The first wave of enterprise AI was a land grab, as firms raced to bolt frontier models onto every workflow. Industry experts see the next wave favors custom tools trained on proprietary data giving rise to AI verticalization, which is reshaping how enterprises build, buy and deploy artificial intelligence.
They see companies moving away from general-purpose frontier models toward purpose-built tools fine-tuned on their own data, driven by data sovereignty, compliance demands and the need for domain-specific accuracy.
According to a 2025 survey, four in five companies are now developing their own customized AI solutions. The Nutanix Enterprise Cloud Index (ECI) report underscores that the problem is especially pronounced for firms in highly regulated fields like healthcare, finance and public services, where data sovereignty and compliance constraints make generic AI tools impractical.
AI verticalization is the practice of customizing artificial intelligence models for specific industries or business functions by fine-tuning them on proprietary, domain-specific data, so the tools produce more accurate and compliant results than general-purpose models.
Rob Enderle, principal analyst at the Enderle Group, puts the current moment in sharp relief.
"We are just leaving the hype cycle phase of the movement," Enderle told The Forecast.
"Enterprises are quickly realizing that general-purpose frontier AI models are too noisy and prone to hallucination for mission-critical workflows."
The shift toward verticalization is still in its early innings, Enderle said, but the push to adopt more internally crafted solutions is inevitable and accelerating as that realization spreads across industries.
A general-purpose AI model remains broadly capable thanks to optimization for diverse tasks and design for accessibility across a wide range of users. But enterprises are discovering that those same properties tend to limit a model's usefulness in detailed, high-stakes, domain-specific contexts. The shift now underway favors purpose-built models, fine-tuned on industry-specific data, optimized for the decisions that actually matter inside a given business.
Companies "don't need a frontier model that can quote 'War and Peace' or extrapolate haikus for different styles," Lynn Comp, head of global sales and go-to-market and vice president of Intel's AI Center of Excellence, told The Forecast.
"They need something that is relevant to their industry that takes their specific data and can essentially make faster decisions with it."
What Does AI Verticalization Mean in Practice?
The term encompasses a spectrum of approaches to AI design and development, and the distinctions matter for how enterprises approach procurement and ownership. At one end is retrieval-augmented generation (RAG), which involves taking an existing foundation model and connecting it to a curated, domain-specific knowledge base so its outputs draw on an organization's own documents, policies and data without retraining the underlying model. At the other end is full fine-tuning: training a model on proprietary datasets that encode domain expertise, industry-specific terminology and organization-specific decision logic into the tool.
Between those poles sits a growing ecosystem of approaches. That includes parameter-efficient fine-tuning methods that allow companies to adapt large models at a fraction of the cost of full retraining, domain-specific foundation models trained on curated vertical datasets, and hybrid architectures that pair lightweight specialized models for domain reasoning with larger, more generalized models for language generation.
What unites all of these approaches is a common recognition: the competitive value of AI inside an enterprise does not come from the model's generalized capabilities. It comes from how strongly the model understands the specific context and real-world environment in which it operates. A model that has ingested a healthcare system's clinical protocols, patient population data and physician documentation patterns is not just marginally better than a general model at performing clinical support tasks. It operates on an entirely different level.
What's Fueling the Push Toward Custom AI Solutions?
Ask practitioners what is prompting enterprises to embrace verticalized AI solutions, and the answer is rarely the one that appears in vendor marketing materials. Enderle is direct about it.
"The issue comes down to data sovereignty and security," Enderle said.
"Cost and performance are nice side benefits, but keeping proprietary data behind the firewall and avoiding massive compliance liabilities are what's actually driving the bus here."
He sees financial services and healthcare leading other sectors.
"They have the strictest regulations and the budgets to match,” Enderle said. “Their lead proves that if you don't solve governance and security challenges as a business first, you can't scale AI in the enterprise."
These firms are not early adopters chasing new capabilities, Enderle said. They are regulated multinational entities responding to hard constraints on where data can travel and who can access it, while also weighing accountability and corporate governance concerns.
The sovereignty issue is reshaping infrastructure decisions in ways that go beyond software procurement. Enderle describes "a return to on-prem and hybrid" IT management as a defining trend.
"Sovereignty is forcing regulated companies away from public cloud APIs," he said.
"It's also driving a massive hardware refresh cycle for localized AI data center clusters and AI PCs to process data strictly within the corporate firewall."
Sending sensitive customer or patient data to a third-party model API is, in many enterprises and regulatory contexts, not practical or permissible. Given those constraints, verticalization is not a competitive choice for these organizations. It is a compliance requirement.
How Is Procurement Changing?
For IT procurement leaders and chief information officers, AI verticalization represents a fundamental change in how smart technology is evaluated and sourced. The benchmark-driven buying logic of the first wave of AI adoption is giving way to something harder to assess but arguably more consequential, as IT leaders ask: Which model performs best on the enterprise's dedicated tasks, with its data, in its regulatory environment?
Enderle identifies the questions that should anchor any serious evaluation of whether to create a custom AI solution, and they are not the ones most buyers are asking.
"Is our data actually clean enough for this?,” Enderle asked. “Who holds the liability when the model hallucinates a bad decision? And what is our exit strategy if this vendor pivots or goes under?"
Such changes in thinking also reshape the competitive landscape of vendors that IT leaders can rely on. A new tier of specialized model providers has emerged, offering foundation models pre-trained on legal text, clinical literature, financial documents or industrial sensor data, on top of which enterprises can apply their own fine-tuning with less financial investment.
The insurance industry offers a useful early case study. Carriers are deploying context-sensitive models fine-tuned on years of claims data, policy language and underwriting decisions that encode the precise risk logic their businesses run on. Such models do not just deliver detailed decisions. They can also be queried about the reasoning behind any given recommendation in ways that satisfy both internal audit requirements and external regulatory scrutiny.
Who Owns Custom AI, and What Do the Rights Entail?
Growing verticalization also changes the politics and economics of AI tool ownership in ways that enterprises are only beginning to comprehend. A company that fine-tunes a model on its own proprietary data occupies a meaningfully different position than one drawing on a generalized API, Enderly explained. Fine-tuned weights encode knowledge specific to that organization, its terminology, its decision patterns and its domain expertise, that does not exist anywhere else and cannot be replicated by a competitor subscribing to the same foundation model.
Enderle urges caution about overstating the durability of that asset.
"You own the fine-tuned weights, but if your solutions are built on a proprietary base model, you're building on rented land," he said. "It's not a highly durable strategy. You will have to continuously retrain and update as base models evolve."
Such a competitive advantage can be real, Enderle said, but it can also be perishable. Maintaining it requires ongoing investment in retraining pipelines, updating data infrastructure and keeping the AI model current as both the underlying technology and the business contexts shift.
Concerns about being locked into a single vendor's ecosystem also affect the question of ownership.
"Lock-in is a massive risk," Enderle said. "Your defense is architectural abstraction. Enterprises need to use open-source models where practical and build containerized deployments so they can port their workloads to different infrastructure when vendors inevitably hike prices."
The organizations most thoughtfully dealing with such challenges are treating model portability as a first-order architectural requirement from the start of their vertical AI programs, he said.
Comp's framing of the ownership opportunity remains instructive. Owning your own intelligence means being "forced individually and as a company to actually write down the process by which we come to answers, and even challenge ourselves on why do I have that judgment call, and what's the basis of that judgment call," she said.
The process of fine-tuning a model that defines how a company makes decisions is a discipline that compounds in value over time, Comp said. Mapping out such efforts makes the organization's reasoning more explicit, more auditable and more defensible.
Will Fragmentation Undermine the Vertical AI Vision?
An obvious concern about AI verticalization is fragmentation: a proliferation of narrow models, each optimized for a specific context, that creates more management complexity, interoperability challenges and a wide range of integration tools for enterprise IT.
"The fragmentation risk is incredibly valid," Enderle said bluntly. "It's like the SaaS sprawl problem all over again, but with higher computing costs."
Forward-looking organizations are taking steps to address the concern architecturally rather than hoping it resolves itself, Enderle said.
"Leading CIOs are handling the issue by deploying AI gateways and unified management layers to centralize auditing, access and security across a fragmented model ecosystem," he said.
That governance layer sits above the individual models, providing a single plane for access control, cost management and audit logging regardless of how many specialized models sit underneath it.
Comp is more sanguine about diversification.
"I know a lot of people look at the issue like fragmentation, but it's such an interesting technology," she said.
"I think this type of fragmentation is necessary for us to achieve the promise in AI capabilities everywhere."
Such a wide diversity of purpose-built AI models is a feature, not a defect, of a maturing ecosystem, Comp said. It is analogous to how the proliferation of specialized databases and microservices created short-term integration complexity for IT leaders but ultimately enabled more capable and more resilient enterprise architectures than any monolithic alternative could have delivered.
Is Enterprise Data Ready for Vertical AI?
Before any discussion of model architecture, vendor selection or governance frameworks, Enderle argues there is a more fundamental question most enterprises are not asking: Is your corporate data actually ready? Any information that an AI model uses must be clean and categorized, he said. If not, AI, which acts as a force multiplier, may prove a hiccup whose concerns can quickly compound.
"Garbage in, garbage out at scale can be a challenge here," Enderle said.
"The biggest failure that enterprises experience with custom AI is feeding models siloed, improperly permissioned or legacy data. If you don't fix basic data hygiene, metadata tagging and access controls, fine-tuning will just automate your existing dysfunctions."
The enterprises that have most successfully and rapidly adopted vertical AI are, almost without exception, the ones that invested in data readiness before they invested in model capability. It is essential to treat the underlying data infrastructure as a strategic asset, and the model as the beneficiary of that investment.
What Does Verticalization Mean for the Future of Business?
The implications for competitive strategy are significant, and for the enterprises moving most rapidly to embrace verticalized AI solutions, they are already visible. Achieving a first-mover advantage is no longer about having access to the most advanced AI model. It is about having access to the most domain-specific one, the model most extensively trained on proprietary data that competitors do not have access to.
Enderle's view on whether advancing frontier models will eventually render verticalization less necessary is unambiguous.
"Domain specificity is the durable moat for enterprises here," Enderle said.
"Frontier AI models will continue to get smarter, but businesses will always require walled, highly specialized models to protect their IP and ensure absolute accuracy in their specific operational verticals. General models are for proof of concept; vertical models are for execution."
The organizations that understand this are investing accordingly, not just in model capability but in the data pipelines, labeling infrastructure and fine-tuned workflows that turn proprietary data into proprietary AI. Such firms are treating verticalization not as a technology upgrade but as a strategic capability-building program, applying the same long-term investment logic they would apply to any other durable competitive asset.
Frontier models will keep advancing and benchmark scores will keep climbing. But the enterprises that will extract the greatest, most long-lasting value from AI going forward are those that have stopped asking which model is the most powerful in general. They will be the ones that ask which AI model is the most useful for the specific decisions they make day in and day out, and on which the business fundamentally relies.
Related:
Scott Steinberg is a business strategist, award-winning professional speaker, trend expert and futurist. He’s the bestselling author of Think Like a Futurist; Make Change Work for You: 10 Ways to Future-Proof Yourself, Fearlessly Innovate, and Succeed Despite Uncertainty; and Fast >> Forward: How to Turbo-Charge Business, Sales, and Career Growth. Find him at www.FuturistsSpeakers.com and LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
