Asia’s smaller chip companies are joining their bigger peers in hiking prices as robust artificial intelligence demand fuels record levels of capital spending in the region, Nikkei Asia has learned.
Leading chipmakers and chip packaging and testing service providers from South Korea, Taiwan, Japan, and China have committed to spending a total of more than USD 136 billion for 2026, up more than 25% from a year ago, according to Nikkei Asia‘s analysis of leading chip suppliers’ plans and estimates by market research company TrendForce.
Equally notable is the wide range of chip and component companies raising prices, something many smaller players have not done since a sudden correction in the chip market in late 2022.
Vanguard International Semiconductor Corporation, a contract chipmaker of less advanced chips affiliated with Taiwan Semiconductor Manufacturing Company (TSMC), recently told some clients it will raise prices about 5% in the first quarter of 2026 with further hikes of 10–15% from the second quarter thanks to robust demand for power-related chips used in AI, sources with direct knowledge of the matter told Nikkei Asia.
Such hikes indicate that the AI infrastructure boom is now benefiting not only chip titans making cutting-edge processors but also suppliers of peripheral and supporting chips.
“We find the demand for power-related chips [for AI data centers] is particularly strong, and they remain in short supply,” Vanguard chairman Leuh Fang said. “As we increase investment to expand capacity and build new plants, our overall costs will inevitably rise. In that context, we will work closely with our customers to ensure our pricing reflects the value of our services, while striving to deliver a win-win outcome that meets their needs.”
Vanguard said it will maintain last year’s record level capital expenditure of up to TWD 70 billion (USD 2.22 billion) this year, but declined to comment on specifics related to price increases.
Unimicron, the world’s biggest chip substrate supplier by capacity, is also upping its prices along with its capital spending. It has raised prices a few times for substrates since the final quarter of 2025 and will launch a bigger hike this quarter.
“We have to reflect the cost increases over materials and metal prices and since the demand for AI remains so strong, we believe there is still room for us to further reflect the rising costs,” Unimicron spokesperson Vincent Chung said during the company’s recent earnings conference.
The company has raised its spending plans from the initial TWD 25.4 billion (USD 805.5 million) to a record TWD 34 billion (USD 1.1 billion) for 2026 to expand capacity for high-end chip substrates used in AI chips. Nikkei Asia first reported that supply constraints tied to high-end glass cloth have led to shortages of chip substrates used in electronic devices, affecting even Apple and Qualcomm.
It is a similar story with specialty memory chipmaker Winbond Electronics, which has committed to spending TWD 42.1 billion (USD 1.3 billion) in 2026, almost eight times more than last year. The Taiwanese memory chipmaker, which specializes in niche NOR flash chips used primarily to store source code in a wide range of electronic devices as well as legacy customized DRAM chips, said its average price increase will exceed 30% in the first quarter of 2026.
“Our spending this year will be the highest ever and we are confident that it’s for growth for 2026 and 2027,” Winbond President Chen Pei-Ming told Nikkei Asia. “To ensure a smooth capacity expansion, I personally visited all key equipment vendors to secure their support in shortening production tools’ delivery lead times.”
Whether and to what extent the recent US attack on Iran impacts the global economy and tech supply chains remains to be seen. For now, however, the world’s biggest chipmakers are all eyeing a bumper year.
China’s top chipmaker Semiconductor Manufacturing International Corporation (SMIC) has told its customers that prices at some of its plants focusing on mature chips will increase.
“SMIC is taking a flexible approach to raising prices. We’ve heard from the company that plants with tighter utilization rates will increase prices,” an executive at a chip developer told Nikkei Asia.
For 2026, the company has again committed to record spending almost equal to annual revenue as it continues to boost production, mainly for domestic needs.
The world’s top contract chipmaker TSMC, meanwhile, said it will spend USD 52–56 billion this year, while Samsung Electronics and SK Hynix of South Korea are also raising capex for 2026.
Leading chip packaging and testing service providers, along with smaller chip makers, are benefiting from “spillover effects” as AI demand increases the need for supporting components and services.
Chip packaging and testing house ASE Technology Holding and chip testing service provider King Yuan Electronics Corporation (KYEC) are both planning record spending for 2026.
Boris Hsieh, president of chip packaging and testing service provider Powertech Technology, which servers Micron, AMD, MediaTek, Kioxia, and others, told Nikkei Asia that in the current environment, having production-ready capacity is extremely valuable.
“We are seeing some packaging houses fully booked with orders from AI chip giants such as Nvidia, forcing them to turn away many mid-sized and smaller clients,” Hsieh said. “The spillover effect is significant. Right now, any provider with available capacity, especially high-quality capacity, is extremely valuable.” Since last year, Powertech has already purchased several existing cleanroom facilities in Taiwan to expand capacity.
“My team has already begun discussing whether we should once again look for additional facilities that could be quickly converted into chip packaging or testing lines,” Hsieh said.
Other spillover effects are being seen in memory and power-related components.
Building AI servers and the related computing ecosystem requires not only the best high-bandwidth memories (HBM), but also a massive amount of NAND and NOR flash memory, power supply components and power regulators capable of handling the massive and sudden change of voltage from grid to chips.
“It’s difficult to quantify exactly how much additional memory is needed specifically for AI server deployments. However, we estimate that a single Nvidia GB200 server rack requires around 120 NOR flash chips, roughly equivalent to the amount used in some 120 personal computers. That gives you a sense of the scale and the enormous appetite for memory driven by AI,” Winbond’s Chen said.
An executive with an optical module supplier serving Nvidia and Amazon Web Services said supply crunch concerns are growing. “Not just the large chips but also some small chips related to AI are in short supply, and we are very worried about it, as it will affect the smoothness of our shipment to end customers,” the executive said. “Those small, overlooked chips and components not only are in constrained supply, but also much more expensive now.”
Nanya Technology, the world’s fifth-ranked DRAM memory chipmaker, said it would more than double its capital spending for 2026, after nearly three years of a severe market downturn.
President Lee Pei-Ing said there are supply gaps in almost every type of DRAM and in all types of markets.
“Some more urgent ones are from consumer electronics and automotive solutions,” Lee said. “We see overall market demand as strong across every region worldwide, particularly from the US and China, as well as in Japan and South Korea.”
Joseph Tung, CFO of ASE, said that demand currently far exceeds supply, creating a favorable pricing environment. He added that the company’s leading-edge chip packaging business is expected to double in 2026, while its mainstream packaging segment, some of which supports chips used in building out the AI ecosystem, is also experiencing solid growth.
While many component makers in the AI space are racing to expand capacity, they also warn there is still some uncertainty.
Ping Cheng, chairman and CEO of Delta Electronics, the world’s biggest maker of power supply solutions, said the company plans to spend more than TWD 46.1 billion (USD 1.5 billion) in 2026 capex, its highest level ever, and is scouting locations to build new plants.
Cheng said Delta always builds facilities two to three years in advance to prepare for visible future demand. “But in the past two years, the demand for AI came so sudden and ferocious that it occupied many of our prepared new capacities ahead of our plans. We have to look for locations now to build more plants for the 2028 demand and beyond,” he said last week at an investors conference.
“The growing momentum for data center buildup is likely to continue this year, but the market is also moving very dynamically,” he added. “The AI data center buildup could still face many challenges, and there’s a shortage of construction crews as well as issues surrounding water and power supplies in many places. Those can all be issues that hinder the speed of massive construction.”
This article first appeared on Nikkei Asia. It has been republished here as part of 36Kr’s ongoing partnership with Nikkei.