Below are the top DIGITIMES Asia stories from August 25 to 31, 2025.
Intel's AI chip comeback: 2nm Jaguar Shores to rival Nvidia
Intel is retooling its AI roadmap, shelving the long-planned Falcon Shores accelerator and shifting focus to a new platform called Jaguar Shores. Test photos of the chip have already surfaced, and at its AI Summit in July 2025, Intel confirmed Jaguar Shores will incorporate SK Hynix's next-generation HBM4 memory. Unlike Falcon Shores, now limited to internal testing, Jaguar Shores is being positioned as Intel's flagship rack-scale AI system, aimed at competing directly with Nvidia and AMD, which are expected to roll out HBM4-based chips around 2026.
Samsung is reportedly considering a stake in Intel to boost chip packaging and challenge TSMC
Samsung Electronics is reportedly considering a strategic investment in Intel to strengthen its foundry business and counter TSMC, with South Korean media citing access to Intel's strengths in packaging technologies like hybrid bonding and glass substrates as key drivers. The potential tie-up comes as Samsung expands its US presence with a US$37 billion Texas fab, while political uncertainty looms after US President Donald Trump disclosed that Washington now holds nearly 10% of Intel.
Both companies are racing toward 2nm-class chips—Samsung with its 2nm node and Intel with its 18A process—and analysts say their combined efforts on packaging and cost-cutting could challenge TSMC's dominance and ultimately lower costs for consumers.
Samsung-Nvidia CEO embrace hints at HBM4 deal amid Seoul's US$150B US pledge
South Korean President Jae-Myung Lee used his first US visit to push for deeper tech and trade ties, with Seoul pledging US$150 billion in US investments spanning semiconductors, shipbuilding, and AI. At the KORUS Business Roundtable in Washington, more than 50 senior officials and executives from both countries gathered, including Samsung chairman Jay Y. Lee, SK Group's Chey Tae-Won, Hyundai's Chung Eui-Sun, and Nvidia CEO Jensen Huang, alongside leaders from GM, IBM, Google, and Applied Materials.
India-based boAt, HrdWyr unveil indigenously designed chip with Tata Electronics support
Indian wearables brand boAt has teamed up with semiconductor startup HrdWyr to unveil the Indus 1011, a high-volume chip fully designed in India, with Tata Electronics providing assembly, packaging, and testing. The chip, which will power boAt's next generation of premium wireless charging cases slated for early 2026, marks a milestone in India's semiconductor push by localizing design, IP creation, and back-end manufacturing.
Executives from both companies said the collaboration underscores India's ability to innovate in low-power, AI-driven chip design while advancing strategic autonomy in the supply chain. Analysts say the move strengthens India's semiconductor ecosystem under the government's "Make in India" and "Atmanirbhar Bharat" initiatives, potentially reducing dependence on global suppliers.
Micron CEO reportedly meets Samsung's MX division head, bolstering grip on Galaxy's DRAM supply
Micron is tightening its foothold in Samsung's Galaxy supply chain, supplying about 40% of the LPDDR5X memory for the flagship S25 after Samsung's own chip unit faced yield issues, according to industry reports. Micron CEO Sanjay Mehrotra met Samsung mobile chief Roh Tae-moon in July to reaffirm ties, and analysts say the US firm is likely to keep its share in the upcoming Galaxy S26 as Samsung hedges against future disruptions—cementing Micron's role as a strategic partner despite thin margins and capacity constraints.
Samsung and SK Hynix slash NAND spending as SK Hynix unveils 321-layer breakthrough
Samsung Electronics and SK Hynix are scaling back investment in advanced NAND flash as weak demand and high costs push them toward higher-margin DRAM and packaging, South Korean media reported. Samsung has delayed ninth-generation NAND upgrades at its Pyeongtaek and Xi'an fabs and postponed hybrid bonding adoption until at least 2026, while SK Hynix is prioritizing DRAM and HBM for AI servers. Even so, SK Hynix this week unveiled a 321-layer QLC NAND chip with double the capacity of its predecessor and major gains in speed, efficiency, and density, positioning it for PCs, smartphones, and eventually ultra-high-capacity AI server SSDs.
China reportedly turning to second-hand Nvidia A100, H100 GPUs after H20 ban
China's AI hardware market is being roiled by Beijing's ban on Nvidia's H20 accelerator, which has unleashed a flood of second-hand A100 and H100 GPUs from overseas, industry sources say. Chinese firms are dismantling and repurposing the chips into inference servers, slashing costs and sidestepping export curbs, with resale prices for used H100 machines already down nearly 25% this year and projected to fall further by 2026.
Analysts warn the influx is pressuring domestic chipmakers and could destabilize the ecosystem, even as cheaper inference lowers barriers for AI developers and fuels broader adoption—reshaping supply chains and competition across the global AI market.
Article edited by Jack Wu