CLOUD & ON-PREM IN THE GEN AI RACE

The story of the hare and the tortoise is a classic that re-appears time and again between different contestants – cars vs horses, oil vs electricity, AC vs DC, mainframes vs PCs, iOS vs Android. Now there is a new pair of contestants in the race for Generative AI in BFSI: cloud and on-prem. Five tech experts from the BFSI sector dive deep into use cases, technical progress, security issues, compliance perspectives, capex, opex, licensing and future growth:

- 

What is the ratio of on-prem to cloud use cases of Gen Al in your organization?

Sandeep Chordia: Around 25% of the use cases are running on cloud. These are mostly in the areas of text processing and summarization.

Rajesh Unde: Currently, our Gen AI use cases a PL Capital are split approximately 60:40 in favor of on-prem deployments. This reflects both legacy application integration requirements and increasing but measured adoption of cloud-native solutions, tailored for scalability and flexible experimentation.

Puneet Asthana: In my view, AI workloads in BFSI industry - the present trend shows on-prem deployments are on a higher side when it comes to customer data-intensive applications such as credit scoring and KYC document parsing. Whereas cloud hosted AI workloads are largely of innovation ideas, customer-facing chatbots, or marketing analytics where personally identifiable information (PII) risk is lower. Based on my experience, there is 70:30 ratio, between on-premise and cloud installations. Major distinction arises when it comes to choosing the LLMs (Large Language Models). As a regulated entity, we choose to go with Agentic AI journeys where the model is dedicated to us and keeps the data in landing zone or in my physical data center.

Dhanasekaran S: As a tech-focused, digital microfinance institution, we strongly believe that digital automation across the value chain has been a key driver of our remarkable growth. Belstar being a cloud-native company, the vast majority of our use cases are hosted in the cloud, while the on-prem deployments are considered only in specific cases where regulatory or operational constraints require it.

Syed Chowdhury: As a financial institution, Prime Bank has regulatory restrictions on sharing data in the cloud. Therefore, it is difficult for us to use cloud-based AI in a meaningful way.