I have written several times recently about the importance of data in supporting artificial intelligence (AI), including generative AI (GenAI) and agentic AI. From a data platforms perspective, this is most evident in the role that analytic data platforms play in supporting the training and fine-tuning of AI models. Operational data platforms also have a role to play in supporting enterprise AI initiatives, however: specifically, by providing the data processing and persistence capabilities to support the inferencing capabilities of intelligent operational applications. The importance of operational data platforms for AI is highlighted by recent acquisition announcements from analytic data platform specialists Databricks and Snowflake, both of which are adding operational database capabilities to their respective data platforms. In this post, I’ll explore the reasoning behind those acquisitions in the context of the capabilities required for operational data platforms to support enterprise AI.
ISG Research defines operational data platforms as environments for organizing and managing the storage and processing of data generated by applications targeted at business users and decision-makers to run the business, including finance, operations and supply chain, sales, human capital management, customer experience and marketing.
Operational data platforms are complemented by analytic data platforms, which are used to run applications used to analyze the business, including decision support, business intelligence, data science, AI and machine learning (ML). Analytic data platforms, including data warehouses and data lakehouses, are also used to support the development, training and fine-tuning of AI models. As I noted in the 2025 ISG Buyers Guide for Operational Data Platforms, the development of intelligent applications infused with contextually relevant recommendations, predictions and forecasting driven by ML, GenAI and agentic AI also impacts the requirements for operational data platforms. I assert that by 2027, two-thirds of enterprises will have adopted new operational database products driven by the need to support the AI inferencing requirements of intelligent operational applications. The impact of this trend is already reshaping the data platforms market. In recent months, two analytic data platform specialists, Databricks and Snowflake, both announced acquisitions to add operational data platform capabilities and expertise to their portfolios, with Databricks announcing its acquisition of serverless PostgreSQL database specialist Neon in mid-May, followed by the subsequent preview launch of the Neon-powered Lakebase offering. In June, Snowflake announced its plan to acquire another PostgreSQL database provider in the form of Crunchy Data. Both companies referenced the importance of the development and deployment of AI agents and applications as significant drivers behind the acquisitions.
So, what are the capabilities required to support intelligent applications, and how do they go beyond the traditional requirements for operational data platforms? The primary function of a data platform is to store and retrieve information relevant to the supported application. Traditionally that has been limited to structured, unstructured and semi-structured data required to support enterprise applications. The emergence of GenAI added the requirement for data platforms used to support operational applications that they be able to interact with GenAI models and access vector embeddings. As I have previously explained, vectors are multi-dimensional mathematical representations of features or attributes of raw data that are used to support GenAI-based applications by enabling rapid identification and retrieval of similar or related data. Support for storing vectors has quickly become a table-stakes requirement for data platforms, as evidenced by 90% of providers assessed in the ISG 2025 Buyers Guide for Operational Data Platforms grading A- or above for vector model support. The other 10% all have the functionality in development.
Vector retrieval is particularly important for improving accuracy and trust with GenAI via retrieval-augmented generation (RAG), which is the process of combining vector embeddings representing factually accurate and up-to-date information from a database with content automatically generated by a GenAI model. 
I assert that through 2027, almost all enterprises developing applications based on GenAI will invest in data platforms with vector search and RAG to complement foundation models with proprietary data and content. Not surprisingly, support for RAG has also rapidly become widespread, with 80% of providers assessed in the ISG 2025 Buyers Guide for Operational Data Platforms grading A- or above for RAG support. Data platform providers seeking to differentiate in relation to vector storage and retrieval can still do so by focusing on vector indexing as a means of improving the performance and accuracy of similarity search results, as well as implementing hybrid search capabilities using both vector and text search to combine similarity search results with other factors (such as text filtering to enhance relevancy for a specific user role or use case).
The rise of agentic AI means that data platforms providers looking to differentiate need to look beyond vector storage and retrieval towards requirements for automated and accelerated database provisioning. Agentic AI requires data infrastructure and operations that are:
- Agile – to enable collaboration and interactivity
- Adaptive – to technology and business change
- Automated – to reduce manual processes
- Active – to drive impact through two-way integration with operational applications
- Agentic – to ensure they are part of the agentic process
The ability to write-back data derived from analytic systems to operational applications is typically delivered via reverse ETL integration tools, while Couchbase has made the capability a core component of its Capella Columnar analytics service. Native agentic capabilities are also being added by several database providers, including support for agentic frameworks and standards such as Model Context Protocol (MCP), which has emerged as a key open standard for enabling agentic AI by providing connectivity between agents and the data and tools needed to support automated action execution.
Data platforms are at the very early stages of supporting agentic AI, with only 27% of providers assessed in the ISG 2025 Buyers Guide for Operational Data Platforms grading A- or above for native agentic AI capabilities. Some providers, such as MongoDB, have MCP Server integration at the preview stage of development. Neon launched its MCP Server implementation in December 2024, providing a set of capabilities that can be used to manage Neon database resources via MCP. In announcing its acquisition of Neon, Databricks noted that 80% of the databases spun up on Neon’s platform in May were created automatically by AI agents, rather than humans—up from 30% when Neon launched general availability in October 2024. New data platform capabilities are required to deliver that level of automation, including instant provisioning and automated scaling of managed or serverless databases, as well as cloning and branching of data and schema to optimize continuous integration and delivery (CI/CD).
Databricks is by no means the only data platform company aware of these requirements. Other providers, including PlanetScale, SingleStore and Supabase, have focused on the development of branching as a means of enhancing developer productivity by replicating the code branching approach enabled by the Git version control development environment. Support for automated database branching is nascent, however, with 27% of providers assessed in the ISG 2025 Buyers Guide for Operational Data Platforms grading A- or above.
In addition to these new innovative features, more traditional capabilities that deliver reliability and manageability will also be critical features for agentic AI. In announcing its acquisition of Crunchy Data, for example, Snowflake emphasized not only the popularity of PostgreSQL among developers (as illustrated by the Stack Overflow Developer Survey) but also key capabilities for production deployment, including backup, high availability, disaster recovery, security and monitoring. These are all capabilities that are assessed as part of the ISG Buyers Guide methodology. I recommend that enterprises evaluating operational data platforms to support their AI development plans continue to assess usability, manageability, reliability and adaptability alongside new capabilities to address vector storage and retrieval as well as automated scaling, cloning and branching, and support for agentic frameworks and protocols.
Regards,
Matt Aslett
Fill out the form to continue reading.