Confluent’s 2026 Data Streaming Report, drawing on 4,625 IT leaders across 14 countries, argues that the AI bottleneck has shifted from model quality to data infrastructure. Rubal Sahni, AVP for India and Emerging Markets, makes the case through sharp numbers: 79% of Indian IT leaders say inadequate real-time data infrastructure is slowing AI growth, only 37% have agentic AI running in production<\/a>, and 87% of respondents report a two-to-five times return on data streaming investment. The argument is that AI strategy and real-time data strategy are now the same decision.<\/p>\n The split between enterprises that scale AI and those that stay permanently in pilot mode is increasingly traceable to one variable: whether live, trusted data reaches the model at the moment of decision. If your organization’s data still moves in batches, structured for periodic reporting rather than continuous intelligence, your AI systems are operating on stale context. That’s not a model problem. It’s a plumbing problem, and the CDO owns the pipes.<\/p>\n Sahni’s framing holds, but it’s worth naming the tilt. Confluent sells data streaming infrastructure, so its research naturally weights the infrastructure gap over, say, model selection or prompt engineering as the primary constraint. That incentive sharpens the real-time data argument while softening how much legacy batch architectures can be extended rather than replaced. The 87% ROI figure is striking, but it’s self-reported by customers of a streaming-first ecosystem, not a controlled comparison against organizations that chose a different path. Take the directional claim seriously; treat the magnitude with more skepticism.<\/p>\n The governance question around agentic AI is where this gets genuinely urgent. When AI moves from generating insights to taking actions, like adjusting pricing, routing a customer, or flagging a transaction, the quality and provenance of the data it consumes stops being an IT footnote and becomes an accountability question the board will eventually ask. Seventy-two percent of IT leaders in this survey already identify data quality as a barrier to agentic deployment. If your data governance framework was designed for compliance reporting rather than real-time decision audit trails, that gap will surface at the worst possible moment, not during a pilot.<\/p>\n The leading indicator to watch is your agentic AI production rate. If your organization is in the 63% that hasn’t moved agentic systems into production, the honest question isn’t whether your models are good enough. It’s whether you can tell, at any given second, where your enterprise data came from, how fresh it is, and who authorized its use. Vendors will keep improving models regardless. The organizations that pull ahead are the ones that can answer those three questions before the agent acts, not after.<\/p>\n Data streaming is infrastructure that moves data continuously from source to destination, the way a live news ticker updates in real time rather than delivering a printed summary each morning. Traditional batch processing collects data over a period, then processes it all at once. Streaming processes each event the moment it occurs. For AI, this distinction determines whether a model sees the world as it is or as it was an hour ago, which matters enormously when the model is making a decision rather than generating a report.<\/p>\nWhat this means for your business<\/h2>\n
Concept deep-dive: Data streaming<\/h2>\n