Sundar Pichai opened Google Cloud Next 2026 with numbers that frame the scale of the shift: $70 billion in annual revenue, 48% year-over-year growth, and a $240 billion backlog—more than double last year's $155 billion. Customers are spending 30% above committed levels. Capital expenditure is projected at $175–185 billion, nearly double 2025's outlay. The message: agentic AI is no longer experimental. It's infrastructure.
Google's architecture stacks custom silicon (Ironwood TPUs), frontier models (Gemini 3 Pro, Flash, 3.1 Pro), and a cloud platform that integrates Workspace, Search, and Android. The bet is full-stack control reduces latency and cost for agent workloads. Gemini now reaches 750 million monthly active users, with the API serving 85 billion requests in January 2026—a 142% increase from March 2025. Eight million paid enterprise seats span 2,800 companies.
On the engineering side, the Universal Commerce Protocol (UCP) is open-source, built on REST and JSON-RPC with Shopify, Etsy, Wayfair, Target, and Walmart. It layers in Agent2Agent (A2A) v1.2, now governed by the Linux Foundation's Agentic AI Foundation and in production at 150 organizations—including Microsoft, AWS, and Salesforce. Google also launched managed MCP servers for Maps, BigQuery, Compute, and Kubernetes, with Apigee bridging APIs to agent tools. The goal: treat commerce as a programmable service where agents transact on behalf of users.
Hardware targets agentic throughput. TPU 8t delivers 3x faster training with 9,600 chips per superpod (121 exaflops). TPU 8i optimizes inference. The Virgo network links up to 134,000 TPUs at 47 petabits per second. NVIDIA's A5X with Vera Rubin cuts inference costs 10x. A $750 million partner fund targets deployment and upskilling with Accenture and KPMG—the latter hit 90% Gemini adoption in a month. GE Appliances runs 800+ agents in manufacturing. ServiceNow signed a $1.2 billion, five-year deal.
Security gets agentic: Threat Hunting and Detection Engineering agents auto-generate rules. Agent Gateway enforces policies on MCP and A2A traffic. Agent Identity uses cryptographic traceability. Model Armor blocks prompt injection. Internally, 75% of Google's new code is AI-generated, up from 25% in late 2024. Agents sped a code migration sixfold. Engineers review; some teams use Claude Code.
Kurian's critique of competitors: "They hand you pieces, not the platform." Google's Agentic Data Cloud shifts data from repository to reasoning engine, querying AWS and Azure data via Iceberg without migration. The cross-cloud lakehouse handles agent traffic on C4N/M4N VMs. With 13 product lines over $1 billion each and an 11% cloud share (trailing AWS at 31%, Azure at 25%), the 48% growth rate and profitability signal that integration—not just model capability—is the moat. The race is now about production-scale agents, interoperability, and governance. Supply constraints on power and land persist through 2026. Demand does not wait.
Source: Webpronews