Topic
Explore the latest artificial intelligence news and research.
Arm has unveiled CSS for Mobile 2, its next-generation smartphone compute platform. With a C2 CPU cluster featuring double the SME2 units and a GPU with a dedicated integrated AI accelerator, it aims to deliver both agentic AI and desktop-class graphics within a 1W power envelope.
Starting from reports of an approximately 10% October price hike on Intel's PC CPUs, this article traces the pattern of incremental price revisions and server-first supply allocation, examining the shift toward prioritizing gross margins and the conditions—such as 18A and 14A mass production—needed to resolve ongoing supply constraints.
Huawei's two-layer stacked Kirin 2026 reportedly delivers major power savings at equivalent performance. But at peak performance, power density exceeds the previous generation, and absolute junction temperature figures remain unpublished.
DRAM industry revenue rose 59.5% in Q2 2026, but bit shipments increased only modestly. Here's why slowing price growth in Q3 shouldn't be read as a sign that supply is catching up, based on company-by-company market share and long-term contracts.
According to Similarweb's August 2026 survey, ChatGPT's web traffic share rebounded to 55.5%. As Gemini's rapid pursuit stalls and Claude surges to 9.3%, the center of gravity in generative AI competition is shifting from browsers to operating systems and apps.
A new report claims Apple has locked in a multi-year NAND supply contract, with market watchers pointing to Kioxia as the likely partner. Here's what's confirmed, what's speculation, and why AI-driven demand may be reshaping how Apple buys memory chips.
An examination of KIOXIA's CXL-based memory expansion module XL1, based on September performance simulations and August shipment plans, clarifies how it divides roles with the GP1 and CM10 product lines.
Linus Torvalds' Linux 7.3-rc2 landed with 681 commits, the largest of any rc2 across 16 cycles since v6.8. AI wasn't among the causes he actually named—his mention of it was a joking aside, not a firm attribution. Here's what the numbers show about the line between scale and cause.
With its integration into Gemini, Lyria 3.5 becomes a music-generation foundation that offers different tools for 30-second clips versus multi-minute compositions. It includes SynthID watermarking and input restrictions, but iterative editing and rights clearance remain the user's responsibility.
Bloomberg reports that DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT chips at a 1GW-class data center in Inner Mongolia. The chips will be used for inference only; training remains on NVIDIA. The 160,000 figure represents roughly 30% of Huawei's published cluster configuration, with component supply—not compute capacity—constraining delivery.
