Aerospike has introduced an agentic AI stack for real-time fraud detection with Google Gemini, Google's Agent Development Kit and AMD EPYC processors. It says the system can cut fraud investigation workflow time by 90%.
The offering combines automated risk scoring, AI-generated case assembly and human review in a single payment investigation flow. Aerospike says the design supports fraud decisions that must be made in milliseconds, even when a single interaction triggers multiple data lookups and model inferences.
The launch reflects a broader infrastructure challenge for companies deploying agentic AI systems. These applications can place heavy demands on compute and memory resources as they orchestrate several steps within one transaction, particularly in areas such as payments, where response times are tightly constrained.
Aerospike says its database works with Google's AI stack, including Gemini, the Agent Development Kit and Cloud C4D virtual machines powered by 5th Gen AMD EPYC processors. The combination is intended to maintain predictable transaction processing under large-scale workloads.
According to Aerospike, the fraud workflow begins when a payment is made, moves through automated risk scoring, into an AI-assembled case file and ends with a human analyst's decision. The company says that sequence reduces the time needed to complete an investigation by 90%.
Aerospike also outlined the commercial case for the stack, saying Google Cloud C4D virtual machines deliver up to 80% higher throughput per vCPU, while its database can reduce infrastructure costs by up to 80% compared with legacy databases.
Srini Srinivasan, Founder and Chief Technology Officer at Aerospike, described the technical challenge of applying agentic AI to fraud systems.
"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," said Srinivasan, Founder and Chief Technology Officer at Aerospike.
"The combination of Aerospike and Google C4D powered by AMD EPYC ensures that fraud detection and other AI applications relying on real-time data stay fast, every time, and for every user, even when a single interaction involves many operations and a fixed deadline," Srinivasan said.
AMD deployment
Aerospike is also using the announcement to highlight an existing deployment at AMD. The chipmaker's grid computing monitoring platform runs on the Aerospike database and tracks CPU use, DRAM consumption, job start times and user activity across compute jobs in real time.
According to Aerospike, AMD's HPC data centre runs more than 20 million jobs a day, with more than 1 million running concurrently at peak. The monitoring workload is designed to manage CPU and memory consumption across those jobs without adding extra strain to scheduling systems.
Rajdeep Sengupta, Senior Director of Application and System Engineering at AMD, said predictable performance was a key factor in the database choice.
"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," said Sengupta, Senior Director of Application and System Engineering at AMD.
"This also allowed us to offload query capabilities from the HPC grid scheduler so that the scheduler does its most important job - scheduling - rather than handling queries," Sengupta said.
The initial deployment runs in a single AMD data centre. Aerospike says the architecture is designed to scale as AMD extends the monitoring system to more data centres and regions to support larger electronic design automation workloads across its grid.
The announcement places Aerospike among database suppliers seeking to tie their products more closely to AI infrastructure, especially in use cases where low-latency data access and predictable performance are central requirements. Fraud detection is one of the clearest examples because institutions must combine model output, transaction data and operational checks in a narrow time window while keeping human investigators in the loop.
Aerospike says efficient management of constrained CPU and DRAM resources at scale is central to its pitch to customers using machine learning, generative AI and agentic AI systems. In AMD's case, that means supporting real-time visibility into more than 20 million compute jobs a day, while more than 1 million jobs can run at the same time.