Latest
A report says SK hynix is struggling with hybrid bonding for next-gen HBM while Samsung has shipped samples. Company announcements and manufacturing details show what polishing, cleanliness and timing mean for moving from prototypes to mass production.
Code in Android 17 QPR3 Beta 1 points to a "Face Active Illumination" feature that could help Pixel face unlock in the dark. The light source is unconfirmed. We compare it with Google's security classes, Galaxy's screen assist and Face ID's infrared approach.
Anduril has announced Arsenal-2, a $3.7 billion submarine manufacturing site. We look at whether supplying parts to existing shipbuilders and using ArsenalOS for quality control can support higher output, along with funding, safety certification and the 2030 timeline.
Huawei claims Ascend has overtaken NVIDIA's share in China. We check its reasons for adopting NPO, on cost and repairability, against OIF materials, and use DeepSeek's public code to separate component-level efficiency from whole training-system performance.
Anthropic's new Cyber Verification Program tiers access to its most capable models by permitted work and customer security requirements. Company case studies show the verification and remediation challenges, and the rules cover authentication deadlines and customer-held monitoring logs.
UK AISI measured two leading open-weight AI models at 4–7 months behind closed frontier models in cyber capability, with the same exercise costing roughly half as much to about 1/71 as much, depending on the model. Weighing overseas attack cases, the disclosed entry points of Japanese leaks and National Police Agency statistics, we ask whether criminals' improved capabilities can be called the cause of Japan's breaches.
Solidigm is adding a contract manufacturing site in Taiwan for data center SSDs. Its official notice sets December 24 as the customer readiness date and says specifications stay unchanged. We look at ties to Taiwanese server makers, CoreWeave's multi-year deal and SK hynix's M17 plan to see where extra SSD capacity helps supply.
OpenAI has released 722 AI-generated math manuscripts on GitHub. We look at what the numbers mean, how far Lean verification extends across the 162 formalized manuscripts, and how the release compares with mathematicians' publication guidelines.
Claude for Google Workspace is in public beta, letting Claude edit files directly and use your saved work skills. Comment and chart support differs by entry point, and enterprises should check data retention terms.
AMD plans to boost chip supply sharply in 2027. Comparing the deployment plans of OpenAI, Meta and Anthropic, and the Helios rack of 72 GPUs and 18 CPUs, we look at what must come together for added output to become working capacity.
