Latest
China is moving forward with limited imports of NVIDIA H200 chips, with roughly 10,000 units each arriving at ByteDance and Tencent. A large gap remains between the 400,000-plus approved units and actual supply, since both US and Chinese approval are required.
IBM cooled two connected modular cryostat cells to below 15 millikelvin simultaneously. Here's what the demonstration actually proves—and what remains unverified for long-distance quantum wiring and qLDPC implementation.
While grid interconnection queues for AI data centers now stretch up to eight years, Amazon and Meta are securing generation capacity outright through nuclear PPAs and self-funded gas turbine investments, bypassing the wait entirely. That cost is increasingly being passed on to household electricity bills through PJM's capacity auctions and Japan's transmission tariffs.
Claude autonomously operated a suite of existing protein design tools, producing binders for 14 of 15 targets. Reproduction would require model access, computational budget, and wet-lab validation.
Flinders University trapped polyiodides in a 'cage' using a starch-derived cyclodextrin polymer, extending the cycle life of an aqueous zinc-iodine battery beyond 60,000 cycles. The nonflammable, low-cost materials bring the technology closer to meeting the requirements of large-scale energy storage, though zinc anode corrosion and scale-up to real cells remain the next hurdles.
A Pew Research Center survey found that 55% of Americans aged 18-29 said their concerns about AI outweigh their excitement. 73% predicted that AI will reduce the number of jobs in the U.S. over the next 20 years, with high usage rates and skeptical views coexisting.
Windows 11's Task Manager can now display per-app NPU utilization, execution engine, and memory usage. Standard tools make it easier to pinpoint where AI processing runs and what's driving the load.
The US critical and emerging technologies list has been reorganized from 18 domains to 14. While explicit references to data centers and battery storage have disappeared, federal R&D funding allocation is now tied to basic research and national security applications.
A ground-up explainer on HBM (High Bandwidth Memory), the technology shaping AI semiconductor performance. Covers the mechanics of stacking DRAM and connecting it via TSVs, the differences from GDDR, the three-way competition and base die logic shift in the HBM4 generation, the market's rapid expansion toward roughly $60 billion, the resulting DRAM price surge, and the business opportunities for Japanese companies.
Reuters reports that new SF4 orders from Samsung rose 10-15% in the US and China, and 5-10% in Taiwan. While demand from China exceeds supply capacity, prioritization of US customers and capacity reserved for in-house use are constraining the production allocation.
