the money in this cycle goes into tools, not buildings, and a flat number of groundbreakings can sit alongside rapidly growing capital commitments.

Capacity by node: AI at the top, China in the middle

The January 2025 forecast said total industry capacity would grow 6.6% to 33.6 million wafers per month (200mm-equivalent) in 2025.12 It credited leading-edge logic for high-performance computing and generative AI spreading into edge devices as the main drivers.12

The node-level numbers showed a three-way split:

• Advanced nodes (7nm and below): expected to grow 16%, adding more than 300,000 wafers per month to reach 2.2 million.19
• Mainstream nodes (8nm–45nm): projected to grow 6% and pass 15 million wafers per month, driven mainly by China's push for chip self-sufficiency and by automotive and IoT demand.11
• Mature nodes (50nm and above): forecast to grow a more cautious 5% to 14 million wafers per month, which SEMI linked to a slow market recovery and low utilization.11

Foundries were set to lead again, with capacity rising 10.9% from 11.3 million to a record 12.6 million wafers per month.17 Memory as a whole looked more restrained, growing 3.5% in 2024 and 2.9% in 2025. Within memory, though, high-bandwidth memory demand was pulling DRAM and NAND apart. DRAM capacity was expected to grow about 7% to 4.5 million wafers per month, and 3D NAND about 5% to 3.7 million.1317

The SK hynix Yongin project is the clearest single example of the HBM effect turning into new buildings.14 The mainstream-node figures are a reminder that much of the steel going up in China is aimed at chips for cars and industrial equipment, not AI accelerators.

How the spending forecasts kept moving up

SEMI's later forecasts show what happened next. In October 2025, its 300mm Fab Outlook said worldwide 300mm fab equipment spending would pass $100 billion for the first time in 2025, at $107 billion. It projected $116 billion for 2026, $120 billion for 2027 and $138 billion for 2028.3 Six months later, the April 2026 update raised the 2026 figure to $133 billion, an 18% increase, and forecast $151 billion for 2027.15 By the third-quarter 2026 edition, released at SEMICON West in October, the 2026 estimate was $149 billion, up 31% year over year, with spending expected to reach $227 billion by 2029.69

That puts the 2026 estimate about 28% above SEMI's own projection from a year earlier. SEMI CEO Ajit Manocha said in April that AI is "resetting the scale" of manufacturing investment.5 The revisions support that view.

Construction spending is now rising too, after the dip in 2024. SEMI projects 300mm fab construction spending to rise 23% to $47.5 billion in 2026 and 33% to $63.3 billion in 2027, then reach $68.4 billion in 2028.89 The group expects 20 new 300mm fabs or lines to start construction in 2026 and 18 more in 2027. The U.S. accounts for three of the 2026 starts and four of the 2027 starts.68

These numbers do not compare directly with the January 2025 figure. The 2025 count of 18 included three 200mm fabs, while the newer projections cover 300mm facilities only.116 Measured on 300mm alone, the move from fifteen starts in 2025 to twenty projected for 2026 is a clear step up.

The handoff from construction to operation

The 2025 forecast has to be judged on whether those fabs reach production on schedule. SEMI expects 13 new 300mm fabs or lines to begin operations or ship first silicon in 2026, rising to 23 in 2027.68 Rising first-silicon counts in 2027 match the 2026-to-2027 window SEMI gave for the 2025 groundbreakings.11 The 2025 class appears to be arriving roughly on time, though SEMI does not publish a project-by-project tally.

The regional split of equipment spending shows where the next phase is going. In its October 2025 outlook, SEMI ranked China first for 300mm equipment spending from 2026 to 2028 at $94 billion, followed by Korea at $86 billion, Taiwan at $75 billion and the Americas at $60 billion.3 Japan, Europe & Middle East and Southeast Asia trailed at $32 billion, $14 billion and $12 billion, although policy incentives are expected to lift those regions' spending by more than 60% by 2028 compared with 2024.3 The latest edition shows Americas equipment investment more than doubling, from $18.1 billion in 2026 to $40.6 billion in 2029.10

The reading

The 18 groundbreakings of 2025 were never a headline number by themselves. Construction activity was flat on the prior year, and the spread of projects across regions looked more balanced than the money behind them.143 The forecast mattered because it set the timing. Those buildings are the shells into which a record, repeatedly raised wave of equipment spending is now flowing.69 The fab count measured the start of the cycle. The rising equipment and construction forecasts since then show how much the AI-driven buildout has outgrown even SEMI's own expectations.

The main risks are physical and geographic, not financial. A majority of new projects remain in Asia-Pacific, and supply bottlenecks such as EUV availability and field-service engineering capacity have been flagged as limits on how quickly tools can be installed.1910 Funding the plants is less of a constraint now than getting them equipped, staffed and producing chips by 2027.