Aerospike debuts agentic AI stack with Gemini, cuts fraud probe time by 90%

Aerospike Inc. is this week at AMD Advancing AI 2026 showcasing an agentic AI stack built on its database, Google's Gemini and Agent Development Kit (ADK), and Cloud C4D virtual machines powered by 5th Gen AMD EPYC processors. The platform addresses a core challenge of agentic AI: managing CPU and memory constraints while delivering sub-millisecond performance for operations that involve dozens of concurrent data lookups and model inferences.

One production use case is agentic fraud detection. A payment flows through three stages: automated risk scoring, AI-assembled case construction, and human analyst review. Because the system orchestrates multiple real-time lookups and inferences before deciding whether to approve, decline, or flag the transaction—all within a fixed deadline measured in milliseconds—Aerospike's multi-threaded, NUMA-aware Hybrid Memory Architecture fully utilizes available hardware. The result is a 90% reduction in investigation workflow time compared to traditional processes. Srini Srinivasan, founder and CTO of Aerospike, stated: "Agentic AI dramatically increases the number of operations, orchestrating dozens of real-time data lookups and model inferences before deciding in milliseconds whether to correctly approve, decline, or flag a transaction."

Infrastructure economics also favor the stack. Google Cloud C4D VMs deliver up to 80% higher throughput per vCPU, while Aerospike delivers up to 80% lower infrastructure costs than legacy databases. AMD itself has deployed Aerospike for grid computing monitoring across its HPC data center, which runs over 20 million jobs daily with more than 1 million running concurrently at peak. The platform tracks CPU utilization, DRAM usage, job start times, and user activity in real time. Rajdeep Sengupta, senior director of application and system engineering at AMD, explained the value: "We selected Aerospike because it can handle large volumes of operational data with predictable performance. That reliability allows us to monitor compute jobs across the grid while maintaining the efficiency required for high-demand modern workloads." The deployment also freed AMD's HPC scheduler from query handling, allowing it to focus on its primary function of job scheduling.

Aerospike's announcement targets a specific pain point in agentic AI deployments: the infrastructure bottleneck created by real-time data lookups and model inference orchestration. When a single transaction decision requires dozens of concurrent operations—risk scoring, database queries, model calls—and must complete in milliseconds, traditional databases strain under memory and CPU constraints. Aerospike's Hybrid Memory Architecture, which the company describes as NUMA-aware and multi-threaded, is designed to maximize utilization of available hardware resources, addressing what the company identifies as the most constrained resources in AI infrastructure: CPU and DRAM.

The partnership with Google Cloud and AMD anchors this positioning in production infrastructure. AMD's own HPC data center, handling over 20 million jobs daily with 1 million+ concurrent jobs at peak, already uses Aerospike for grid monitoring—a real-world validation of the database's ability to maintain predictable performance under extreme operational load. The fraud detection workflow reduction from traditional timelines to 90% faster turnaround exemplifies the concrete value proposition: agents can investigate and decide in real time, moving decisions to AI automation rather than waiting for human review.

No comments yet. Be the first to share your thoughts!