According to the latest report published by DIGITIMES Asia, global data center AI chip shipments are projected to grow from 30.5 million units in 2024 to 53.4 million units in 2030. This data center AI chip category includes high-end and mid-range GPUs, application-specific AI chips (such as Google's TPUs), AI server CPUs, and networking/interconnect-related chips (e.g., Switch ASICs/rack-scale-up Interconnect Chips/DPUs & NICs).
Geopolitical realignments are accelerating a reshaping of the global outsourced assembly and test (OSAT) landscape. Industry analysts say that over the next five years, advanced AI-chip packaging capacity will consolidate around two critical hubs: Taiwan and the United States. The key beneficiaries are expected to be ASE Holdings and Siliconware Precision Industries (SPIL) in Taiwan, and Amkor Technology in the US and South Korea.
SK Hynix has reportedly postponed the mass production timeline for its sixth-generation high bandwidth memory (HBM4) from the end of the second quarter of 2026 to the third quarter of 2026, coinciding with Nvidia Corporation's planned 2026 launch of its "Rubin" AI accelerator that will use HBM4. The delay appears to be a strategic adjustment influenced by ongoing market demand and supply negotiations.
These are the most-read DIGITIMES Asia stories in the week of December 1 to December 7, 2025.
Samsung Electronics has shifted its internal strategy in response to intensified HBM market competition**,** reallocating capacity toward DDR5 RDIMM modules and freeing up around 80,000 DRAM wafers monthly to yield stronger profits.
Ritek Group CEO Wang Ting-chang said the group has long invested in AI-linked fields such as power, semiconductor materials, and packaging. Although invisible at the consumer end, Ritek has become an "invisible champion", supplying the tooling, materials, and power backup systems that underpin customers' AI deployment.
China's storage industry is at a critical juncture. Surging AI workloads have fuelled a global shortage of high-bandwidth memory (HBM) and pushed storage to the forefront of the semiconductor market. Yet while overseas suppliers have filled HBM capacity through 2027, many Chinese vendors remain trapped in margin-draining price competition that limits innovation.
Japanese and Korean passive-component makers are diverting more capacity to high-end capacitors used in AI servers, tightening supply in mainstream electronics, and creating new openings for Taiwanese suppliers. The pivot reflects how booming AI server demand is reshaping global component production.
Tight memory supply has lifted prices across product lines, allowing Elite Semiconductor Microelectronics Technology (ESMT) to break a two-quarter losing streak in the third quarter of 2025. With wafer cycles running four to six months and contract pricing lagging by a quarter, the real pricing uplift is expected to appear from the fourth quarter of 2025. ESMT is on track to return to core profitability, with gross margin set to rise materially from third-quarter levels.
South Korea is racing to slash its more than 90% dependence on imported power semiconductors. Industry leaders warn that this reliance threatens the country's competitiveness in electric vehicles (EVs), data centers, and emerging mobility markets. Airport ground support equipment (GSE) has emerged as a strategic platform to validate and scale domestic power semiconductor technologies.
A newly released filing from China's National Intellectual Property Administration (CNIPA) shows Huawei submitted a patent in June 2022 outlining a method to achieve 2nm-class metal line patterning using only deep ultraviolet (DUV) lithography — potentially bypassing US export restrictions on advanced tools. The South China Morning Post reports the technique can create metal pitches under 21nm, a requirement typically met only with ASML's top-tier extreme ultraviolet (EUV) exposure systems, which remain off-limits to Huawei.
More coverage
