AI-native infrastructure: Time to end the legacy IT thinking

Imagine this: It is early 2026, and a bank in Mumbai launches a new lending product overnight. No marathon development cycles. No months of approvals. Instead, an agentic AI system, threaded across a sovereign cloud, core banking engines, and compliance models, interprets regulations, designs workflows, assesses risk, drafts documentation, and prepares a complete rollout plan before dawn.

Nearly 1,300 Km away, in Delhi, a major hospital is undergoing a similar shift. Clinical monitors and emergency devices begin streaming anomalies into a central AI hub. Before doctors arrive for morning rounds, the system has already prioritised cases, checked insurance eligibility, suggested treatment pathways, and coordinated beds and lab workflows.

While the two organisations represent contrasting worlds, what ties them together is the potential of AI-native infrastructure to transform business operations. Work once scattered across analysts, coders, and coordinators could soon be handled by reasoning agents that instantly grasp data, rules, and intent. Decisions may flow through a unified digital spine—an organism-like backbone capable of reshaping how institutions think, act, and respond.

These scenarios might feel extraordinary today, yet they echo a reality that is fast taking shape across the technology landscape. They were well-painted and discussed by major tech companies—from AWS, HPE, and NetSuite, to Oracle, SAP, Salesforce, and Snowflake—at their annual tech events during the OND quarter. Seen separately, each update feels incremental. But together, they point to a deeper change: the world is no longer adapting AI to existing IT systems—it is rebuilding those systems to be AI-native.

Three shifts capture this transition: infrastructure engineered for intelligence, data platforms purpose-built for AI reasoning, and enterprise software evolving into ecosystems of autonomous agents.

Infrastructure Learns to Think

Oracle Chairman and CTO Larry Ellison expressed the scale of this shift most vividly. At Oracle AI World 2025, he described AI as a network of “electronic brains” that will augment human capability. Oracle is building data centres with more than 450,000 NVIDIA GPUs and developing billion-watt power plants dedicated to AI workloads—projects more akin to national infrastructure than cloud expansion. The company’s Helios racks and Zettascale GPU superclusters reflect this ambition, combining extreme density, energy-efficient cooling, and Acceleron. This networking fabric merges NIC and DPU pathways into a unified, zero-trust, low-latency backbone.

HPE has taken a similarly architectural view. The company’s announcement of Grenoble AI Factory Lab with NVIDIA in France and Private AI Lab in London, developed in partnership with Carbon3.ai, reframes the data centre as a sovereign AI factory—an industrial environment for training and deploying regulated AI systems. Similarly, HPE’s networking evolution reinforces this mindset. With its Tomahawk-6-powered QFX5250 switch and an integrated Aruba–Juniper Mist fabric, networking becomes an intelligent workload-shaping layer rather than mere connectivity.

HPE CEO Antonio Neri has described this transformation as a generational shift, arguing that enterprises will increasingly rely on AI factories—sovereign, hybrid environments capable of producing and governing intelligence at scale. His vision underpins HPE’s efforts to merge accelerated computing, Ethernet-based AI fabrics, and regulated AI stacks into a unified operational model.

AWS offered the broadest demonstration of scale and ambition. Under CEO Matt Garman, the company revealed the deployment of more than one million Trainium chips worldwide and outlined an accelerated roadmap through Trainium3 Ultra and Trainium4. But the most striking leap was AWS’s AI Factories—private AWS regions installed inside a customer’s own data centre, combining sovereign control with hyperscale AI capabilities. For India’s regulated sectors, AI Factories mark a turning point: cloud-grade intelligence without data ever leaving the premises.

Infrastructure is no longer a neutral substrate buried beneath software. It is becoming a sovereign, self-optimising, and deeply intelligent foundation designed for the physics of modern AI.

The Rise of AI-Native Data Fabrics

If infrastructure is learning to think, the data layer is being re-crafted into a living fabric capable of feeding AI systems with real-time, structured context.

Ellison’s focus on Oracle’s AI Data Platform underscores this. Oracle now “vectorises” enterprise data from multiple sources, enabling retrieval-augmented generation without sacrificing confidentiality. This approach brings models to the data rather than the other way around, a crucial distinction for sectors defined by regulation and risk.

Snowflake, meanwhile, is evolving from a data warehouse into an AI Data Cloud. Snowflake Intelligence, OpenFlow, and an expanded Horizon Catalog allow employees to interact conversationally with all organisational data while ensuring governance. Thousands of AI agents deployed on the platform hint at how enterprises now expect AI to work inside their data flows, not outside them. Snowflake’s upcoming zero-copy integration with SAP Business Data Cloud will be one of 2026’s most significant milestones, enabling SAP’s semantic business data to interact seamlessly with analytical and AI systems.

SAP’s BTP is also becoming an AI-native core. With vector search, knowledge graphs, and unified governance feeding into Joule, SAP is turning ERP into an active intelligence layer capable of reasoning over business processes rather than simply recording them.

This vision was sharpened at SAP Connect in Las Vegas, where SAP unveiled a new generation of role-specific Joule assistants—more than 40 of them planned through mid-2026, each capable of collaborating across departments to automate complex decisions. SAP also introduced Business Data Cloud Connect, enabling zero-copy, bidirectional data sharing across platforms such as Databricks and Google BigQuery, reinforcing its shift toward an open, intelligence-ready data ecosystem.

This vision was sharpened at SAP Connect in Las Vegas, where SAP unveiled a new generation of role-specific Joule assistants—over 40 planned through mid-2026—to automate cross-department decisions. SAP also introduced Business Data Cloud Connect for zero-copy sharing with platforms like Databricks and Google BigQuery. As SAP Executive Board Member Muhammad Alam noted, these advances show “the power of bringing AI, data and applications together to propel smarter decisions.”

All these, while NetSuite Next extends this to mid-sized enterprises, where a unified ERP-CRM-HR data model can allow suite agents and Ask Oracle to execute tasks with native contextual understanding of workflows, rules, geographies, and compliance.

At the frontier sits AWS’s Nova Forge, which allows enterprises to infuse their proprietary data directly into the pre-training of Nova foundation models. This shift—from augmentation to core model shaping—marks an entirely new dimension in enterprise AI.

A quick analysis and review of the announcement also indicates that across vendors, there is only one emerging reality: data strategy is AI strategy. The organisations that prosper in 2026 will be those that build continuously updated, semantically rich, governance-driven data fabrics ready for reasoning systems, not those that merely accumulate data.

Platforms Become Agentic AI Ecosystems

The final shift is taking place at the platform level, where enterprise software is evolving from workflow engines into orchestrators of autonomous agents.

Salesforce’s transformation is the clearest example. With Agentforce 360, the company has embedded agents directly into customer service and sales workflows. Under its President, Chief Engineering, and Customer Success Officer Srini Tallapragada, Salesforce’s internal operations have moved from zero AI-handled cases to 1.8 million and from manual lead qualification to fully autonomous orchestration of tens of thousands of opportunities. AI is no longer an add-on; it is the operational fabric of the platform.

Similarly, SAP’s Joule functions inside the semantic structure of its applications, enabling it to take decisions and perform tasks across finance, procurement, and supply chain with procedural awareness. AWS, on the other hand, has formalised this new paradigm with AgentCore, a governed runtime where agent capabilities and constraints can be defined in natural language and continuously monitored through evaluation loops.

During his keynote at the Oracle AI World, Ellison went a step ahead, articulating how entire industries—from healthcare to agriculture—could become multi-agent ecosystems. He imagines autonomous diagnostic agents coordinating with financial and regulatory systems, drones delivering samples, and AI-designed crops reshaping agriculture. This is not software modernisation—it is a reordering of how industries operate.

2026: A New Digital Order Takes Shape

Viewed together, these developments reveal the quiet emergence of a new digital order. Infrastructure behaves like an intelligent engine, data operates as a unified context fabric, and enterprise platforms orchestrate fleets of autonomous agents. The bank and the hospital analogies are only early expressions of this shift, not anomalies.

As 2026 unfolds, telcos will deploy self-supervising networks, manufacturers will recalibrate supply chains in minutes, financial institutions will run AI-guided compliance engines, and public services will increasingly rely on sovereign AI factories.

The message is clear: the world is not simply adding AI on top of existing systems but rebuilding itself around AI—constructing infrastructure that will not only support intelligence but generate and refine it, becoming the unseen engine of the next digital transformation era.