Semiconductor Plant Construction Drops 45% as Fab Pledges Swell

Two Numbers Moving in Opposite Directions

U.S. chipmaking investment in 2026 shows two trends that point in opposite directions. Companies keep announcing more money for semiconductor plants. TSMC's Arizona commitment reached $265 billion in July. Micron broke ground on a $100 billion memory complex in New York. SpaceX and Tesla approved a Texas chip campus. Over the same period, the money actually spent building factories has fallen faster than in any other construction category the Census Bureau tracks.

Both trends are real, and the gap between them is the main story of this year's factory investment cycle. Announcements describe plans for the end of the decade. Construction spending describes what crews are pouring and erecting this month.

The Census Data: One Category Explains Most of the Drop

Manufacturing construction ran at a seasonally adjusted annual rate of $169.8 billion in July 2026. Spending for the first seven months was 22.2% below the same period of 2025, the weakest of the 17 categories Census publishes15. The total is about 32% below its September 2024 peak of roughly $250 billion1517.

The figures differ slightly by source and month, but they agree. June spending came in at $172.674 billion, down 21.4% from a year earlier1318. First-half spending was $87.0 billion, against $111.7 billion in the same period of 202512.

Nearly all of the decline comes from one subcategory. Census groups chip fabs under "computer, electronic and electrical" construction. That line fell 45.4% in the first seven months of 2026, which accounts for $27.7 billion of the $29.1 billion drop in private manufacturing construction15. Without it, the rest of factory building is down only about 2.1%15. Measured from the September 2024 peak, the fab category makes up about $73 billion of an $81 billion decline19. The fab line peaked at $126.4 billion in June 2024 and had fallen to $52.9 billion by June 2026, a drop of more than 58%12.

Other kinds of construction have moved ahead of factories. Data center construction overtook fab construction in April 2026 and hit a record $68.3 billion annual rate in June12. Power construction also passed total manufacturing construction in May18. Investment money has not left the country. It has moved toward the electricity and computing that AI depends on.

A Completed Cycle, Not a Collapse

Nearly every analysis agrees that this is the downside of the CHIPS-era boom rather than a broad retreat from U.S. manufacturing. Associated Builders and Contractors chief economist Anirban Basu has tied the decline to CHIPS Act subsidies winding down11. In January he also pointed to trade policy as a headwind14. Other analysts explain it through the way fabs are built. Spending is heaviest while the shell and cleanroom go up, then falls off sharply once work shifts to installing tools13.

That view fits the rest of the economy. ISM data showed factory production expanding for a tenth straight month in August, and core capital goods orders reached record levels15. Existing plants are running harder, and the spending has shifted from buildings to the equipment inside them. Even after the pullback, manufacturing construction is still more than double its 2019 average of about $81 billion15.

The politics of the data are less settled. A FactCheck review found that factory construction spending has declined since President Trump took office, even though he has claimed a 41% increase20. Critics count the drop at about 32% from the late-Biden peak16. Defenders of the cycle explanation note that today's construction spending reflects board decisions made 18 to 30 months ago18. Both points hold. In our view, the timing argument is stronger, but it creates a test of its own. If 2026's announcements are real, construction spending should turn up in 2027 and 2028.

The Project Pipeline

The largest U.S. projects show both the scale of the plans and how far they still are from finished.

TSMC, Arizona. On July 16 the company pledged another $100 billion for four more advanced facilities. That brings its U.S. total to $265 billion across 12 facilities. TSMC also raised its 2026 capital budget to $60–64 billion10. The first fab has been in volume N4 production since late 2024, with yields described as comparable to Taiwan23. Satellite imagery tells a more modest story. Of the ten fabs announced for the north Phoenix campus, one is producing chips. A second has its equipment moving in, and a third was being roofed as of mid-2026. Fabs 4 and 6 show only early ground clearing, and fabs 7 through 10 have not broken ground10.

The second fab's start date is disputed. Phoenix city materials and several outlets point to 3nm volume production in 2027231. One analysis cites a 2028 date, two years later than the original target10. Another report says pilot production started about six months early6. TSMC is also reportedly considering a multi-fab campus in Texas, though nothing has been approved9.

Micron, New York and Idaho. Micron's U.S. plan totals $200 billion: $150 billion for fabs and $50 billion for research and development across Idaho, New York and Virginia21. The Clay, New York megafab broke ground in January. It carries a $100 billion price tag, and its first fab is expected to run around 20302326. Idaho production is expected in 202724.

Samsung, Taylor, Texas. Samsung paused construction and resumed it in mid-202527. It cut its planned Texas investment from $44 billion to $37 billion, and its CHIPS award dropped from $6.4 billion to $4.7 billion as a result24. The plant has since shifted to 2nm production45. Samsung's latest earnings report points to commercial mass production at Taylor Fab 1 in 2027. It also says a second fab will break ground by the end of 2026, aiming for 2030 output7. If those dates hold, Taylor would make 2nm chips on U.S. soil roughly two years before TSMC's Arizona 2nm fab7.

Intel, Ohio. Intel is the clearest example of a delay. Ohio One has been slowed to a 2030–2031 operating window23, more than two years behind the original 2027–2028 plan4. Intel's Fab 52 in Chandler, Arizona, did reach high-volume 18A production in early 202654.

New entrants. SpaceX plans to spend $16.8 billion on the first phase of "Terafab," a chip complex with Tesla in Grimes County, Texas. It is meant to combine logic, memory and packaging under one roof, and construction is due to start before the end of the year29. Another database lists a related Austin-area project at $119 billion, still in planning25. SK hynix broke ground in August on a $4 billion facility in Indiana for packaging high-bandwidth memory (HBM)23. Texas Instruments' Sherman fab began production in December 202523.

Where the Coverage Diverges

Not all reporting on project status holds up. Some procurement-focused scorecards describe Intel's Ohio Module 1 as already in volume production2 and Samsung's Taylor fab as making 3nm chips since late 20253. Those claims conflict with Intel's own slowdown23, with the reported 2nm pivot4, and with Samsung's 2027 mass-production guidance7. Readers should treat aggregated scorecards with care and give more weight to company filings and municipal records.

Total announced investment also depends on who is counting. One manufacturing-announcement tracker puts the figure at $1.966 trillion. At the current pace of all U.S. manufacturing construction, building that out would take more than 11 years12. The Peterson Institute found that fewer than 30% of more than $5 trillion in pledged foreign investment can be tied to a specific project15. For semiconductors, the typical gap from announcement to groundbreaking is 28 to 39 months, so this year's pledges mostly represent construction spending in 2027 and 202815.

What Comes Next

The forward indicators point to a recovery, not a permanent decline. SEMI projects that equipment spending for 300mm fabs in the Americas will more than double, from $18.1 billion in 2026 to $40.6 billion in 2029. It expects three new U.S. 300mm fabs or production lines to start construction this year and four in 202728. Global fab construction spending is forecast to rise 33% in 202728.

There are risks. Dodge expects manufacturing construction starts to fall 24% this year15. A possible interest-rate increase in 2027 could slow private construction further11. Building fabs in the U.S. remains far more expensive than in Asia. One estimate put the cost of TSMC's first Arizona phase at about twice that of a comparable Taiwan fab2.

Our view: the fall in fab construction is real and steep, but it signals the end of one phase of the boom rather than its failure. AI demand is now driving chip investment, and the CHIPS Act subsidies are winding down. Whether the industry is healthy will depend on how many of the ten announced Phoenix fabs, the four planned Clay fabs and the second Taylor fab actually reach vertical construction by 2028. Until those numbers appear in the Census data, the pledges remain plans rather than built capacity.