TrendForce

👀 With NVIDIA’s H200 now cleared for China, AI chip momentum in the country is also heating up #Alibaba’s T-Head GPUs have reportedly shipped 470K units by early 2026, with 60%+ powering external clients.💡More: buff.ly/ky7Ay1j 🔗

🔥 NVIDIA doubles its 2025–2027 AI chip revenue target to $1T—but CSPs are coming after Team Green. As #Google and #AWS accelerate #ASIC push in 2H26, #MediaTek and #Alchip may benefit. 💡More: buff.ly/ntt75Sr 🔗

🔥 #Samsung's HBM4 bet is getting serious: Over half of its Pyeongtaek foundry capacity reportedly going to in-house #HBM4 base dies — just as #OpenAI joins NVIDIA and AMD as a rumored major customer. 💡More: buff.ly/jkonUCy 🔗

👀 As memory giants move away from avoiding long-term lock-ins, they may shift to 3–5yr binding deals. #Samsung is reportedly in talks with #Google and #Microsoft, with a rumored $10B+ prepayment to prevent cancellations.💡More: pse.is/8useqs 🔗

#Micron’s $5 billion #CapEx hike for FY2026 (totaling over $25 billion) has raised market concerns regarding excessive capital spending. The primary driver of this expanded CapEx is the acquisition of #PSMC’s Tongluo facility, which is slated to begin contributing to production capacity by FY2028. #TrendForce indicates that this expenditure increase is imperative to accommodate future market demand—most notably trajectories extending past 2029.

🚀 Breakthrough in magnetic memory from #China: Tsinghua researchers reportedly achieved zero-field switching of chiral antiferromagnets—overcoming a key bottleneck of field dependence and low efficiency. 💡More: pse.is/8uqntp 🔗

🔥 Fresh off #HBM4 deliveries to NVIDIA, the memory big three are already eyeing the next front: Samsung leans on 1c DRAM and its in-house 4nm logic dies, while #SKhynix is reportedly considering a leap to #TSMC’s 3nm for its logic dies.💡More: pse.is/8urckz 🔗

👀 As AI chips stack up in #CoWoS 2.5D/3D packages, a business boom emerges: Japan's Rix is gearing up to mass-produce advanced cleaning machines, scrubbing flux residues from tiny gaps between chiplets, interposers, and substrates.💡More: pse.is/8uqgdy 🔗

🔥 TSMC and Samsung’s sub-5/4nm nodes are tight with price hikes, while mature nodes see diverging demand—how will this reshape end-market inventory strategies?💡More analysis from #TrendForce: buff.ly/bxQAI4l 🔗

👀 Decoupling accelerates—#Murata reportedly plans to separate #China rare earth operations within 3 years to mitigate geopolitical risk. The company supplies ~40% of global #MLCCs.💡More: buff.ly/kb3Hwdk 🔗

👀 The world’s most advanced lithography machine has landed at #imec. #ASML’s EXE:5200 High-NA #EUV reportedly has fewer than a dozen units worldwide—placing Belgium alongside Intel, SK hynix, and Samsung.💡More: buff.ly/aLYIkr2 🔗

💥 Another #NAND giant retiring legacy tech? #Kioxia reportedly plans to phase out Thin Small Outline Package (TSOP) products—mainly used for MLC NAND—according to a client notice from China. 💡More: pse.is/8un87j 🔗

🚀#China’s #SiC milestone: Tiancheng Semiconductor develops a 14-inch SiC single crystal with 30 mm thickness, marking a shift beyond the 12-inch era toward early 14-inch development. 💡More: pse.is/8ukser 🔗

👀 Another #Samsung Foundry moment? After NVIDIA, #AMD also secures Samsung’s #HBM4—but it’s not just about memory. The Korean giant may be leveraging supply to pull some AI chip production away from #TSMC. 💡More: pse.is/8uladb 🔗

Record margins, earnings, and big bets on Capex—#Micron is going all in.

🤖 #NVIDIA’s #GTC keynote spotlights #physicalAI, integrating Cosmos, Isaac, and GR00T to bridge simulation and real-world deployment—a look at key innovations that may accelerate large-scale robotics adoption.💡More: pse.is/8ukpkj 🔗

🚀 As CSPs expand in-house chips, #NVIDIA GTC 2026 shifts to #AI inference. To tackle latency bottlenecks, how can “Disaggregated Inference” combine Rubin and LPUs to push compute limits?💡More analysis from #TrendForce: buff.ly/G78qQiC 🔗

🚀 Rising advanced packaging for AI: #NVIDIA’s Rubin Ultra and Feynman may drive #SoIC, with #TSMC reportedly boosting capacity, potentially lifting Besi, Applied Materials, and TEL.💡More: buff.ly/kOuuiXb 🔗

🚀 #Samsung showcased its #HBM4 at #NVIDIA #GTC, but it's already looking ahead: 2nm base die for #HBM5 and 1d DRAM for HBM5E. The roadmap is ambitious. 💡More: buff.ly/UBgE4IA 🔗

🤔 South Korea’s memory giants shifting gears? Just two months after #Samsung reportedly rejected 2–3 year deals with clients, the company is now said to be exploring multi-year memory contracts of up to 5 years.💡More: pse.is/8ueh8k 🔗