By Poulomi Chatterjee
As enterprises head into the new year, artificial intelligence is expected to move out of its experimental phase into a more measured, execution-heavy cycle. After inflated expectations around agentic AI and autonomous workflows in 2025, companies would recalibrate around cost, security, governance and returns. Experts FE spoke to said that this year is likely to be defined less by radical breakthroughs and more by how organisations industrialise AI at scale without destabilising core operations.
Agentic AI likely to slowdown:
The promise of multi-agent systems replacing human-led workflows has collided with practical realities. Deployments have proved expensive, disruptive and fraught with security risks, particularly in regulated sectors such as banking and healthcare. Autonomous agents can expose enterprises to data leakage, corruption and deletion, while prompt injection attacks on AI browsers and workflows have added to risk concerns. As a result, enterprises are expected to prioritise agent observability over rapid rollout, focusing on monitoring execution, detecting threats and tracking data flows.
AI governance to become an enterprise mandate:
The year is likely to see formal AI governance frameworks move from policy documents into operating practice. Enterprises are assigning responsibility for AI outcomes, defining escalation paths for failures, and embedding controls into workflows rather than retrofitting them. This shift reflects a growing recognition that AI systems behave more like infrastructure than software features, requiring continuous oversight, audits and risk management comparable to core IT systems.
Data centre expansion to turn selective:
AI-driven demand for data centres continues, but the pace is becoming more uneven. A recent report by KPMG projected India’s installed data centre capacity to cross 2 GW in 2026, up from just over 1 GW currently, signalling sustained investment interest. However, signs of restraint are visible globally. Oracle has reportedly delayed some buildouts, highlighting concerns around financing, power availability and construction timelines. According to Bain & Company, hyperscalers are shifting from an early scramble to a more disciplined, execution-focused phase centred on service delivery rather than raw capacity addition.
Monetisation pressure to reshape AI platforms:
The cost of running large-scale AI models is pushing platform companies to explore new revenue streams. OpenAI, which lacks the advertising backstop of rivals such as Google and Meta, is reportedly considering introducing sponsored content on ChatGPT. For enterprises, this would raise new questions around neutrality, data use and commercial influence within AI-driven decision-making tools that are increasingly embedded in workflows.
Budgets to tighten, smaller models will gain ground:
Enterprise AI spending is coming under closer scrutiny, with boards demanding clearer links between investment and productivity gains. This is accelerating interest in smaller, task-specific and open-source models that are cheaper to run and easier to control than large foundational models. For India, these alternatives are particularly relevant given constraints around compute costs, GPU availability and fragmented local-language data, as highlighted by executives at EY.
India’s enterprise AI stack will deepen:
India is expected to see stronger momentum in enterprise applications and AI-led services rather than frontier model development. Industry body Nasscom has pointed to continued growth in global capability centres as a stabilising force for tech employment, even as entry-level roles will remain under pressure. On the funding side, investors such as Antler expect capital to remain concentrated in enterprise applications, AI-enabled services and consumer-facing use cases, with seed-stage activity staying healthy. Together, these trends suggest that the year ahead will be less about AI spectacle and more about making the technology work, securely and sustainably, inside enterprises.