A single word "fab" now carries a price tag that rivals the economy of a small nation. Semiconductor manufacturers are no longer negotiating in the billions; they are negotiating in the hundreds of billions. In 2026, the real cost of a new fab has stopped being a construction line item and has become a multi-year bet on equipment inflation, government cash flow, an unfinished workforce, and a power grid that doesn't exist yet. Every fresh groundbreaking announcement now arrives with a headline number, but that number rarely captures the full capital stack a chipmaker actually has to assemble before a wafer ever moves through the line.
An advanced fab still carries a baseline cost of $10 billion to $20 billion to build and equip, according to SEMI's industry tracking. But that baseline no longer describes what leading-edge players are actually committing. TSMC's own investor disclosures show its Arizona campus grew from an initial $12 billion project in 2020 to a $165 billion commitment by March 2025, and at its Q2 2026 earnings call the company raised that figure again to $265 billion, alongside a 2026 capital expenditure guide of $60 billion to $64 billion. What used to be a single-site decision has become a multi-fab, multi-decade capital program.
Contrary to popular assumption, the building itself is the smaller line item. Speaking at SEMI's Industry Strategy Symposium, Exyte executive Herbert Blaschitz put the physical structure's share at roughly $4 billion to $6 billion of a $20 billion-plus fab, with equipment absorbing the majority of the rest. SEMI's 300mm Fab Outlook confirms this at the industry level, projecting that global equipment spending will climb 18% to $133 billion in 2026 and a further 14% to $151 billion in 2027, driven largely by AI-related advanced-node demand.
The CHIPS and Science Act was meant to blunt this cost curve, but its cash is arriving slower than its press releases imply.
The Congressional Research Service found that by January 2025, the Commerce Department had funded 19 companies across 40 fab projects with $30.7 billion in direct awards plus $5.5 billion in loans, with expected completion dates for those projects stretching from late 2024 all the way to October 2033. By mid-2026, the CHIPS Program Office had allocated roughly $38.7 billion of its $39 billion manufacturing pool yet only about $11 billion had actually been disbursed against verified construction milestones, since funds are paid out only as builders clear negotiated checkpoints, not on signing. For a company mapping out a multi-year build budget, that gap between an announced award and cash in hand is itself a financing cost.
A joint study from SEMI, the National Science Foundation, and McKinsey & Company found that the roughly $390 billion in announced U.S. semiconductor investment through 2030 will require close to 189,000 additional workers, against a talent pool where a mere 3% of engineering graduates enter the industry at all leaving a projected shortfall of up to 157,000 skilled workers by 2030. That gap isn't theoretical: TSMC's Arizona build has already faced repeated construction delays tied to shortages of skilled engineers, according to reporting on the joint study. Every month a fab sits short-staffed pushes its yield ramp and revenue timeline further out, turning a hiring gap into a direct hit on the return profile of the entire project.
Power and Water Are Now Capital Decisions, Not Utility Bills
McKinsey's operations research puts peak fab electricity draw as high as 100 megawatt-hours per hour more than many automotive plants or oil refineries with power sometimes accounting for up to 30% of total operating cost. Water is an equally binding constraint: asset manager Robeco cites industry estimates of 20 million to 38 million liters of water used per fab per day, a volume that can rival a small city's consumption and forces new sites into multi-year negotiations over grid interconnects and water-reclamation infrastructure before a single wafer ships.
Where a fab gets built remains the single biggest cost lever available to a chipmaker. Blaschitz's SEMI ISS remarks put it bluntly: building a wafer fab in the West costs roughly twice as much and takes twice as long as building the same fab in Taiwan, even though process-equipment costs are broadly similar across regions. That gap is why every new U.S. or European site announcement now comes bundled with incentive packages, workforce pipelines, and infrastructure guarantees because the concrete and cleanroom alone no longer explain the bill.
The real cost of a 2026 fab isn't the number printed on the groundbreaking press release. It's the sum of equipment inflation, delayed subsidy disbursements, a stretched labor market, and power-and-water infrastructure that has to be built before the fab can even open its doors.