Getting enterprise AI right means making simultaneous decisions across silicon, workload placement, data governance, and network architecture. Most organizations are still learning what the stack actually demands, says Ron Westfall, vice president of Infrastructure and Networking at HyperFRAME Research.
Westfall has a front-row seat for that scramble. He sees companies piloting AI deployments that ultimately stumble and fall. A 2025 MIT study reported 95% of enterprise generative AI pilots are failing to reach production, but things are evolving quickly, although not always in the right direction. The 2026 Enterprise Cloud Index shows 79% of IT leaders encounter unauthorized AI deployments, and this familiar pattern of Shadow IT puts them at risk.
Turn to trusted advisors before making critical AI deployment decisions, advises Westfall. Early missteps of building AI stacks tend to compound, leading to costly operational challenges, he said.
“The AI stack is distinct from previous network implementations,” Westfall told The Forecast during Nutanix’s recent NEXT 2026 event in Chicago.
“It has added intricacies, and you want to minimize them, so they don’t become overwhelmingly complex.”
IT teams are feeling overwhelmed, according to The HyperFRAME Research Lens: The State of the AI Stack. Reporting on the first half of 2026 survey findings from 544 qualified enterprises show that only 14% classify their core data architecture as fully modernized for AI workloads today, while 23% are still running a legacy on-premises data warehouse.
There’s much work to be done.
A More Complex Stack
The AI stack poses specific challenges for organizations. The decisions it demands span multiple layers simultaneously, with each layer potentially affecting others.
Silicon is one example. Westfall said organizations must now choose among CPUs (central processing units, the traditional workhorses of enterprise computing), GPUs (graphics processing units, the dominant engines for AI training and inference), or a combination of the two. That decision touches every other layer of the stack.
Workload placement is another example. Westfall said organizations are settling into a hybrid approach where sensitive or regulated workloads stay on-premises, legal mandates require direct control, and public-facing use cases move to the cloud. In fact, 97% of senior decision-makers surveyed by NTT DATA believe critical workloads are best kept in private or on-premises environments, with less-sensitive tasks handled elsewhere.
Distributing workloads that way puts pressure on the rest of the stack.
“The AI stack is driving more awareness of how the network has to work with the other layers of the stack,” Westfall said. “That includes the data layer, the cloud layer, the orchestration layer.”
Managing Data Sovereignty
Before any of those infrastructure decisions pay off, the data feeding the AI has to be right. Westfall was direct.
“The quality of AI outputs depends on the quality of what goes in,” he said. Organizations that skip the data fundamentals will feel it downstream.
The deeper issue, he said, is data sovereignty, an organization’s ability to govern its data before AI gets layered on top. It covers more ground than most enterprises realize.
That includes knowing where sensitive data resides, who can access it, how it moves across systems, and whether employees are introducing unsanctioned AI tools into workflows. Cybersecurity is part of the equation, along with shadow AI, the practice of employees or partners using AI tools without IT oversight. When that happens, sensitive company data and intellectual property can flow into public large language models (LLMs) unnoticed. It’s already a significant issue, with 69% of organizations suspecting or confirming employees are using prohibited public generative AI tools at work, according to Gartner.
Westfall said organizations need to get ahead of shadow AI or risk increased security and operational vulnerabilities. Without proper oversight, the exposure risk grows every time someone on the team introduces an unauthorized tool, he noted.
AI observability and governance platforms offer one way to address the problem by giving CIOs greater visibility into how employees use AI tools, what data is shared, and whether sensitive information leaves approved environments. These platforms can help organizations detect unauthorized applications, monitor prompts and outputs for risky behavior, enforce data handling policies, and identify potential security or compliance violations before they escalate.
The technology is also becoming increasingly important for regulatory compliance. Westfall said organizations operating in the EU already face growing pressure to demonstrate that AI systems and data practices comply with government mandates around transparency, privacy, and risk management. Observability platforms can help by creating audit trails, documenting AI usage, tracking data lineage, and providing reporting capabilities that support compliance efforts.
Organizations are paying close attention: 57% of C-suite leaders in an EY survey identified non-compliance with AI regulations as a top AI-related risk.
“In the EU, we’re seeing the best first steps being taken that could actually end up as best practices for other parts of the world to emulate,” he said, referring to the EU AI Act.
An Advisor Advantage
Getting ahead of those requirements takes expertise that most organizations are still building, which is where trusted advisors add value. Westfall called this the “lead integrator role,” referring to the coordination of decisions across interconnected AI stacks.
Effective advisor relationships provide a consistent management layer that runs AI workloads across on-premises and cloud environments from a single control plane, Westfall said. This keeps hardware decisions flexible as the stack evolves. A software-defined foundation enables organizations to add capabilities without having to rebuild from scratch as requirements change, he added.
Advisors can also help organizations navigate the AI partner ecosystem itself. The vendors covering model operations, governance, data services, and inference are multiplying fast, and most enterprises don't have the bandwidth to evaluate them. A trusted advisor can provide a vetted view of what's available and what works. In fact, AI implementations built on external partnerships succeeded at roughly twice the rate of those developed internally, according to the same MIT study.
Recognizing this, industry leaders have jumped into the fray with AI service offerings. In May 2026, Anthropic launched an enterprise AI services company backed by Blackstone, Hellman & Friedman, and Goldman Sachs to help organizations move AI from proof of concept to production. Around the same time, OpenAI launched its Deployment Company, a roughly $4 billion initiative with forward-deployed engineers aimed at the same problem. Nutanix similarly expanded its professional services for AI/ML deployments and created a certified AI Partner Program designed to guide organizations through exactly those stack decisions.
“You really do need a trusted advisor to help streamline and simplify the tools needed to make this work,” Westfall said. Overall, this architectural friction reveals that enterprise AI maturity is no longer hindered by model capability, but rather by an operational gap where fragmented infrastructure, unmanaged shadow AI, and compliance mandates compel IT leaders to trade lone-wolf buildouts for ecosystem integration.
David Rand is a business and technology reporter whose work has appeared in major publications around the world. He specializes in spotting and digging into what’s coming next–and helping executives in organizations of all sizes know what to do about it.
Ken Kaplan contributed to this story. He is Editor in Chief for The Forecast by Nutanix. Find him on X @kenekaplan and LinkedIn.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
