Google Cloud announced a range of new compute capabilities at Next '26 aimed at supporting both general-purpose and agentic AI workloads. Key launches include: Google Axion N4A GA (up to 2x better price-performance vs x86 VMs), GKE Agent Sandbox GA with Axion N4A (30% better price-performance vs competitors), C4A.metal bare metal instances in preview, expanded C4 Intel Xeon 6 support, and Flexible CUDs expanded GA. New network-optimized instances C4N (95M packets/sec, 400 Gbps bandwidth) and memory-optimized M4N are in preview, alongside new Z4D and Z4M storage-focused instances. Storage improvements include Hyperdisk Balanced with 160K IOPS, Hyperdisk ML boosted to 2 TiB/s aggregate throughput, and Hyperdisk Exapools GA. Customer case studies from Unity, Deutsche Börse, WP Engine, eDreams ODIGEO, Shopify, and others highlight real-world performance and cost gains.
Table of contents
Run AI and general purpose workloads togetherRun I/O and latency-sensitive workloads togetherHandle demanding storage requirementsFluid infrastructure for the agentic eraSort: