🛰️ SpaceX and Nvidia target orbit with Vera Rubin 🛰️
The two companies plan to roll out AI compute infrastructure in orbit by late 2027.

SpaceX and Nvidia have officially partnered to build orbital data centers, with SpaceX's Starmind facilities running Nvidia's Vera Rubin NVL72 rack — 72 chips operating as a single computer, each delivering up to 25× the performance of the H100. Elon Musk described the space-optimized rack as simpler, denser and lighter, engineered for orbital radiation and heat. Analysts estimate orbital compute still costs over 4× more than ground-based alternatives, though Musk claims the economics will flip within years. Removing terrestrial power and cooling constraints could reshape datacenter strategy for hyperscalers and sovereign AI programs.

⚡ OpenAI's Jalapeño chip beats Nvidia GPUs by 3.6× ⚡
OpenAI's first custom chip targets inference speed and energy efficiency.

OpenAI published the first performance benchmarks for Jalapeño, its custom inference accelerator built for low-latency agent workloads. The 700-watt chip delivers up to 3.6× faster responses and 1.9× better energy efficiency than Nvidia's 1,200-watt flagship GPUs, a direct challenge to Nvidia's inference dominance. AI itself designed circuits and programmed kernels during an accelerated nine-month development cycle. Deployment starts inside OpenAI's own infrastructure by year-end, with two more generations through 2027 — and the chip won't be sold externally, making this a pure vertical-integration play.

🤗 Nvidia buys Hugging Face for $12.9 billion 🤗
The $12.9B deal reshapes open-source AI platform access forever.

Nvidia has agreed to acquire Hugging Face, the open-source platform hosting millions of models and datasets, for $12.9 billion — nearly three times its $4.5 billion 2023 valuation, after an earlier $500 million minority-stake offer was declined. The deal hands the chip giant control of one of the most critical platforms in the open AI ecosystem, raising immediate questions about platform neutrality: Hugging Face today supports models running on Nvidia, AMD and Google hardware alike. Founded in France, it is also one of Europe's most significant AI assets to change hands to a US acquirer.

INTERESTING TO KNOW

🔍 Mystery model Ox Alpha offers free 1M-token access 🔍

An anonymous model called Ox Alpha appeared on OpenRouter with free access, a 1M-token context window and multimodal input, triggering a community-wide hunt for its creator. Built for coding and sustained agentic work, it scored 63% on full DeepSWE testing while consuming far fewer tokens per task. Forensic analysis — response patterns, Chinese zodiac naming — points to China's Zhipu AI, possibly GLM-5.3 Flash or GLM-6. All four anonymous model drops on OpenRouter in the past six months have traced back to Chinese labs.

🎬 Gemini Omni 1.1 Flash ships 4K video upscaling 🎬

Google launched Gemini Omni 1.1 Flash, a major upgrade to its video-generation stack now available through the Gemini API. Developers can extend scenes up to 40 seconds, interpolate between first and last frames, and run 4K upscaling for higher-resolution output. The model has also climbed to the top of Arena's text-to-image leaderboard, applying direct competitive pressure on Runway and Sora.