What Nutanix's Agentic AI Stack Means for Financial Services

Nutanix has introduced the Nutanix Agentic Al solution, designed to help enterprises build, run, and govern Al workloads at scale. The announcement reflects a shift financial services leaders have anticipated: the primary bottleneck has moved from algorithmic capability to architectural capability. While models have become sufficiently intelligent, the conversation now centers on how to run autonomous agents in production without being hindered by brittle data foundations that were originally designed for an earlier era.

That shift matters. Banking, financial services, and insurance (BFSI) firms are wrestling with a scaling challenge that better models alone do not solve. Before the industry can move from chatbots to autonomous reasoning engines, it has to escape what is increasingly being called 'pilot purgatory', the gap between launched AI pilots and durable production deployments.

So why are 95% of AI pilots stuck?

The headline numbers look impressive. By late 2025, 85% of global banks had integrated some form of AI into their operations, and the financial sector is projected to invest $97 billion in AI infrastructure by 2027. But beneath the adoption figures sits a sobering reality: MIT research suggests up to 95% of AI pilots never reach production. Capgemini’s 2026 research adds context: 43% of corporate and investment bank IT budgets are still consumed maintaining legacy systems, while only 29% goes toward transformative technologies.

The root cause is rarely a lack of ambition. More often, it is infrastructure built for an earlier era. Financial institutions are attempting to run autonomous agents that support real‑time credit decisioning, continuous risk monitoring, and security operations at machine speed on systems originally designed for batch processing and human‑initiated queries.

• The result is a compounding set of barriers, including:
  Legacy technical debt that struggles to support modern AI compute demands
• Critical data trapped in silos and difficult to move across environments
• Exploding token volumes are making AI more expensive to run
• Agent sprawl, with machine identities now outnumbering human ones 96:1, often with inadequate controls
• A tightening regulatory landscape that increasingly governs where data must reside (data sovereignty) and how AI‑driven decisions must be documented

Other risks are also emerging as the market matures. Just as the Digital Operational Resilience Act (DORA) elevated attention on cloud concentration risk, regulators are beginning to focus on its AI equivalent: model concentration risk.

The release of Anthropic™ Claude Mythos Preview model, designed to identify vulnerabilities in critical software and infrastructure code, has rattled the industry and further reinforces how quickly AI governance is converging with operational resilience mandates. For many BFSI CIOs, the challenge is increasingly centered on how to permission, audit, and contain AI agents at enterprise scale.

This is not a pilot problem. It’s an infrastructure problem.

Introducing a turnkey path: Nutanix Agentic AI

To move from the lab to the banking branch office or insurance headquarters, the Nutanix Agent Gateway solution, now generally available as part of Nutanix Enterprise AI 2.7, is designed to provide a turnkey way to help manage the architectural chaos BFSI IT teams often face and aims to deliver a validated, full-stack environment for AI and agents that can support predictable scaling.

It adds an orchestration layer for fleets of enterprise Al agents, designed to provide builders shared access to models, tools, and GPUs while helping platform teams maintain governance over security, performance, and token costs. This is engineered to establish a common foundation where experimentation, deployment, and lifecycle management can be managed through one consistent control plane rather than a sprawl of bespoke projects. 

For the agent-sprawl problem specifically, this is where governance becomes concrete: granular access policies for the Model Context Protocol servers that agents use to reach business tools, centralized token observability across model vendors, and audit logging of agent requests for compliance.

The Nutanix Enterprise AI  solution then brings a cloud operating model to enterprise Al, allowing GPU, storage, and data services to be unified into a shared, governed pool that teams can consume on demand instead of building one‑off stacks. It lets enterprises run any mix of private and cloud‑hosted models, RAG, agents, and classic inference through a single secure endpoint and dashboard, with builtin guardrails for security, governance, and cost intended to help Al scale while mitigating operational chaos or vendor lock-in.

This is the kind of architecture our BFSI AI whitepaper develops in depth, under what is increasingly being called in the industry as the Trusted Stack.

The hybrid reality: infrastructure as the moat

While much of the AI conversation centers on which Large Language Model is the "smartest" this week, in the high-stakes world of banking, capital markets, and insurance the model may not be the moat. Your infrastructure may be.

This is the conclusion a growing body of strategic analysis is reaching. As McKinsey recently put it, when everyone has access to the same models, advantage goes to those who turn cognitive work into infrastructure: the data pipelines, governed models, and integrated workflows that scale at low incremental cost and that competitors cannot easily copy. For BFSI institutions, that infrastructure is where placement, governance, and resilience decisions actually get made.

A "Cloud Smart" approach treats placement as a strategic decision rather than a default. Public cloud is well-suited to training and experimentation, where elasticity matters most. And  inference, where proprietary data meets production workloads under regulatory scrutiny, may belong closer to the institution, on-premises or in a hybrid core.  The Nutanix Cloud Platform solution is designed to allow for that fluidity, providing institutions a path to deploy across environments rather than committing to a single vendor or model stack.

Three principles going forward

As your institution evaluates its AI roadmap, keep these in mind. Notably, each maps to a distinct competitive moat that strategic analysts identify as durable and hard to replicate:

• Trust is the primary currency. Any AI application that lacks auditability by design can be a regulatory liability. Governance is not easily bolted on after a pilot; it is often best to be engineered into the foundational stack from day one.
• Resilience and sovereignty are non‑negotiable. With 79% of BFSI IT leaders prioritizing data sovereignty, infrastructure that supports air‑gapped operations, localized data residency, and survivability under stress is no longer optional. It is a prerequisite for production‑grade AI.
• Modernize the core to scale the edge. You cannot run a 21st‑century reasoning engine on a 20th‑century substrate. Legacy modernization is not just about reducing maintenance costs; it is about creating the headroom for AI to function securely and reliably across supported environments.

Is your current infrastructure ready for what comes next?

Go deeper: Engineering the AI Trusted Stack for Financial Services.