Topic
Explore the latest artificial intelligence news and research.
Google's September Android Drop introduces five features, each with its own minimum OS, RCS, and regional conditions. Here's a breakdown of the requirements, staged rollout, and safety limits for users in Japan, based on primary sources.
Google Pics is now generally available, bringing Nano Banana-powered object editing, version history, and collaboration to Workspace. Here's a breakdown of the September phased rollout, undisclosed generation limits, and the boundaries of AI provenance data.
Claude Fable 5.1 outperforms its predecessor on long-horizon tasks, but the price cut applies only to cached reads. Comparing it with the same-foundation Mythos 5.1 reveals how deployment conditions, including safety systems, shape real-world performance.
Android 17 introduces a new permission separating access to home Wi-Fi networks from internet access. However, the protection only applies to apps targeting API 37, meaning its real-world effectiveness depends on how quickly developers update their apps.
We break down Apple's official performance multipliers for the M5 Max Mac Studio down to the footnotes, weighing the ¥170,000 domestic price increase from the M1 Max generation and the limits of pre-release benchmarks to determine which tasks actually benefit from upgrading.
The spot price of 36GB HBM3E has reportedly reached about $2,100 per unit, four to five times the long-term contract price. Samsung plans to allocate 60-70% of its memory production capacity to long-term contracts running through 2031, and whether a buyer is inside or outside those contracts is starting to determine the terms under which AI memory can be procured.
NVIDIA's $3.5 billion purchase of MediaTek convertible bonds is examined as a strategy linking MediaTek's advanced packaging capabilities with NVLink infrastructure, including how the funds will be used, the non-exclusive nature of the deal, and mass-production constraints.
We break down TSMC's projected "50x" leap by 2029 into the roles played by CoWoS, SoIC, and optical interconnects, examining where effective performance ends and mass-production conditions begin. We also clarify the difference between the 50% silicon photonics market forecast and actual CPO adoption rates.
Broadcom's VMware AI Factory for AMD touts GPU sharing and usage visibility, but pricing, release dates, and measured cost reductions remain undisclosed, leaving key conditions for verifying the savings still missing.
Intel expects revenue from EMIB-T-based advanced packaging to grow starting late 2027 and scale up by 2029. Customer qualification, yield, and financial disclosure will determine whether the technology becomes a real external business.
