These are the most-read DIGITIMES Asia stories in the week of December 1 to December 7, 2025.
A TSMC pioneer questions the judgment behind Wei-jen Lo's move to Intel at 75
Former TSMC senior strategy executive Wei-jen Lo is facing a lawsuit from the chipmaker after reports that he took confidential sub-2nm documents before joining Intel at age 75, a move that has drawn scrutiny across the industry. Intel has expressed confidence in Lo, while Chin-tay Shih, a founding architect of TSMC and former ITRI head, questioned the judgment behind such a late-career shift and said the case raises issues that go beyond normal legal and ethical boundaries.
Shih said the facts will have to be clarified through litigation, including whether Lo's future work relates to TSMC technologies and how non-compete rules in Taiwan and the US apply. He added that talent mobility is inevitable as global demand for Taiwanese engineers rises, but warned that Taiwan's shrinking population and strained education pipeline will deepen workforce shortages unless the country steps up efforts to cultivate and attract skilled workers.
Broadcom CEO visits South Korea as Samsung HBM4 reportedly obtains Google TPU certification
Samsung Electronics reportedly passed quality tests for high-bandwidth memory used in Google's next-generation Tensor Processing Unit (TPU) and secured supply commitments through 2026, although the company later denied the report to DIGITIMES and declined to elaborate.
The Chosun Daily reported that Broadcom CEO Hock Tan visited South Korea this week to sign an HBM4 supply agreement with Samsung Device Solutions head Jun Young-Hyun, and that Broadcom sought volumes extending to 2028, but only 2026 quantities were finalized. Broadcom, which designs and manufactures Google's TPUs, is expected to use HBM3E for the seventh-generation chip and move to HBM4 in the 2026 model.
Samsung, SK Hynix set profit-first DRAM regime
Samsung Electronics and SK Hynix told global investors they will maintain tight control over DRAM output as shortages spread from AI servers to PCs and smartphones, signalling that the memory upcycle could extend beyond 2028. Samsung said at a Morgan Stanley event that it will prioritise profitability and limit capacity additions, a stance reflected in its inability to meet full DRAM order volumes and its refusal to sign multi-year mobile DRAM contracts as prices rise.
SK Hynix delivered a similar message at a Bank of America Merrill Lynch meeting, saying it will accelerate its transition to 1c DRAM while projecting that industry shortages will persist at least through the first half of 2027. Both companies are resisting long-term deals, keeping contract terms short to preserve pricing flexibility.
December memory prices surge as the real crunch lies ahead
Global memory supply has tightened so sharply that December contract prices for major DRAM and NAND products have jumped as much as 80 to 100%, TeamGroup General Manager Gerry Chen said, warning that a far more severe shortage will emerge in the first half of 2026 once inventory is cleared.
Chen said the imbalance is being driven by structural supply limits and rising demand from AI and data center servers, which are already pushing up PC and notebook component costs and could cut mainstream shipments by as much as 40 percent next year. He added that cloud providers have begun securing DRAM and NAND supply for 2026 and 2027, signaling continued strength in server demand.
Jensen Huang visits Taiwan and refutes ODM capacity concerns
Nvidia CEO Jensen Huang's recent visit to Taiwan has fueled speculation about GB300 shipment issues, although supply chain companies said deliveries are on track and that any yield or logistics challenges remain manageable. ODMs are working to meet Nvidia's rising volume requirements as suppliers enter the company's fiscal fourth quarter, a period that typically demands rapid output growth.
Wistron and Foxconn both reported strong AI server momentum, with Wistron expanding its Hsinchu plant and Foxconn posting a 300 percent quarterly jump in third-quarter AI rack shipments. Quanta said customers began shifting to the GB300 in the third quarter and expects AI products to drive triple-digit revenue growth in 2026. Huang has said demand for the Blackwell GPU is exceptionally strong and that cloud GPU supply is fully allocated, with Nvidia securing about US$500 billion in AI chip orders through 2026.
Samsung reportedly wins majority of Nvidia's 2026 SOCAMM2 supply
Samsung Electronics is expected to supply more than half of Nvidia's SOCAMM2 memory modules for 2026, according to Hankyung and ICsmart, positioning the company as the largest contributor to Nvidia's next-generation AI server CPU ecosystem. Industry sources say Samsung surpassed Micron after achieving strong yields on its 1b DRAM and secured a 100 billion Gb allocation, which represents half of Nvidia's planned 200 billion Gb SOCAMM requirement for next year.
Samsung would need 30,000 to 40,000 wafers per month to meet the order, while SK Hynix is expected to take most of the remaining share, and Micron only a small portion. SOCAMM2, designed for Nvidia's upcoming Vera CPU, promises higher capacity and better efficiency and is set to play a central role in expanding CPU-side memory bandwidth in future AI servers. With tightening inventories and rising demand for LPDDR and HBM, Samsung is also ramping up 1c DRAM output, reinforcing its renewed lead in the AI memory race heading into 2026.
Apple's potential Intel partnership hinges on 18A-P process maturity by 2027
Apple is weighing a return to Intel for entry-level M-series chip production by 2027, a move that hinges on whether Intel can mature its 18A-P process within the next two years, according to reports from PC Gamer, TechSpot, and BusinessWorld. The plan would shift a portion of MacBook Air and iPad chip manufacturing to the US and reduce Apple's dependence on TSMC, although the company will only proceed if Intel meets its strict quality and cost standards.
Intel has reported steady progress on 18A yields, but its history of process delays and continued foundry losses adds uncertainty. If Intel fails to deliver, analysts say Apple may deepen its reliance on TSMC and Samsung, and US policymakers could redirect their domestic chip strategy toward foreign suppliers expanding production in Arizona and Texas.
Article edited by Jack Wu
