For years, enterprises enthusiastically adopted a cloud-first approach, eager to take advantage of the flexibility and rapid innovation public cloud platforms offered. But as artificial intelligence operations transition from experiments to business imperatives, organizations are discovering that the old assumptions about where AI belongs (primarily in the public cloud) no longer hold true. AI is, in fact, becoming the great normalizer of platform selection. Its unprecedented costs, demands for low latency, and new concerns about resilience and data sovereignty are pushing enterprises to reconsider the cloud as the default choice for their most strategic workloads. Having long advocated this perspective, even as it ran counter to prevailing wisdom, I’m heartened to see industry consensus shifting in this direction.

AI workloads are different from traditional applications in many ways, most notably their hunger for compute and data. Running large-scale training or inference jobs in the public cloud can quickly lead to skyrocketing costs, sometimes outstripping what comparable on-premises infrastructure would demand. According to Deloitte, “some enterprises are seeing monthly [cloud] bills in the tens of millions,” with costs often surpassing 60% to 70% of the total cost to acquire and maintain equivalent on-premises systems. That kind of economic pressure openly challenges the cloud-first edict, particularly when predictability and budget control are required.

Moreover, AI’s need for ultra-low latency to support operations such as real-time decision-making and its requirement for reliable, always-on performance in mission-critical environments often make on-premises solutions more viable. The same Deloitte report observes that “applications requiring response times of 10 milliseconds or below cannot tolerate the inherent delays of cloud-based processing.”

Conventional, not controversial

It wasn’t so long ago that expressing skepticism about cloud-only strategies drew resistance. For the better part of a decade, cloud held near-religious status in enterprise strategy meetings. I recall regularly making the case for hybrid cloud, influenced not by nostalgia but by a pragmatic understanding of the architectural realities many organizations faced. Today, it’s clear that more decision-makers and respected consulting voices are coming to similar conclusions.

Notably, Deloitte’s analysis now explicitly recommends a “three-tier approach” that combines cloud for elasticity, on-premises for consistency in production workloads, and edge deployments for immediate or ultra-low-latency AI needs. This is architecture tailored for a world where AI—not generic IT workloads—drives technology decisions. ZDNet, summarizing these industry movements, notes that “cloud-first strategies can’t handle AI economics,” and forward-looking companies are now “contemplating a shift away from mainly cloud to a hybrid mix of cloud and on-premises.”

AI as the platform equalizer

The radical resource requirements and operational constraints of AI have stripped away much of the mythology that cloud is best for everything or that on-premises is purely legacy. Every choice is now fundamentally workload-driven: AI must go where it can run most cost-effectively, safely, and responsively. Sometimes that’s in the cloud; just as often, it’s in non-cloud deployments or a combination thereof.

This new reality is forcing organizations to undertake careful assessments before making platform decisions for AI. The days when IT leaders could simply sign off on wholesale cloud migrations, confident it was always the most strategic choice, are over. In the age of AI, the optimal approach is usually hybrid.

Having openly championed this hybrid path even when it was unpopular, I welcome the growing acceptance of these ideas among decision-makers and industry analysts. Enterprises now have a rationale for an integrated, best-of-both-worlds platform, not as a retreat from cloud, but as a progression toward mature, sustainable AI. Hybrid approaches allow organizations to optimize costs, address regulatory and latency needs, and retain vital security controls, while also continuing to leverage the cloud’s strengths for experimentation and growth.

Artificial intelligence, with its intense resource demands and complex risk profile, has normalized a pragmatic approach to platform architecture. Ignore the rhetoric. Hybrid is the future for organizations that intend to scale AI, strike the right cost-performance balance, and adapt to ever-changing requirements. With authoritative references both validating and reinforcing this view, it’s clear that what was once dissent is now fast becoming conventional wisdom. I’m pleased to witness this change. Proof that the path to successful AI lies not in cloud-only solutions, but in the thoughtful combination of cloud and non-cloud strategies.