Latest
Samsung's preliminary Q3 2026 operating profit is 107.4 trillion won. Official figures show profit margins, the roles of HBM4 and conventional DRAM, and how chip profits coexist with device-side losses.
Researchers at the University of Electro-Communications and Nichia report that RbLi(Li₃SiO₄)₂ has a theoretical density 46% lower than LLZO and a Li-ion migration barrier of 0.30 eV. Measured conductivity is only 2.9×10⁻⁸ S/cm at 150 °C, so lithium-rich compositions that overcome defect formation are the next test.
Deployable Energy's Unity small reactor will move to full-power demonstration at the DOME facility in 2027. We examine the 4.95% enriched fuel, the roles of water and helium, how the design's 2 MW of surplus heat is handled, and the plan from fueled transport to post-operation relocation, based on primary sources.
An unconfirmed report says Anthropic's custom AI chip may be made on Samsung's 2nm process. We check the official supplier ties and a precedent for Japanese customers, then look at HBM capacity, inference memory bandwidth and the 1,300W design figure to see what it would take to make the chip work and pay off.
A comparison of OpenDots and Open Dot examines scheduling, memory and approvals. Together with OpenAI's safety evaluation and user reports, it shows that choosing a model alone doesn't solve the problem of continuous work.
Nous Research has raised $90 million and announced a push to take Hermes into the enterprise. We look at how its memory and skills enable self-improvement, how Business and Enterprise differ in operation, and the permissions and cost controls needed to share knowledge across an organization.
NVIDIA's DGX Station for Windows is due in Q4 2026. We break down the 748GB memory split, the conditions behind the 20 PFLOPS figure, and what a trillion-parameter model needs in NVFP4, plus enterprise management, Arm compatibility, long-context and concurrent inference.
Chrome 155 will officially support JPEG XL. We look at the shift to Rust, lossless recompression of existing JPEGs, how it compares with AVIF, decoding speed, and differences between browsers.
Pre-orders for Surface Laptop Ultra have begun. Japan's ¥513,480 starting price is for the 24GB model, not the 128GB one. We examine the conditions behind Microsoft's AI comparison and weigh chips from Apple, AMD, Intel and Qualcomm.
3DCenter's German GPU price comparison shows GeForce RTX 50 cards up 66% on average. We compare the extra cost of 16GB models against Radeon and look at memory supply and value-for-money limits.
