NXP and Vanguard Open S$8.5 Billion Singapore Fab as Specialty Chip Demand Outruns Supply

Introduction

A wafer fabrication plant is one of the least glamorous and most consequential buildings in the global economy. No consumer will ever see one, photograph one, or name one. Yet every smartphone, every vehicle, every industrial robot and every data centre depends on chips that came out of facilities precisely like this one — vast, hermetically clean buildings where 300-millimetre circles of silicon are etched, doped, deposited upon and diced, tens of thousands of times, with a consistency that borders on the absurd.

On Monday, 28 September 2026, VisionPower Semiconductor Manufacturing Company (VSMC) opened the doors to exactly such a building. The plant sits in Tampines, in Singapore's eastern industrial heartland, and it is the first fab of its kind for a joint venture between two companies that between them represent two of the most strategically important franchises in semiconductors: Taiwan's TSMC-backed Vanguard International Semiconductor (VIS, TPEx: 5347) and the Netherlands' NXP Semiconductors. The facility represents a capital commitment of S$8.5 billion, roughly US$6.7 billion, according to Singapore's Channel NewsAsia.

What makes VSMC interesting is not the size of the cheque. It is what the plant is not building. While much of the semiconductor world's attention — including a great deal of coverage in our semiconductors category — has been locked onto the leading edge, the two-nanometre and sub-two-nanometre logic nodes that power the latest generation of artificial intelligence accelerators, VSMC will run 130-nanometre down to 40-nanometre processes. These are mature-node technologies, and they are in violent shortage.

The company is unambiguous about the reason. VSMC and VIS chairman Fang Leuh (方略) said on Tuesday that demand for specialty node chips "has gone through the roof" as AI buildout has squeezed non-AI manufacturing capacity, and that the resulting chip supply crunch "is expected to persist for a long time." The most revealing remark in the coverage, however, was an expression of regret rather than triumph. "We really should have just built both fabs at the same time," Fang said, according to the Taipei Times.

That single sentence captures the central paradox of the 2026 semiconductor cycle. Leading-edge capacity is not scarce — it is being built at a pace that has surprised even its own architects. The scarce capacity is everywhere else.

Main Content

What VSMC actually opened

The technical specification of the plant is worth setting out plainly, because the gap between these numbers and the leading edge is the whole story. VSMC's Tampines facility will fabricate 300-millimetre wafers — the industry standard "big wafer" format — and will do so using process technologies ranging from 130 nanometre down to 40 nanometre. Those nodes serve mixed-signal, power management, analog and interposer applications. An interposer, the company explained, is a critical component used in advanced chip-on-wafer-on-substrate packaging — the connective tissue that allows different dies to be bonded together into a single higher-performance package.

The output figures are substantial in physical terms even if modest in lithographic terms. The fab is expected to produce about 44,000 12-inch wafers per month and to create roughly 1,600 jobs. The construction timeline itself is a datapoint: VSMC took 22 months to build the facility, according to CNA. Risk production is likely to begin in the next quarter, with mass production scheduled for the first quarter of 2027, and full capacity utilisation is projected ahead of the original 2029 schedule.

The market mix is the part that should interest anyone following the Internet of Things and robotics sectors. CNA reported that the chips will serve high-performance computing, mobile, automotive, industrial and consumer markets. Read together, that list describes the sensors, motor drivers, power regulators, analog front-ends and interface silicon that the electrification of vehicles, the growth of connected devices and industrial automation all depend on — none of which needs a nanometre node, and all of which needs a great deal of 40nm and above capacity.

Singapore's Minister for Trade and Industry, Dr Tan See Leng, made precisely this argument at the opening ceremony. While much of the attention today is on the most advanced logic chips that power artificial intelligence, he said, specialty semiconductors are just as essential. Citing industry forecasts, he noted the semiconductor industry is set to double to US$1.6 trillion by 2030, and that demand for specialty chips is growing with trends like electrification, connected devices and industrial automation. Of the 1,600 new jobs, 75 per cent will be professional, managerial, executive and technician roles — engineers, chemists and data scientists who operate advanced manufacturing equipment, develop and fabricate wafers, and optimise the manufacturing process.

The decision to fast-track a second fab

The more consequential news broke a day after the ribbon-cutting. VSMC said it is fast-tracking the timeline for its second Singapore fab. Fang said the Singapore fab has the potential to break even earlier than anticipated, with full capacity utilisation projected ahead of the 2029 schedule, and that progress at the facility has exceeded expectations. Phase 1 capacity is already entirely allocated, with additional demand to be accommodated in a phase 2 expansion.

The company was careful not to over-commit. Asked whether the expansion plan would be finalised by the end of the year, Fang said VSMC would not rush a decision. To support such a capital-intensive 12-inch fab project, the company is carefully structuring its financial strategy and evaluating future product applications and technology portfolios, including 28-nanometre or 22-nanometre technologies. Notably, the company is evaluating whether to develop those nodes internally or license them — Fang said licensing provides a faster route to scale production. More than 10 customers have already visited the Singapore site to discuss future demand and capacity requirements.

A move from 40nm up toward 22nm is a meaningful upgrade in capability, and the fact that a mature-node specialist is contemplating it speaks to how much the mid-range is being pulled upward by AI-adjacent demand. Interposers and advanced packaging substrates sit squarely in that middle, and they are the connective tissue of the chiplet era. When NXP and Vanguard built the first fab, they positioned it for analog and power. The second one appears to be aimed somewhat higher.

On the AI-slowdown debate

Fang was also asked directly about the recurring debate over whether AI investment is slowing. His answer was flatly dismissive of the premise. AI's trajectory is "unstoppable and irreversible," he said, and a future slowdown in growth would not leave foundry capacity idle.

This is a common refrain across the industry in 2026, and it deserves more scrutiny than it usually receives. The bull case — dominant across our recent coverage, including analysis of how TSMC's Arizona expansion and Asian foundries are racing to convert announced capacity into working production — assumes that the specific artefact driving memory and advanced packaging demand stays in fashion for long enough to absorb the capital now being committed.

Fang's own remarks, read carefully, support a narrower and more defensible claim than his rhetoric suggests. Even in a scenario where AI growth decelerated sharply, his argument is not that AI demand would persist at current levels. It is that the non-AI capacity already being built would not sit idle, because electrification, connected devices and industrial automation demand are independent of the AI cycle. That is the same logic that makes specialty-node capacity defensive. It is also a reminder that the fab under discussion is not, strictly speaking, an AI bet at all — it is the other bet, and the one that has been chronically under-invested.

Why the mature nodes are the tight ones

It is tempting to assume that chip supply tightness is a memory story, because the headlines of 2026 have largely been about HBM, DRAM pricing and long-term supply agreements. That is real, but it is a narrower phenomenon than the one VSMC is responding to.

Mature-node capacity is scarce for structural reasons that have nothing to do with AI enthusiasm. A 40nm fab is a large, expensive building full of equipment that is, by leading-edge standards, technologically obsolete. It is economically rational to run such a plant at high utilisation indefinitely, because the capital is sunk and the marginal cost of an additional wafer is low. That makes mature capacity sticky at existing fabs but almost impossible to create quickly at new ones — the 22 months VSMC took to build is typical, and a leading-edge fab takes considerably longer. Meanwhile the demand side has changed character: every vehicle gaining an electronic control unit, every grid inverter, every industrial sensor is buying more analog and power silicon than the vehicles, inverters and sensors of a decade ago.

There is a second, subtler effect. When a leading-edge fab's capacity is consumed by AI accelerator production, the engineering and packaging resources clustered around those programmes — and, critically, the specialist equipment supply chain that serves them — become a bottleneck for everyone else. The wafer starts life on a 40nm line and ends up inside a package competing for the same advanced-packaging capacity as a GPU. Fang's observation that AI buildout "squeezes non-AI manufacturing capacity" describes precisely this.

The result is a market in which a Taiwan-based foundry operator and a Dutch automotive and industrial chipmaker can jointly build US$6.7 billion of 40nm-and-above capacity in Singapore, sell every wafer of phase 1 before the doors officially open, and describe a second fab as an obvious priority. GlobalFoundries, speaking separately in Hsinchu, has said much the same thing: the company plans to accelerate global capacity expansion to accommodate exponential growth in radio frequency and other chips driven by the AI boom, with Singapore playing a vital role.

The Singapore bet, and what it is worth

It is worth pausing on the location, because the choice says something about where the industry thinks durable capacity will be needed.

Singapore is small — a city-state of under six million people — and yet, per the Economic Development Board factsheet cited by CNA, the semiconductor industry contributes close to 6 per cent of the nation's economy and employs more than 35,000 people. The country accounts for about one in every 10 chips made globally and produces one-fifth of global semiconductor equipment annually. Tan See Leng framed the new fab as reinforcing "Singapore's role as a trusted node in the global supply chain," and a month into September 2026 the government had also unveiled a national entity, SG Semiconductor, intended to strengthen that role further.

The structural advantages are legible: political stability, strong rule of law, deep technical labour pool, an established equipment ecosystem, and a willingness to impose the kind of industrial discipline that keeps a small country competitive. The vulnerability is equally legible — cost, and a tiny domestic market, and the simple fact that a fab is a fixed asset with a fifteen-to-twenty-year economic life being designed against demand forecasts that have been volatile for two years.

NXP's chief executive framed the investment in personal terms. Rafael Sotomayor said the plant represents a long-term commitment to strengthening the global semiconductor ecosystem, enhancing supply chain resilience and creating the capacity needed to support the technologies that will define the future. He described the facility as a catalyst both for innovation and for developing the industry's next generation of engineering talent. Then he made the scale explicit: the investment in the VSMC fab is one of the most significant milestones in NXP's history, and it is the largest investment NXP has made since its acquisition of Freescale in 2015.

That is a striking figure for a company that was, until recently, a largely automotive and industrial supplier operating in a market most investors considered mature. NXP is making its largest bet in a decade on a category of chip that is not sexy, not visible and — on current evidence — not going away. Sotomayor's framing of the moment is telling in its own way: he said the world is entering an era where artificial intelligence is moving beyond the cloud and into the physical world, with physical AI applications to vehicles, robots and factories. "Semiconductors are at the heart of this transformation," he said, "providing the foundation for trusted physical AI."

The technical translation of that claim is a 40-nanometre mixed-signal part with a power management section and a sensor interface. It will not be photographed. It will, however, be in a very large number of things.

Conclusion

There is a useful discipline in looking at any major capital announcement in semiconductors and asking one question: what would have to be true for this to be a bad idea? For a leading-edge AI fab, the answer is that the AI capex cycle breaks. For VSMC, the answer is considerably harder to construct, because the demand it serves — vehicles, grid equipment, industrial automation, connected devices — is growing on its own logic rather than riding on anyone's model bet.

The risks are real and should not be waved away. Phase 2 has no firm decision date. The potential move into 28nm and 22nm could be licensed rather than internally developed, which is faster but places VSMC's technology roadmap in someone else's hands. Mature-node fabs have a long history of arriving just as a cycle turns, and a facility that takes 22 months to build and 10% of a decade to amortise carries genuine timing risk. Singapore's cost base is not the world's lowest, and the country has no captive domestic demand of any scale to cushion a downturn.

But the balance of evidence currently favours the build. Phase 1 is fully allocated before opening. More than ten customers have visited the site. The first fab is running ahead of schedule. A second is being fast-tracked. And the people who own both fabs are saying publicly that they wish they had started sooner — which is not something capital discipline usually produces.

The lesson generalises beyond Singapore. The most under-appreciated fact about the 2026 chip cycle is that the AI boom's most durable side effect may be the sudden visibility of a bottleneck nobody was looking at. Everyone has a thesis on leading-edge capacity. The 40nm lines that make the power management, the mixed-signal front-ends and the interposers are where the waiting lists actually are — and they are being filled, one US$6.7 billion building at a time.

Images

A semiconductor wafer being handled at a wet-processing station inside a fabrication facility. Illustrative photograph of wafer handling in a generic fab cleanroom; it is not the VSMC plant in Singapore. Photo: Wikimedia Commons.

A macro photograph of a decapped integrated circuit, showing the silicon die at the centre of the package with fine gold bond wires fanning out to the package leads. This is the view of a chip that, once produced, is sealed and invisible — the kind of part VSMC's 40nm and 130nm lines are built to make. Photo: Wikimedia Commons.

Decapped integrated circuit showing die and bond wires

NXP's facility in Kaohsiung, Taiwan, with the company's logo visible on the rooftop. Illustrative photograph of an NXP site in Taiwan; it is not the new Tampines fab in Singapore. Photo: Wikimedia Commons.