The ‘All Cloud’ AI Strategy Is Dead — 75% of Companies Are Going Hybrid by 2028 [Forecast]

What Happened

A convergence of analyst forecasts from IDC, Deloitte, and Gartner points to a single conclusion: the era of running all enterprise AI workloads in the public cloud is ending. IDC projects that by 2028, 75% of enterprise AI workloads will be deployed on fit-for-purpose hybrid infrastructure — a mix of public cloud, private data centers, and edge computing environments selected based on the specific requirements of each AI task.

Deloitte’s enterprise AI infrastructure survey, conducted in December 2025 with 515 US leaders from enterprises with over $500 million in annual revenue, found that over 70% expect to scale both “AI factory” and “AI at the edge” deployments by 2028. That would roughly double current adoption levels in three years. Today, 36% of respondents have scaled AI at the edge; by 2028, 72% expect to reach that milestone.

Separately, Gartner predicts that by 2028, over 40% of leading enterprises will have adopted hybrid computing paradigm architectures into critical business workflows, up from approximately 8% today.

Why It Matters

The shift reflects a hard economic reality. As companies move from AI experimentation to production-scale deployment, the cost profile of cloud-only infrastructure becomes increasingly difficult to justify. Deloitte’s research identifies a specific tipping point: when cloud costs reach 60% to 70% of equivalent hardware costs, enterprises should seriously evaluate alternatives including colocation providers and managed service providers.

The math is straightforward. Training large AI models requires burst capacity that cloud providers handle well — spinning up hundreds of GPUs for weeks, then releasing them. But production inference — running trained models to serve predictions at scale — generates predictable, continuous compute demand. Paying cloud premium pricing for that steady-state workload erodes margins over time.

According to the Deloitte survey, 24% of respondents said they plan to move off the cloud when costs reach 25% to 50% of the relative cost of alternatives. Additionally, 78% of respondents anticipate their organizations will boost use of edge technology in the next 12 months, reflecting efforts to manage computing costs as AI scales up.

Technical Details

Leading organizations are responding by adopting what Deloitte describes as a three-tier hybrid architecture. The first tier is public cloud, used for elastic training workloads, burst capacity needs, and experimentation phases. The second tier is private or on-premises infrastructure, running production inference at predictable costs for high-volume, continuous workloads. The third tier is edge computing, handling time-critical decision-making where latency requirements rule out round trips to centralized data centers.

Each tier addresses different constraints. Cloud excels at variable workloads where the ability to scale up and down matters more than per-unit cost. On-premises infrastructure wins when utilization rates are consistently high — above 60-70% — making the capital expenditure worthwhile. Edge deployments are driven by physics: some AI decisions, particularly in manufacturing, autonomous systems, and real-time video analysis, must happen in milliseconds.

Industry analysts project that AI infrastructure spending will surpass $200 billion annually by 2028. A significant portion of that investment will go toward building or expanding private AI compute facilities. A DataBank survey found that 76% of enterprises plan geographic expansion of their AI infrastructure, while 53% are actively adding colocation to their deployment strategies.

Who’s Affected

The hybrid shift has the most immediate implications for organizations that have gone all-in on public cloud for AI. Companies running large-scale inference workloads — customer-facing recommendation engines, real-time fraud detection, content generation systems — face the starkest cost pressure. For these workloads, the cloud premium over equivalent on-premises compute compounds month after month.

Cloud providers themselves are adapting. AWS, Azure, and Google Cloud have all introduced on-premises and edge extensions of their platforms, recognizing that retaining enterprise customers means meeting them where their workloads need to run, not insisting everything stays in the public cloud.

Infrastructure teams at enterprises with $500 million or more in revenue are most affected by this transition. The Deloitte survey respondents represent this tier, and their investment plans suggest a significant rebalancing of IT budgets toward private and edge infrastructure over the next two to three years.

Smaller organizations with lighter AI workloads may continue to find cloud-only approaches cost-effective. The hybrid calculus changes primarily at scale, where the volume of inference requests justifies the capital investment in dedicated hardware.

What’s Next

The transition to hybrid AI infrastructure will accelerate through 2026 and 2027 as more enterprises move past pilot projects into production deployment. Deloitte’s survey indicates that the gap between current adoption and 2028 expectations is widest for edge AI, suggesting that edge infrastructure buildout will be the fastest-growing segment.

Organizations planning their AI infrastructure strategy should start by profiling their workloads. Training runs with variable demand belong in the cloud. High-volume inference with predictable patterns is a candidate for on-premises or colocation. Latency-sensitive applications require edge deployment. The 75% forecast from IDC reflects not a rejection of cloud, but a recognition that different AI tasks need different compute environments — and the companies that match workloads to infrastructure correctly will have a meaningful cost and performance advantage over those that do not.