Daily AI RoundupsSaturday, 4 July 2026

The 04 Jul 2026 roundup leads with CNBC's report that Nvidia is entering revenue-sharing agreements with AI startups, offering compute credits in exchange for a share of future profits. It also covers Reuters' report that Alibaba and Tencent are backing Kuaishou's Kling AI video model in a $2.8B fundraise, Bloomberg's report on Micron's ¥1.5T Japan factory expansion for advanced memory chips, llama.cpp's topk-MoE fusion for 288 CUDA experts, Thariq's field guide to Claude Fable 5, and VittoStack's public jailbreak review describing Fable 5's layered defenses.

• AI compute
• AI funding
• AI video
• AI chips
• open-source inference
• coding agents
• AI safety

Stories in this roundup

1. 01
   Nvidia enters revenue-sharing agreements with AI startups, offering compute credits in exchange for a share of future profitsCNBC / X
2. 02
   Alibaba and Tencent back Kuaishou's Kling AI video model in a $2.8 billion fundraiseReuters / X
3. 03
   Micron breaks ground on ¥1.5 trillion ($9.3B) factory expansion in Japan for advanced memory chipsBloomberg / X
4. 04
   llama.cpp enables topk-MoE fusion for 288 experts on CUDA, expanding efficient local inference to large mixture-of-experts modelsGitHub
5. 05
   Anthropic’s Thariq publishes a field guide to Claude Fable 5, framing prompts, skills, and context as a map that has to be tested against unknownsThariq / X
6. 06
   VittoStack reports Fable 5’s jailbreak defenses are heavily layered, with most attempts blocked and remaining bypasses requiring long, costly, multilingual probingVittoStack / X