Google I/O Unveils Gemini Spark; Nvidia Delivers First Vera CPUs to AI Labs

Google's biggest I/O keynote in years introduces an always-on AI agent and a new multimodal model family as Nvidia crosses a hardware milestone with its first purpose-built agentic AI CPUs.

🤖 Google I/O 2026: Gemini Spark, Omni Flash, and 3.5 Models Reshape the AI Stack

Decoded: Google unveiled its most significant AI product expansion in years at Google I/O 2026 on May 19, centered on three new releases. Gemini Spark is a new always-on AI agent that runs 24/7 on Google Cloud virtual machines and can autonomously write emails, build continuously updated study guides, and monitor credit card statements for hidden subscription fees — connecting to Google Workspace (Docs, Gmail, Sheets, Slides) and third-party apps including Canva and Instacart. Gemini 3.5 Flash — described by Google as significantly faster, better at agentic tasks, and with improved coding capabilities — becomes the default model for the Gemini app and AI Mode in Search starting today; Gemini 3.5 Pro follows next month. Gemini Omni Flash, a new model family capable of generating video from any combination of text, photos, video, and audio inputs, is rolling out in the Gemini app, Google Flow, and YouTube Shorts. Google also announced Android vibe-coding via AI Studio, letting users build, test, and publish native Android apps to the Play Store using natural language prompts. Project Aura smart glasses, developed in collaboration with Xreal, received an updated preview. (The Verge, May 19, 2026)

Why it matters: Gemini Spark is Google's direct entry into the always-on personal AI agent market — competing with OpenAI's memory-enabled GPT models and Microsoft's Copilot 365. Spark runs as a persistent cloud service that monitors and acts on Google Workspace data continuously rather than responding to prompts, repositioning Gemini from a conversational assistant to a background execution layer. For Alphabet (GOOGL), this is a higher-retention, higher-engagement product model. Gemini Omni's any-input-to-video generation, if it matches previews, would be the first production model to unify multimodal generation at scale. The AI Studio Android vibe-coding pathway lowers the barrier to Play Store publishing by orders of magnitude.

🗄️ Nvidia Hand-Delivers First Vera CPUs to Anthropic, OpenAI, SpaceXAI, and OCI

Decoded: Nvidia hand-delivered its first Vera CPUs — the company's first processor designed specifically for agentic AI workloads — to Anthropic in San Francisco, OpenAI in Mission Bay, SpaceXAI in Palo Alto, and Oracle Cloud Infrastructure in Santa Clara last week, Nvidia's official blog confirmed on May 20. CEO Jensen Huang told Dell Technologies World that AI infrastructure demand is now "going parabolic, utterly parabolic." Vera is Nvidia's first CPU architecture, extending its silicon portfolio beyond GPUs into the processor tier that orchestrates AI agent workflows. Designed to pair with Nvidia's GB300 Grace Blackwell Superchip architecture, Vera delivers memory bandwidth and latency characteristics matched to AI inference and agent coordination tasks. The four hand-delivery recipients — three AI labs plus a major cloud provider — are Nvidia's designated launch customers for the agent-era silicon cycle. (NVIDIA Blog, NVIDIA Newsroom, May 2026)

Why it matters: Nvidia entering the CPU market means the company now competes at every layer of the AI computing stack — GPU, CPU, and interconnect — a direct threat to Intel and AMD in the processor segment. Vera's early delivery to Anthropic, OpenAI, and SpaceXAI creates deep hardware-software integration before alternatives can ship. Agentic AI systems require tight CPU-GPU coordination; Nvidia controlling both gives it an architectural advantage that is harder to replicate on mixed vendor stacks. For investors, Jensen Huang's "parabolic" demand commentary at Dell Technologies World suggests Nvidia is already quoting commercial volumes on its next silicon cycle.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team