Bilingual

China's LongCat-2.0 Becomes the Biggest AI Model Without Nvidia Chips
21 days ago
- Meituan released LongCat-2.0, a 1.6 trillion parameter open-source LLM, trained end-to-end on domestic Chinese hardware without NVIDIA chips.
- It's the industry's first trillion-parameter model to complete both training and inference on domestic hardware, marking a major milestone for China's tech self-reliance.
- The model uses Huawei's Atlas-950 SuperPoDs and HCCL software, mirroring NVIDIA's setup, and shows strong performance but still trails top global models like GPT-5.5.
- The launch demonstrates frontier-scale training is viable on Chinese hardware, potentially narrowing the gap with Western systems faster than expected.
- Challenges remain, including a less mature software ecosystem and memory limits compared to NVIDIA's banned H800 chips.
Top Companies in the Global Hybrid Memory Cube High-Bandwidth Memory Market
21 days ago
- AI, ML, and HPC drive demand for HBM/HMC due to high bandwidth and low latency needs.
- HBM market size estimated at USD 4.56B in 2025, projected to reach USD 36.6B by 2035 with 23.1% CAGR.
- Asia-Pacific dominates market due to semiconductor manufacturing in South Korea, Taiwan, Japan, and China.
- North America shows fastest growth, fueled by hyperscale data centers and AI infrastructure.
- Key trends include AI-driven demand, shift to memory-centric architectures, and rapid evolution from HBM3 to HBM4.
- Top companies include Samsung Electronics, SK hynix, Micron Technology, Intel, and NVIDIA.
- 3D stacking and chiplet packaging technologies advance HBM integration and efficiency.
- HBM is critical in GPUs, AI accelerators, and data centers for handling parallel processing workloads.
Samsung chip division's 1-year profit beat past 40 years of profits, combined
17 days ago
- Samsung's chip division is projected to achieve greater operating profit in 2026 than its cumulative profit over the past 40 years in the semiconductor business, driven by increased memory and storage prices.
- Samsung surpassed Nvidia to become the world's most profitable company in the second quarter of 2026, with a 19-fold increase in quarterly operating profit, largely due to AI-driven memory demand.
- The DS division accounted for about 94% of Samsung's total operating profit in the first quarter of 2026, with full-year 2026 operating profit estimated near 300 trillion won ($196 billion).
- Contract prices for DRAM and NAND have risen sharply in 2026, with AI server demand leading to supply shortages and operating margins of 40% to 50% on NAND in the first half.
- Samsung is negotiating further price increases for commodity DRAM in the third quarter and expects tight memory supply to continue through at least 2027.
US seeks share of Korean chipmakers' 'excess profits'
8 days ago
- The U.S. seeks a share in profits from Korean semiconductor companies, citing high demand for their chips.
- U.S. Deputy Trade Representative argued American companies contributed to Korean chipmakers' earnings by purchasing large volumes.
- Korean semiconductor exports hit record highs in the first half of the year, with shipments to the U.S. surging significantly.
- Publicly, the U.S. has focused on urging Korean firms to expand manufacturing in America rather than profit-sharing.
- Debates in Korea center on redistributing 'excess' profits to subcontractors, suppliers, or the public due to taxpayer support.

Hasty Briefsbeta

semiconductors

China's LongCat-2.0 Becomes the Biggest AI Model Without Nvidia Chips

Top Companies in the Global Hybrid Memory Cube High-Bandwidth Memory Market

Samsung chip division's 1-year profit beat past 40 years of profits, combined

US seeks share of Korean chipmakers' 'excess profits'