Semiconductor Capacity Is a Story of Bottlenecks, Not Just Fab Announcements

Every few months a government announces a semiconductor package, a company breaks ground on a fab, and the headline number is billions of dollars. The harder story is underneath: what actually limits how many chips the world can make, and how fast. Three supply-chain details published over the past two years suggest the answer is not just the number of fabs.

Take the industry's own history. IEEE Spectrum's December 2024 feature, "The Forgotten Story of How IBM Invented the Automated Fab," describes Project SWIFT, a manufacturing research program led by IBM manager Bill Harding at East Fishkill, New York, in 1970. Harding's goal was a fully automated wafer-fabrication line that could produce integrated circuits in less than one day, at a time when ICs such as RAM chips typically moved through dozens of manual work stations over about a month.

SWIFT averaged 5 hours to complete each layer of its fabrication process, while the fastest modern fabs take 19 hours per processing layer, and the industry average is 36 hours.

That comparison comes directly from IEEE Spectrum, which notes today's chips are built with more layers on larger wafers and with more complex processing. Those factors do not close the gap, the article argues. Many of SWIFT's innovations are now standard in automated plants, but its turnaround time has never been matched. The lesson for capacity planners is uncomfortable: throughput per square metre of cleanroom is a design problem, not simply a capital problem.

Where the raw material comes from

Then there is Spruce Pine, North Carolina. Tom's Hardware reported on 1 October 2024 that Hurricane Helene brought more than two feet of rain to the town, cutting roads and power and leaving the status of the local high-purity quartz mines unclear. Two companies mine ultra-pure quartz there: Sibelco North America and The Quartz Corp. Tom's Hardware said a Quartz Corp spokesperson told The Verge it was too early to say whether production had been affected, and that Sibelco had purportedly confirmed a suspension of operations. Repeated attempts to reach Sibelco failed because of the lack of phone service in the area.

Why does one small town matter? Quartz crucibles hold molten silicon during the Czochralski process that turns polysilicon into monocrystalline ingots, which are then sliced into wafers. Impurities in the crucible migrate into the melt, and semiconductors are extremely sensitive to contamination. Brian Potter's Construction Physics analysis, published on 8 November 2024, puts the trade in context. Spruce Pine is estimated to produce 180,000 to 200,000 tons of high-purity quartz a year, and high-purity quartz costs on the order of $10,000 a ton. Crucibles cost several thousand dollars each and last around 400 hours of operation. A single CZ furnace consumes 20 to 25 crucibles a year, and one paper estimated crucibles account for 30% of the manufacturing cost of monocrystalline silicon ingots.

Potter is careful about the apocalyptic version of the story. Cutting off Spruce Pine would probably not stop semiconductor production completely, he writes, but it would mean lower yields and higher costs. Alternatives exist, but they are not yet developed, not quite as good, or not as cheap. He also notes the high-purity quartz industry is extremely secretive and that specific numbers are hard to come by, which is worth remembering the next time a single mine is described as the foundation of the modern economy.

Adding capacity in different places

Meanwhile, capacity is being added in ways that do not look like a leading-edge logic fab. Texas Instruments said on 24 October 2024 that it had started producing gallium nitride power semiconductors at its factory in Aizu, Japan. Combined with existing GaN manufacturing in Dallas, Texas, TI said it would internally manufacture four times more GaN-based power semiconductors as Aizu ramps to production. The company said it had qualified 200mm GaN technology for mass production.

TI's release quotes Mohammad Yunus, senior vice president of technology and manufacturing, saying the milestone supports a plan to grow internal manufacturing to more than 95% by 2030 while sourcing from multiple TI locations. TI also said it had piloted GaN manufacturing processes on 300mm wafers and that its expanded processes are fully transferable to 300mm. GaN is an alternative to silicon for power conversion, used in laptop and phone adapters, motors for heating and air conditioning, and server power supplies. It does not compete with leading-edge logic, but it is part of the broader semiconductor capacity picture.

The most explicit government push comes from India. TechCrunch reported on 15 July 2026 that New Delhi unveiled a Mobile Phone Manufacturing Scheme worth ₹625 billion, about $6.5 billion, running five years, with incentives of 2.25% to 5% of eligible sales and an extra 1.5% for sourcing key components and sub-assemblies locally. The government also committed a further ₹1.28 trillion, around $13.3 billion, to domestic semiconductor manufacturing, expanding a $10 billion chip incentive programme launched in 2021, with greater support for chip equipment, materials, design and research.

The scale of the gap is in the same article. China accounted for 63% of global smartphone production in 2025, compared with India's 18%, according to Counterpoint Research. Apple began assembling iPhones in India in 2017, and about 25% of its iPhones are now made there, TechCrunch reports. The smartphone scheme runs through March 2031. The government expects mobile-phone production over that period to total about ₹39 trillion, roughly $405 billion, and the scheme to create about 60,000 direct jobs. Navkendar Singh of IDC tells TechCrunch the programme marks a shift from "assemble more" toward depth, R&D and local value capture, and that India has excelled at final assembly while remaining reliant on imported components.

Malaysia offers a third model. CNBC reported on 18 April 2024 that second finance minister Amir Hamzah Azizan said the country is not fazed by competition from China or the United States as it moves into higher-end semiconductor manufacturing. "The semiconductor space is now in the upcycle, and Malaysia will be a beneficiary of that," he told CNBC's Karen Tso at the IMF spring meetings in Washington. Malaysia holds 13% of the global market for chip packaging, assembly and testing services, according to a 18 February report from the Malaysian Investment Development Authority cited by CNBC. The minister said Malaysia would not compete head-on at the high end, and framed the opportunity as supply-chain diversification by end users.

What capacity really means

Put these together and the capacity question looks less like a single race and more like a set of linked constraints. Leading-edge fab construction gets the announcements. But wafer starts also depend on crucible supply from a handful of mines, on how many hours each process layer takes, on whether assembly and test capacity exists in places like Malaysia, and on whether countries such as India can move beyond final assembly into components and design.

None of this makes the billion-dollar fab numbers wrong. It makes them incomplete. The useful question is not how much money is being committed, but which of these bottlenecks the money actually loosens.