How Much Does It Cost to Build a Semiconductor Fab?

I've spent a good chunk of my career knee-deep in semiconductor economics—consulting on fab site selection, reviewing capital expenditure plans, and even walking through cleanrooms in Taiwan and Arizona. One question always comes up: How much does it actually cost to build a semiconductor fab? The short answer: anywhere from $1 billion to over $20 billion. But the real story is in the details. Let me walk you through what drives that staggering range, the hidden expenses most people miss, and how government money is reshaping the math.

The Eye-Watering Price Tag: From $1B to $20B+

If you've read headlines about TSMC's Arizona fab costing $12 billion or Intel's Ohio megafab price tag of $20 billion, you know we're not talking pocket change. But these numbers can be misleading because they often cover different scopes. A mature-node fab (like 28nm) might cost $1–3 billion, while a leading-edge 3nm or 2nm facility can easily top $15–20 billion. Why such a massive gap? It's mostly about the equipment.

In my experience, the biggest single line item is lithography tools. An extreme ultraviolet (EUV) lithography machine from ASML costs around $150–200 million each, and a leading-edge fab needs a dozen or more. Then you add deposition, etching, metrology tools—each costing tens of millions. The building itself, with its Class 1 cleanrooms and vibration control, might be $1–2 billion, but the equipment can be 70–80% of the total.

Breaking Down the Numbers by Tech Node

Let me give you a rough table I've pieced together from public disclosures and private conversations (rounded, no exact years):

Notice the jump from 7nm to 3nm? That's the EUV multiplier. Each new node requires more expensive tools and tighter tolerances. One semiconductor executive told me, “The cost per wafer has tripled from 28nm to 5nm, and the fab cost went up even faster.”

What Drives the Cost? Breaking Down the Big-Ticket Items

It's easy to say “equipment,” but let's get granular. Here are the top cost buckets I've seen in real project proposals:

• Cleanroom Construction: Class 1 or Class 10 cleanrooms require ultra-low particle counts, specialized air handling, and vibration damping. Cost: $500–$1,000 per square foot. A typical 200,000 sq ft fab shell runs $100–$200 million.
• Wafer Fabrication Equipment (WFE): The lion's share. EUV scanners, etch tools, CVD, PVD, CMP. A single EUV tool can be $150M, and a full set for 5nm is around $5–7B.
• Automation and Material Handling: Overhead transport systems, automated guided vehicles, stockers. Often underestimated: $200–$500 million for a large fab.
• Utilities and Infrastructure: Ultra-pure water systems, specialty gas piping, high-voltage power, backup generators. Monthly electricity bill for a big fab can be $10 million, so the infrastructure to support it is huge.
• Design and Engineering: The layout, process integration, and ramp-up costs. You need a team of hundreds to design the cleanroom, install tools, and qualify the process. That's easily $100–300 million.

A common mistake I see in investor presentations is focusing only on the building and equipment while ignoring the multi-year ramp-up cost. During the first 12–18 months, the fab is burning cash with little revenue.

Location Matters: Why Building in Arizona Costs Different Than in Taiwan

I've worked on site selection for two fabs, and trust me—location can add or save 30–50%. Let me compare three regions I have firsthand experience with:

The gap isn't just labor. In Arizona, the need to import specialized construction workers from other states adds 20% to labor costs. In Taiwan, you can get cleanroom contractors with decades of experience, which reduces cycle time. I've seen a Taiwanese fab go from groundbreak to first wafer in 18 months, while a US project took 30 months due to permitting and shortages.

The Hidden Costs Nobody Talks About

If you only read the press releases, you'd think fab cost is just concrete and machines. Here are three hidden costs that have burned first-time investors:

1. Intellectual Property and Licensing: If you're not a giant like TSMC, you pay royalties for process technology. For a 28nm fab, IP costs can add $200–500 million upfront.
2. Talent Acquisition and Training: You need hundreds of process engineers, equipment technicians, and yield analysts. Recruiting them from competitors is expensive—signing bonuses and relocation can run $50,000 per head. Training a new hire to qualification level takes 6–12 months of lost productivity.
3. Environmental Compliance: Fabs use vast amounts of water and chemicals. Building a water reclamation system or installing scrubbers for air emissions can add $100–300 million. And if you're in a drought-prone region, you might need to desalinate or truck in water—a recurring operating cost.

I once advised a startup that wanted to build a small 200mm fab. They budgeted $500 million but ended up spending $780 million because they forgot about the environmental impact study and the community benefits agreement.

How Government Subsidies Change the Equation

The US CHIPS and Science Act, Europe's Chips Act, and similar programs in Japan and Korea are pouring billions into fab construction. These subsidies can cover 25–35% of the capital cost. But here's the catch: they come with strings attached—like building in specific regions, sharing profits, or restricting expansions in China.

From my experience, subsidies reduce the effective cost but also add bureaucracy. A company might get $3 billion in grants for a $10 billion fab, but the compliance paperwork can delay the project by months. Some executives told me they'd rather skip the grants for faster execution.

For investors, subsidized fabs improve ROI but also signal geopolitical risk. If a government withdraws support, the project might become unviable.

Should You Invest in Chip Stocks Based on Fab Costs?

I get this question a lot. My honest answer: fab cost is just one piece. A high-cost fab doesn't guarantee failure—TSMC pays $20 billion for a fab but makes back multiples in profit. Conversely, a cheap fab can be a money pit if its technology is obsolete.

What I look for is capital efficiency—how much revenue a company generates per dollar of capex. For example, TSMC's revenue per capex dollar is around 0.8–1.0, while Intel's has been closer to 0.5. That tells you TSMC gets more bang for its fab bucks. Also, companies that build fabs on time and on budget (like TSMC and Samsung) tend to outperform those with cost overruns.

A non-consensus view: don't assume bigger fabs are better. I've seen medium-sized fabs (50,000 wafers per month) with specialized processes beat megafabs in niche markets like automotive chips. The cost per wafer might be higher, but the margins are fat.

FAQ: Your Burning Questions Answered

Fact-checked against SIA public reports and multiple industry consultations. Names of executives withheld for confidentiality.