01 — The Verge
Google says searches are growing at their fastest rate in five years, but Search revenue growth slowed to 12% in the third quarter. The Verge reports that AI Mode earns less per query than traditional search; Google says currency effects were the biggest reason for the slowdown. Advertisers cannot separately measure Google's AI ad performance, leaving the returns on its expanding AI search business unclear.
Why it matters
Google's search makeover could change the economics of the advertising business funding its AI expansion.
Read the source →02 — The Verge
AMD reported record quarterly revenue of $13.4 billion, including $8.5 billion from data centers, according to The Verge. Gaming revenue fell 40% from a year earlier to $773 million. CEO Lisa Su acknowledged that rising component prices are slowing consumer purchases, while the company forecasts another revenue record next quarter.
Why it matters
AMD's results show the widening gap between demand for data center hardware and price-sensitive consumer technology.
Read the source →03 — Ars Technica
California is accepting $1.42 billion in federal broadband funding while reserving the right to challenge restrictions on its telecom laws, Ars Technica reports. The federal conditions bar enforcement of net neutrality and rate regulations against providers receiving BEAD funds. The dispute could delay or cancel grants intended to bring broadband to 270,571 locations.
Why it matters
The fight puts broadband expansion and states' authority to regulate funded internet providers in the same legal dispute.
Read the source →04 — Ars Technica
Russian technology companies are considering moving data centers underground after Ukrainian drone strikes damaged facilities and disrupted cloud services, Ars Technica reports. Rostelecom has discussed the idea with mining company Norilsk, according to reporting cited by Ars. These remain proposals; meanwhile, the Kremlin is requiring operators to reserve computing capacity for emergency state needs.
Why it matters
Wartime attacks are forcing cloud operators to reconsider physical protection and the state's access to their computing capacity.
Read the source →05 — Meta Engineering
Meta's F3 Lightning lets ads-ranking features be compiled, validated, deployed and rolled back independently, avoiding releases blocked by one faulty definition. Shared execution is reconstructed at startup using stable computation identifiers, a common execution plan and shared memory. Meta reports that comparable extraction features reach production about 3.6 times faster than before, without increasing infrastructure compute costs.
Why it matters
The design offers a useful pattern for separating deployment boundaries from execution efficiency in systems with heavily shared computation.
Read the source →06 — Cloudflare
Cloudflare introduced Zero Wire on Radar, bringing vulnerability explanations, affected deployments, recommended actions and supporting sources into short briefings. It focuses on critical flaws in internet-facing software rather than cataloging every CVE, and is available to noncustomers. Briefings distinguish exploitation evidence from severity and explain where Cloudflare controls can help, while directing readers to vendor fixes.
Why it matters
Security teams get another public starting point for deciding which disclosures affect their systems and what to investigate next.
Read the source →07 — The Register
Apple says fiscal 2026 set a Mac sales record, with September-quarter Mac revenue rising 24 percent to $10.8 billion, The Register reports. The company expects memory shortages to constrain Mac and iPad supply in the December quarter. Higher memory costs are also squeezing margins despite earlier hardware price increases.
Why it matters
Strong demand is running into component limits that Apple says will restrict supply and raise its costs.
Read the source →08 — Simon Willison
Simon Willison reports that Qwen 3.8 27B running on his laptop identified simplified Pac-Man, Pong and Breakout networks from their raw numerical weights, without executing code. Each successful attempt took roughly 46 to 65 minutes; it misidentified Freeway. His experiment reproduces a small part of a broader weight-space programming study while questioning claims that such results herald the end of source code.
Why it matters
The experiment makes a surprising research claim reproducible while keeping its small scope and substantial runtime visible.
Read the source →