Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Anil Inamdar is Global Head of Data Services at NetApp Instaclustr
Enterprises are crossing a critical threshold in 2026 as they move AI from experimentation into production at scale. Those pulling ahead will be the ones making deliberate choices about freedom, composability, and the fundamentals of data infrastructure.
PREDICTION 1: Kafka Optimization Will Replace AI Streaming Platform Purchases
Enterprise tech leaders are being pitched specialized “AI workflow platforms” almost daily, promising to revolutionize data streaming with purpose-built pipelines and intelligent orchestration. The reality is simpler. The fully open source version of Apache Kafka already provides the streaming backbone that most AI systems require, with performance and cost benefits that purpose-built platforms cannot match.
The fundamental challenge these platforms claim to solve (feeding AI models stale data) is exactly what Kafka was designed to prevent. Real-time recommendation engines need current user behavior, and fraud detection systems must respond to threats as they happen. Well, Kafka’s real-time streaming capabilities solve this at scale, handling the continuous data flows that keep AI systems accurate and responsive.
Before writing checks for “AI-native” streaming platforms, I think more enterprises in the coming year will be auditing what their existing Kafka infrastructure already delivers. Kafka processes real-time data ingestion with latencies as low as 2 milliseconds, handles real-time transformations through Kafka Streams, and scales to trillions of messages daily. Purpose-built AI platforms essentially rebuild these proven patterns with more constraints and less performance. Most run on Kafka anyway, adding abstraction layers that increase latency.
PREDICTION 2: Composable Architectures on Open Foundations Will Outpace Proprietary AI Platforms
The winning technical playbook in 2026 for agentic AI will center on composable architectures built from established open source components (Kubernetes, Postgres with vector capabilities, Kafka etc) rather than proprietary “agentic-AI-ready” platforms.
Engineering teams will deploy AI agents faster and more cost-effectively by making targeted upgrades to their current stacks. The playbook is there for the taking: integrate pgvector with Postgres for vector search, utilize Apache Cassandra 5.0’s native vector indexing for hybrid workloads, deploy OpenSearch for semantic search capabilities. The competitive edge comes from mastering the essential but often overlooked aspects (streaming patterns, retrieval optimization, zero-trust security implementations) within these proven open source systems.
As agentic AI becomes production-ready, organizations maintaining control over their data and avoiding vendor lock-in will have the flexibility to evolve their AI agents outside of the lock-in constraints of proprietary ecosystems. This incremental enhancement strategy will distinguish the leaders from those who invested in expensive platform overhauls that ultimately limit future adaptability.
PREDICTION 3: Open Source Will Define the Enterprise AI Data Stack
AI is fundamentally reshaping what “enterprise-ready” means for data infrastructure, and open source is winning that transformation. Postgres, Cassandra, Redis, OpenSearch, and ClickHouse are replacing proprietary databases because they’re AI-extensible, cloud-neutral, cost-efficient, and community-driven.
The shift will accelerate across multiple dimensions, simultaneously, in 2026. AI demands multi-modal capabilities, including vector, graph, and semantic stores that provide the context-rich retrieval LLMs need, not just rows and columns. Tools like Postgres with pgvector and Cassandra with native vector indexing are becoming mandatory, not optional.
Real-time is becoming standard, not special. Kafka-driven streaming-first architectures power agents, automations, and analytics that can’t wait for batch processing. Event-driven workloads require always-on, globally distributed stores, where Cassandra continues to dominate.
Governance is getting automated because it has to. Enterprises are blending metadata engines with systems like OpenSearch and OpenLineage to automate tagging, PII detection, and access monitoring. Critically, AI models are only as safe as the data they consume, and manual governance can’t keep pace.
The throughline here is that 2026 will reveal that AI readiness isn’t about buying tons of new shiny platforms. Rather, viability is a question of strategic freedom, incremental intelligence, and open foundations. The organizations with the best success stories will be those who master the fundamentals while preserving flexibility to evolve. They’ll optimize the flexible and scalable infrastructure they already have, build on open architectures that prevent lock-in, and recognize that the most sophisticated AI strategy often looks like smart infrastructure choices, not revolutionary overhauls.
##
ABOUT THE AUTHOR
Anil Inamdar is Global Head of Data Services at NetApp Instaclustr, which provides a managed platform around open source data technologies including Cassandra, Kafka, Postgres, ClickHouse, and OpenSearch. Anil has 20+ years of experience in data and analytics roles. Joining Instaclustr in 2019, he works with organizations to drive successful data-centric digital transformations via the right cultural, operational, architectural, and technological roadmaps. Before Instaclustr, he held data & analytics leadership roles at Dell EMC, Accenture, and Visa. Anil lives and works in the Bay Area.
