By Wan Ge in Tokyo

Introduction | “The AI investment boom is driving an expansion of Asian manufacturing, with semiconductor demand becoming a critical buffer for regional economies.” This has become a widely shared assessment of Asia’s artificial intelligence industry. Based on that premise, this series of reports seeks to examine the issue in depth and reveal the industrial system behind the AI boom.

Much of the current media coverage of AI in China focuses on algorithmic breakthroughs, the race for computing power and geopolitical competition, while paying far less attention to the people and factories behind the technology. This series attempts to fill that gap by bringing the AI boom down from “the cloud” and back onto the production line.

AI Is Not Replacing Workers—It Needs More of Them

When an application on your phone answers a question, have you ever imagined that only a few lines of code are moving through the cloud?

The reality is very different. Tens of thousands of servers are operating around the clock inside facilities covering tens of thousands of square metres. Production-line workers are working three shifts to turn sand into wafers, wafers into chips and chips into servers. Outside the data centres, substations and cooling towers are supplying the electricity and temperature control required to keep the entire system running.

AI’s “intelligence” may reside in the cloud, but its physical body is built on production lines. What appears to be an algorithm is, in reality, also a vast manufacturing industry.

The following is a supply-chain inventory stretching from sand to servers. It reveals just how many factories are working behind every AI service.

Penang, Malaysia: Three Shifts at the Packaging Plant

Inside the cleanroom of a semiconductor packaging and testing plant in Penang, workers rotate through three shifts while the equipment operates 24 hours a day. The packaging of high-bandwidth memory, or HBM, proceeds minute by minute. A delay in the delivery of a single chip could affect the launch schedule of an entire AI data centre.

Malaysia accounts for approximately one-eighth of the global semiconductor supply chain, with a large proportion of the country’s semiconductor companies concentrated in Penang. Long known as the “Silicon Valley of the East,” the state is now entering what local officials describe as the “Silicon Valley of the East 2.0” stage, gradually moving from contract manufacturing towards chip design. Penang currently has 45 integrated-circuit design companies in operation.

The strength of demand is reflected in production schedules. According to the Malaysia Semiconductor Industry Association, strong AI-related demand has already filled available capacity for certain memory products, with new orders scheduled as far ahead as 2028 or even 2029. The order book of local semiconductor equipment manufacturer UWC increased from less than 150 million Malaysian ringgit in 2024 to 180 million ringgit in 2025 and is expected to reach 250 million ringgit by mid-2026. The company has begun subcontracting part of its manufacturing work to local suppliers and extending its supply chain into neighbouring countries.

Global semiconductor companies are also accelerating their investment in the region. In December 2025, Intel committed an additional 860 million ringgit, or approximately US$208 million, to expand its packaging and testing operations in Penang and Kulim, with the aim of making Malaysia a core global centre for Intel’s assembly and testing business.

In January 2026, low-power AI solutions provider Syntiant opened a new manufacturing and research-and-development campus in Penang. Covering approximately 220,000 square feet, the facility doubled the company’s annual production capacity to around 1.6 billion units and created 800 high-technology jobs. In February, Taiwan-based display-driver integrated-circuit packaging and testing company Chipbond Technology opened a new plant in Penang following an investment of approximately US$200 million.

In June, Singapore-based Galatek Technologies established its first manufacturing, assembly and delivery centre in Penang, with plans to increase its local investment to US$100 million over five years. Chinese semiconductor packaging and testing company Forehope Electronic has also announced plans to invest up to RMB2.1 billion in a Malaysian production base.

At night, Penang remains illuminated by the lights of cleanrooms that never close. Every chip leaving these plants will eventually become part of the computing core of an AI server somewhere in the world.

Giheung, South Korea: An Arms Race in Wafer Fabrication

At Samsung Electronics’ Giheung campus in Gyeonggi Province, plans are under way for a new DRAM wafer fabrication plant. In the existing production facilities nearby, lithography machines etch circuits with nanometre-level precision. Each wafer carries the foundations of tens of thousands of AI chips—the physical “brains” of the computing industry.

On July 15, 2026, The Korea Economic Daily, citing industry sources, reported that Samsung Electronics was planning to construct a new DRAM plant at the Giheung campus. The facility is expected to have a monthly production capacity of approximately 100,000 wafers, with investment estimated at tens of trillions of won. Construction could begin as early as the third quarter of 2026.

The site had originally been designated for a research and development centre. Samsung later revised the plan and allocated the land to a large-scale DRAM production facility in response to growing demand for memory products generated by investment in AI computing infrastructure.

Giheung represents only one part of Samsung’s much larger semiconductor expansion. The company plans to build six wafer fabrication plants in the Yongin semiconductor cluster, which is intended to become a core manufacturing base for next-generation advanced processes and HBM memory.

On July 12, Samsung announced that production at the first Yongin plant would begin in 2029, one to two years earlier than originally planned. The company is investing a combined KRW2,030 trillion in its Pyeongtaek and Yongin clusters, while also planning two new wafer plants in Gwangju with an additional investment of KRW400 trillion.

Samsung Foundry’s medium- and long-term order backlog has approached KRW50 trillion.

SK hynix is also expanding aggressively. In July 2026, the company disclosed plans to invest KRW100 trillion in additional production capacity in Cheongju, North Chungcheong Province.

In a written response issued on July 13, the Bank of Korea stated: “The semiconductor market is continuing to expand, primarily because rapid investment in AI infrastructure is driving sustained growth in semiconductor demand, while the pace of supply expansion continues to lag behind demand.”

At night, the lights inside South Korea’s semiconductor cleanrooms never go out. Samsung’s KRW50 trillion order backlog is waiting to move through those facilities. Every wafer emerging from the production line represents another stake in the global race for AI computing power.

Chengdu: Accelerating the AI Server Assembly Line

Inside an AI server assembly plant in Chengdu, robotic arms and human workers operate side by side. GPU servers take shape rapidly along the production line before being packaged, loaded onto trucks and shipped overnight to intelligent-computing centres across China.

Driven by the rapid expansion of the computing-power industry and rising demand for AI servers, a company in Chengdu’s Pidu District recorded a sharp increase in orders during the first quarter, keeping its production lines in continuous operation.

Chengdu pioneered China’s “computing-power voucher” programme and has already issued vouchers worth more than RMB20 million. Beginning in 2026, the city plans to distribute at least RMB100 million in computing-power vouchers every year, covering as much as 60% of eligible computing expenses.

Huaxia Kunpeng won the contract to operate Huawei’s Western Fulfilment Centre in 2022. It has since established ten final-assembly and testing lines producing Huawei servers, storage systems, network switches and other products.Huakun Zhenyu has also opened its supply chain to local companies and reached preliminary cooperation agreements with more than ten Chengdu-based businesses in areas including storage products, integrated circuits and high-speed cables.

At night, workers on Chengdu’s server assembly lines are racing against time. Every delayed server delivery could extend the training cycle of an AI model.

Commentary

The Battle for the AI Industry Is in Full Swing

The semiconductor packaging cleanrooms of Penang, the wafer fabrication plants of Giheung and the server assembly lines of Chengdu span three time zones and involve three different categories of work. Yet they all follow the same rhythm: accelerate production, expand capacity and accelerate again.A detailed semiconductor industry report published by Nomura Securities in early July 2026 warned that the chip cycle remained far from its peak and that the second half of 2026 could bring an “epic” mismatch between supply and demand.

TSMC’s monthly CoWoS production capacity is expected to increase from approximately 70,000 wafers at the end of 2025 to 120,000 by the end of 2026. Even with that expansion, the capacity shortfall is expected to remain as high as 400,000 wafers, or more than 20%. Delivery schedules for HBM products have commonly been extended into 2027.

AI’s “intelligence” may reside in the cloud, but its body is built on production lines. At 10 p.m., when most people are already asleep, workers and machines across several Asian production centres are still operating at full speed.This is not a temporary period of overtime at a handful of factories. It is a multinational industrial campaign that could continue for years. In the AI era, the production line is the most tangible battlefield.

Next in the Series

In the next report, “At 10 p.m., the AI Chip Production Lines Are Still Racing Against Time,” we will enter semiconductor packaging plants in Penang, wafer fabrication facilities in Giheung and server assembly lines in Chengdu to examine why the lights on Asia’s AI production lines remain on deep into the night.