Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
AMD CEO Lisa Su estimated in July 2025 that chips TSMC makes in Arizona would cost more than 5% but less than 20% more than comparable Taiwan-made chips. That is an upper-bound estimate for a particular sourcing comparison—not a confirmed 20% increase across all TSMC chips, or a forecast that AI hardware, cloud computing or consumer devices will rise by the same amount.
The premium could add pressure to AI infrastructure costs, especially if it persists as deployment scales. But it may also be the price of diversifying production away from Taiwan. Whether it slows AI growth depends on which costs rise, who absorbs them, and whether customers value a more geographically resilient supply chain enough to pay.
What Lisa Su’s estimate does—and does not—say
Su’s estimate concerned AMD’s expected sourcing of chips made at TSMC’s Arizona facilities: more than 5%, but less than 20% more than comparable chips made in Taiwan. The range was reported from her July 2025 remarks; it was not a published TSMC price list or a claim that every U.S.-made chip costs exactly 20% more. Techmeme’s coverage of the remarks and Tom’s Hardware’s reporting on the range describe the comparison.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteThe public figure does not establish whether the comparison is for wafers, individual dies or packaged chips, or whether production volume, packaging, testing and other terms are identical. It also does not show how much of any additional cost AMD would absorb, pass to customers or offset elsewhere. The estimate cannot be applied mechanically to the retail price of a CPU, GPU, server, cloud instance or AI service.
#1 Best Overall
Why Arizona production can cost more
Arizona starts without the depth of Taiwan’s established semiconductor cluster. A new fab must be built, staffed and supplied; its costs are spread over output that may initially be below a mature site’s level. Industry reporting has cited U.S. construction, labor and staffing costs among the reasons for the premium. Asiae’s July 2025 report discusses those factors.
- Construction and capital: TSMC’s initial Arizona plan called for three fabs and more than $65 billion in investment. Greenfield facilities require substantial upfront spending, and depreciation is part of the economics of producing chips there. The U.S. Commerce Department’s November 2024 award announcement describes that initial plan.
- Labor and staffing: Specialized fab work requires engineers, technicians, maintenance teams and experienced operators. Building a local workforce while transferring expertise to a new site adds costs; U.S. labor and construction costs are also part of the reported price difference.
- Supplier ecosystem: Taiwan has dense networks of equipment, materials, chemical, clean-room, packaging and testing suppliers, alongside a large pool of experienced workers. Arizona’s cluster is being built, so sourcing and support can be less local or less economical in its earlier stages.
- Scale and utilization: Fabs have large fixed costs. When a new facility is ramping up and producing fewer wafers than a mature, highly utilized site, those costs are spread across less output.
- Process transfer and parallel capacity: Qualifying production at a second site takes engineering, training and time. Maintaining capacity in more than one geography also means duplicating some infrastructure rather than relying solely on the lowest-cost established base.
Comparable yields show technical progress, not equal costs
TSMC says its first Arizona fab entered high-volume production in the fourth quarter of 2024 using its N4 process. The company has also reported Arizona yields comparable with its Taiwan fabs. TSMC’s Arizona site overview describes the production milestone, while its first-quarter 2025 transcript discusses yield comparability.
Comparable yields are significant: they weaken the idea that Arizona’s cost premium is simply a result of poor process performance. But yield measures how many usable chips a process produces; it does not account for every cost of making them. Labor, depreciation, utilization, supplier economics, logistics and the expense of maintaining a second production footprint can differ even when yields are comparable. Technical parity is not economic parity.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
How a chip premium reaches AI infrastructure costs
An AI system’s cost is not just its accelerator. It may include the processor, high-bandwidth memory, advanced packaging, networking, server chassis, power delivery, storage, cooling and data-center facilities. Operators also spend on electricity, software, engineering and staffing. A higher manufacturing cost for one chip therefore does not translate automatically into an equally large increase in the cost of a complete system, much less into the same percentage increase in a cloud service or AI product.
The effect depends on the chip’s share of total cost and on who can absorb or pass through the difference. It is more consequential if the affected chip dominates a system’s bill of materials, if many units are deployed, or if customers are running lower-margin workloads. It can matter less where the chip is a smaller share of cost, the product carries strong margins, or reliable access to supply is worth more than the premium.
- Chip designers: AMD and other designers may face higher input costs on products made at Arizona facilities. The public estimate does not establish that all of AMD’s products will be made there or that the company will raise retail prices. Nvidia is also a major TSMC customer, but the available reporting does not establish a specific Nvidia Arizona product, allocation or price. The same caution applies to Apple, Qualcomm, Broadcom and other fabless designers.
- Cloud providers: Providers buying affected hardware could absorb costs, share them with customers through cloud pricing, or adjust their investment plans. Their response depends on hardware utilization, contracts, electricity costs, competition and demand; the manufacturing estimate alone does not determine GPU-hour prices.
- AI companies and users: More expensive infrastructure could raise the cost of training or serving models, but whether that changes the price of an AI service depends on the rest of the cost stack, provider margins and competition.
A useful way to frame the potential effect is: manufacturing premium × affected chip share × pass-through rate × deployment scale. AI deployment is most at risk where those factors combine to make a project’s total cost exceed the value it is expected to generate. There is no evidence in the estimate alone that this has happened broadly.
Why customers might pay more for U.S. production
Producing advanced chips in Arizona can diversify wafer manufacturing away from a supply chain heavily concentrated in Taiwan. That is valuable if a disruption would cause costly shortages or delay critical systems. For a customer facing high downtime costs, a government buyer or a business that needs assured domestic supply, paying more may be preferable to depending entirely on a single region.
But Arizona does not make the whole AI semiconductor supply chain domestic. Production still relies on TSMC’s process expertise, global equipment suppliers and materials, as well as packaging, memory, transport, electricity and water infrastructure. A wafer made in the United States may still depend on other regions for steps needed to turn it into a finished AI accelerator.
TSMC’s plan to expand its U.S. investment to $165 billion includes six wafer fabs, two advanced-packaging facilities and an R&D center, according to the company’s announcement. That scope suggests a longer-term effort to build a cluster rather than a standalone fab, but it is a plan, not completed capacity. TSMC’s March 2025 announcement sets out the investment plan; its Arizona overview describes the wider project.
Rank #4
What CHIPS funding changes—and what it does not
The Commerce Department’s November 2024 final award included up to $6.6 billion in direct CHIPS funding and up to $5 billion in proposed loans for TSMC Arizona. The department also said the project could benefit from the Treasury Department’s semiconductor investment tax credit, potentially worth up to 25% of qualifying capital expenditures under the terms it described. The grant, proposed loans and tax credit are distinct forms of support, and the award is tied to project conditions and milestones. Commerce’s announcement provides the award details.
Public support can reduce the cost of building U.S. capacity, but that does not mean customers receive an equivalent discount on each chip. The available disclosures do not establish exactly how the incentives factor into Su’s 5%-to-under-20% estimate. The policy choice is broader than making a particular chip cheaper: public funding can be understood as paying for domestic capacity, jobs and reduced geographic concentration even if production remains more expensive than in Taiwan.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesCould the premium narrow—or persist?
The estimate is best treated as a current sourcing premium whose long-term trajectory is uncertain. More utilization, a larger trained workforce, deeper local suppliers, process learning and domestic packaging could reduce per-unit costs over time. Scaling across multiple Arizona facilities may also spread fixed costs more efficiently.
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
Some differences may be harder to eliminate. U.S. labor, construction and compliance costs may remain higher; Taiwan’s supplier density and experience are difficult to replicate quickly; and a geographically diversified company must maintain capacity in more than one place. Government incentives may help the economics of building a cluster, but they do not guarantee that the cost of making a chip will converge with Taiwan’s.
For readers judging whether the cost gap is shrinking, the useful indicators are fab utilization, production and yield disclosures, local supplier and workforce development, and progress on advanced packaging. TSMC has also announced plans for advanced U.S. process capacity: its second Arizona fab is associated with 3nm and 2nm production plans, with timing and the exact node mix subject to customer demand. Commerce’s April 2024 preliminary terms announcement describes the expansion plans. The third Arizona fab broke ground in April 2025, according to Commerce’s account of the groundbreaking.
When the premium is easiest—or hardest—to justify
| Situation | Why paying more may or may not make sense |
|---|---|
| Critical infrastructure, government or security-sensitive systems | Domestic access and reduced exposure to a regional disruption may justify higher manufacturing costs. |
| High-margin accelerators with constrained supply | Strong demand and pricing power may make the added input cost easier to absorb or pass through. |
| Price-sensitive consumer products or low-margin inference | Customers may have less room to absorb higher costs, particularly where alternatives exist and supply is ample. |
| Products with ample alternative capacity | The value of domestic production may not outweigh its premium if supply continuity is already adequate. |
The broader trade-off is not simply between a cheap chip and an expensive one. Taiwan-made production offers a mature, dense ecosystem and lower current cost, while Arizona production offers U.S. capacity and geographic diversification at a reported premium. Other foundries may provide alternatives, but product qualification, process performance, capacity and design compatibility must be assessed case by case; they are not automatic substitutes for a particular TSMC product.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →What the estimate means for AI’s growth
A higher Arizona manufacturing cost is a real economic consideration, especially if it remains structural as AI hardware deployment grows. It could squeeze margins, affect cloud prices or make some marginal projects less attractive. But Su’s estimate does not show that AI growth has slowed, nor does it establish a 20% increase in AI hardware or service prices.
The key question is whether the additional cost, after accounting for the affected chip’s share of a system and any pass-through, changes investment decisions. If demand and expected returns remain strong, customers may pay for capacity and resilience. If AI returns are weak or workloads compete mainly on low cost, the premium could push some deployments below their economic threshold. The trade-off is between the lowest current manufacturing cost and the value of a less geographically concentrated supply chain.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
