Below are the most-read DIGITIMES Asia stories from the week of June 8-14, 2026:
Samsung foundry profit rebound may come in 3Q26 as 2nm orders rise 130%
Samsung Electronics' foundry business is on track to return to profitability as early as the third quarter of 2026, ahead of previous expectations, driven by stronger demand for advanced manufacturing processes and high-bandwidth memory (HBM) components. Rising utilization of its 4nm production lines, fueled by growing HBM4 base-die orders for AI accelerators from customers such as Nvidia and AMD, is boosting revenue while reducing the impact of depreciation costs.
At the same time, Samsung has improved yields on its 2nm gate-all-around process and is preparing to ramp production for major customers, including Tesla, with additional opportunities emerging from companies seeking alternatives to TSMC's capacity-constrained advanced nodes.
The company's US$37 billion Taylor fab in Texas is expected to further improve cost efficiency as production scales, though accounting treatment and startup expenses remain variables. After several years of trillion-won operating losses, Samsung's foundry division appears to be benefiting from the AI-driven semiconductor boom, improving manufacturing performance, and diversifying its customer base.
Nvidia unveils AI PC vision as Intel offers reflection instead of product news
Computex 2026 highlighted the contrasting positions of Intel and Nvidia as the AI era reshapes the technology industry. While Intel CEO Lip-Bu Tan focused largely on his personal history with Taiwan and Intel's long-standing ties to the island's technology ecosystem, offering few major product or strategy announcements, Nvidia used the event to showcase its expanding AI ambitions.
The company unveiled RTX Spark, a new processor platform designed to run AI agents locally on PCs, signaling a push beyond data-center accelerators into AI-powered personal computing. However, analysts questioned whether the high expected price of RTX Spark-based laptops could limit adoption, particularly as many users remain comfortable relying on cloud-based AI services.
The launch also underscored Nvidia's growing partnerships with MediaTek and Intel, reflecting a strategy of leveraging external expertise in PC platforms while concentrating its own resources on higher-growth opportunities such as AI accelerators, autonomous vehicles, and robotics. Together, the developments illustrate how Nvidia is broadening its influence across the computing industry, while Intel continues to search for a clearer path to renewed growth.
TSMC hit by US patent suit, Taiwan ministry pledges support
TSMC has become the target of a patent infringement complaint before the US International Trade Commission filed by Ireland-based patent licensing firms Longitude Licensing and Marlin Semiconductor, which argue that alleged patent violations could justify restrictions on imports of TSMC-manufactured chips into the US.
The case has attracted attention after the firms enlisted support from several US lawmakers and raised the prospect of an import ban. Both complainants are non-practicing entities (NPEs), commonly known as patent-licensing firms that acquire intellectual property from distressed companies and seek licensing fees or settlements through litigation.
Such firms often use the US Tariff Act of 1930 to pursue ITC complaints against overseas semiconductor manufacturers and major technology companies, with previous targets including Apple, Qualcomm, and Amazon. In response, Taiwan's Ministry of Economic Affairs emphasized that TSMC and other Taiwanese semiconductor companies operate in compliance with local laws and intellectual-property regulations worldwide, pledging to monitor the case closely and provide support if necessary to safeguard the stability and competitiveness of Taiwan's semiconductor industry and supply chain.
Analysis: ASICs are coming for Nvidia's GPU dominance — and it could happen next year, says DIGITIMES analyst
Key announcements made by Nvidia CEO Jensen Huang at GTC Taipei and Computex 2026 may represent the company's roadmap from generative AI to agentic AI and ultimately physical AI. According to DIGITIMES analyst Joyce Chen, Nvidia's influence now permeates nearly every corner of Taiwan's technology ecosystem, but new bottlenecks are emerging in networking and data movement, particularly around co-packaged optics (CPO), creating opportunities for players such as Marvell Technology and Broadcom.
Chen also predicts that AI inference workloads will accelerate the adoption of application-specific integrated circuits (ASICs), with shipment volumes potentially surpassing GPUs as companies seek more power-efficient alternatives, benefiting firms including MediaTek and Google's TPU ecosystem. While Nvidia remains dominant in AI revenue, products such as RTX Spark may initially appeal only to niche high-performance users, and the Vera CPU is viewed primarily as a way to deepen customer dependence on Nvidia's GPU and CUDA ecosystem.
Chen further dismissed speculation that Nvidia is building a parallel AI supply chain in South Korea, arguing that Taiwan's manufacturing ecosystem remains unmatched, and highlighted how Europe's AI sovereignty ambitions, exemplified by French chip developer SiPearl, continue to rely heavily on Taiwan's semiconductor design expertise and supply chain.
Nvidia's AI ramp deepens memory squeeze as cloud providers lock up supply through 2028
Global memory shortages are expected to worsen through 2027 and into 2028 as demand from AI infrastructure continues to surge, driven largely by Nvidia's upcoming Vera Rubin platform and the mass adoption of HBM4 memory. Major cloud service providers have already locked up nearly all available long-term memory supply for 2027 and are now securing 2028 capacity, leaving OEMs, module makers, and other device manufacturers facing tighter availability of both DRAM and NAND.
The supply squeeze is being intensified by concurrent demand from AI servers, edge AI devices, and consumer product launches, while memory manufacturers remain confident enough in future pricing to negotiate long-term volume commitments without fixing prices far in advance. The shift of production capacity toward high-margin server and HBM products is also crowding out standard PC memory, reshaping the DRAM market, and pushing some AI server customers to adopt lower-capacity memory configurations to control costs.
With SK Hynix, Samsung, and Micron already ramping HBM4 production for Nvidia's next-generation systems, industry sources expect memory prices to remain elevated and shortages to become more severe than those experienced during the 2025-2026 supply crunch, creating ongoing cost pressures across the broader technology sector.
Cerebras outpaces Nvidia in video showdown at SuperAI Singapore, making its case against GPU dominance
At SuperAI Singapore 2026, Cerebras Systems used a dramatic on-stage demonstration of its wafer-scale AI processor to argue that traditional GPU architectures are increasingly inadequate for the next generation of AI workloads. Chief Strategy Officer Andy Hock showcased the company's Wafer Scale Engine, positioning it as a faster and more efficient alternative to GPU clusters for AI training and inference
Cerebras contends that the rise of reasoning models and agentic AI applications, which can require up to 1,000 times more compute than conventional AI queries, makes low-latency inference a critical competitive advantage. The company highlighted demonstrations showing its systems completing tasks significantly faster than leading GPU-based implementations and pointed to major partnerships with OpenAI and Amazon Web Services, including a disaggregated inference architecture that combines AWS Trainium chips with Cerebras hardware.
Fresh from a blockbuster IPO that valued the company at nearly US$100 billion, Cerebras also underscored the importance of its long-standing manufacturing partnership with TSMC, which enabled the creation of what it describes as the world's largest and fastest AI processor, highlighting a growing push within the industry toward specialized AI silicon designed for inference performance rather than general-purpose computing.
J&V Energy moves into AI data center power infrastructure to tap Taiwan's AI electricity boom
J&V Energy Technology is expanding beyond its core renewable energy businesses to capitalize on the growing power demands of AI, data centers, and semiconductor manufacturing, positioning itself as a provider of both clean energy and AI-related infrastructure. At its annual shareholders' meeting, the company reaffirmed its focus on green electricity trading, energy storage, energy management, and overseas expansion, while highlighting new investments in AI energy infrastructure through its subsidiary J&V Super Computing.
Management expects the AI-related business to begin generating revenue in 2026, accelerate significantly in 2027, and contribute a double-digit share of group revenue by 2030. The company also emphasized the strong performance of its energy businesses, noting that its electricity retail subsidiary GREENET has secured more than 34.1 billion kWh of green power wheeling contracts, while energy storage subsidiary Recharge Power has accumulated over 412 MW/1,055 MWh of projects across Taiwan and international markets.
As AI drives a surge in demand for reliable, low-carbon electricity, J&V Energy aims to leverage its expertise in renewable power, storage, and energy management to create multiple growth engines and strengthen its presence in both domestic and overseas markets.
Article edited by Jack Wu
