Oracle just fired up to 30,000 people with a 6 a.m. email to fund its AI data center obsession, a tiny French startup's 10-billion-parameter model is outperforming GPT-5.4 at actually using a computer, and Q1 venture funding hit a staggering $297 billion with AI swallowing 81% of every dollar. Plus, Google wants your voice assistant to finally sound human, and NVIDIA is shipping AI chips to orbit.
Oracle Lays Off Up to 30,000 Workers to Fund $156 Billion AI Data Center Buildout
Oracle began executing what may be the largest layoff in its history on March 31, sending termination emails to employees across the US, India, Canada, and Mexico at 6 a.m. with no prior warning. TD Cowen estimates 20,000 to 30,000 workers were cut, roughly 18% of Oracle's 162,000-person workforce, with the move expected to free up $8-10 billion in cash flow to help fund $156 billion in committed AI infrastructure spending. Oracle disclosed a $2.1 billion restructuring charge in its latest SEC filing. It's the starkest example yet of a legacy tech giant gutting its workforce to bet everything on AI compute.
H Company's Holo3 Sets New State of the Art for Desktop Computer Use, Beating GPT-5.4 at 1/10th the Cost
French AI startup H Company released Holo3, a 122-billion-parameter mixture-of-experts model with only 10 billion active parameters that scored 78.85% on the OSWorld-Verified benchmark for desktop computer use, surpassing both GPT-5.4 and Claude Opus 4.6. The model can open applications, click buttons, fill out forms, and complete multi-step workflows across emails, PDFs, and spreadsheets, just like a human operator. It's available on Hugging Face and through H Company's inference API. This is a notable upset: a relatively small open model beating frontier proprietary systems at one of the most commercially important AI tasks at a fraction of the inference cost.
Q1 2026 Venture Funding Hits Record $297 Billion as AI Swallows 81% of All Investment
Global venture capital hit $297 billion in Q1 2026, a sum that exceeds all full-year venture totals before 2018 and equals nearly 70% of everything deployed in all of 2025. AI accounted for 81% of that capital, up from 55% a year earlier. Four frontier AI labs alone collected $186 billion: OpenAI ($120B), Anthropic ($30B), xAI ($20B), and Waymo ($16B), representing 64% of total global venture activity for the quarter. US-based companies captured 83% of all dollars invested. This isn't just a boom anymore, it's a wholesale rearrangement of global capital allocation around AI.
Google Launches Gemini 3.1 Flash Live, Its Most Human-Sounding Voice AI Model Yet
Google released Gemini 3.1 Flash Live, a real-time voice and audio model that collapses the traditional voice processing stack by handling acoustic nuances like pitch, pace, and background noise natively rather than relying on text transcription. The model powers an upgraded Gemini Live experience with fewer awkward pauses and twice the conversational memory, and is rolling out globally across Search Live in 200+ countries and 90+ languages. Verizon, LiveKit, and Home Depot are already using it. This is Google's play to make voice the default AI interface, and the native audio processing approach could make current speech-to-text pipelines look primitive.
NVIDIA Launches Vera Rubin Space Module for Orbital AI Data Centers
NVIDIA unveiled the Vera Rubin Space Module, a space-grade AI computing platform delivering up to 25x the inference performance of an H100 in a form factor designed for the size, weight, and power constraints of orbit. Partners including Aetherflux, Axiom Space, Kepler Communications, and Planet Labs are already building on the platform for orbital data centers, geospatial intelligence, and autonomous space operations. With Earth-based data centers running into energy and cooling constraints, orbital facilities powered by solar energy represent a serious long-term alternative. Jensen Huang noted the unique challenge: in space there is no convection for cooling, only radiation.
