Drone Strikes on Data Centers, Qwen3.5 Punches Above Its Weight, DeepSeek Goes Huawei, and Apple Unifies Visual Tokenization
Published on 20.03.2026
Drones Hit Persian Gulf Data Centers
TLDR: Iranian drones struck at least three Amazon Web Services data centers in Bahrain and the UAE, disrupting banking, payments, and business software across the region. The attacks mark the first known military strikes against cloud infrastructure and put trillions of dollars of AI data center investments at risk.
Drones Hit Persian Gulf Data Centers
Qwen3.5 Outperforms Bigger Models, Leads Vision Benchmarks
TLDR: Alibaba released the Qwen3.5 family of eight open-weights vision-language models, with the 9-billion parameter version outperforming OpenAI's 120-billion parameter gpt-oss-120b on most language benchmarks. The largest model, Qwen3.5-397B, beats GPT-5.2, Claude 4.5 Opus, and Gemini-3 Pro on 28 of 44 vision benchmarks.
Qwen3.5 Outperforms Bigger Models, Leads Vision Benchmarks
DeepSeek Snubs Nvidia for Huawei
TLDR: DeepSeek withheld a prerelease version of its upcoming V4 flagship model from Nvidia and AMD, instead sharing it exclusively with Huawei for hardware optimization. The move signals an accelerating decoupling of the US and Chinese AI ecosystems.
DeepSeek Snubs Nvidia for Huawei
A Single Tokenizer for Visual Media (Apple's AToken)
TLDR: Apple researchers built AToken, a unified visual tokenizer that maps images, videos, and 3D objects into a single shared token space with four-dimensional coordinates. The system approaches specialized model performance across all three media types for both generation and classification tasks.