Cloudy with a chance of repatriation: enterprises flee public cloud for sovereign AI control

By early 2026, NTT DATA's research highlights a striking paradox in enterprise AI: while over 95% of organizations acknowledge the critical importance of private and sovereign AI, a mere 29% are actively prioritizing sovereign AI in their near-term strategies. This gap underscores a broader challenge where awareness does not translate into decisive action, yet those enterprises that do proactively redesign their AI infrastructure—focusing on control, data locality, and security—are carving out a competitive edge by aligning infrastructure, governance, and operating models early.

Data sovereignty and privacy regulations have evolved into formidable architectural constraints that disrupt traditional cloud models reliant on seamless cross-border data flows. According to NTT DATA, while data can still move, it no longer does so in ways AI systems require, forcing enterprises to rethink where data resides, how models operate, and the governance frameworks that oversee them. This regulatory landscape compels a fundamental redesign of AI systems, shifting from fluid, global data access to localized, jurisdiction-compliant architectures.

The complexity of integrating private and sovereign AI solutions emerges as a significant barrier, with over half of surveyed organizations citing it as their top challenge. This complexity stems from the need to orchestrate tightly coupled ecosystems involving multiple partners and diverse technology stacks. Compounding this challenge is the pervasive lack of confidence in cloud security—only 38% of enterprises express strong trust in their cloud security posture—undermining the foundational trust necessary for robust private and sovereign AI deployments.

By early 2026, AMD and Dell have emerged as pivotal players driving hybrid cloud solutions that empower enterprises to dynamically allocate AI workloads between on-premises infrastructure and multiple public clouds. This approach not only addresses the critical demands for scale, speed, and control but also lowers the barrier to entry for customers previously unable to invest in large-scale GPU configurations, thereby broadening access to on-prem AI inference capabilities. As AMD highlights, customers who couldn’t previously consume eight-way GPU setups now have expanded options, illustrating how hybrid architectures blend flexibility with inclusivity.

The rise of hybrid and sovereign edge computing models reflects a strategic enterprise response to the complex balancing act of cost, performance, compliance, and AI workload demands. Businesses increasingly migrate predictable, steady-state workloads from public clouds to sovereign private clouds or colocation facilities to reduce costs and maintain data residency, while leveraging edge computing—especially sovereign regional edge facilities—to run latency-sensitive AI inference workloads closer to end-users. This nuanced placement optimizes operational efficiency and compliance, as public hyperscale clouds remain indispensable for large-scale AI training, underscoring a sophisticated, multi-layered cloud strategy.

Re-platforming applications into containerized workloads running on regional sovereign edge facilities enables enterprises to maintain low latency and regulatory compliance while retaining the flexibility to burst to public clouds for massive scale when needed. Partnerships between AMD and Dell exemplify this trend, facilitating phased migrations that reduce operational risk and ensure local control. By anchoring AI inference workloads in sovereign edge environments, organizations can optimize performance and cost efficiency without sacrificing compliance, illustrating how hybrid and sovereign edge computing have become strategic pillars in modern enterprise AI infrastructure.

By mid-2026, CIOs across Asia are increasingly adopting capability-based budgeting to navigate the twin pressures of AI cost inflation and geopolitical shocks, shifting focus from traditional line-item allocations to investments in a targeted set of high-value business capabilities. This strategic pivot, championed by experts like Fred Giron, emphasizes ruthless prioritization—cutting application sprawl and technical debt—to free resources for AI initiatives that directly enhance customer or employee outcomes, rather than chasing experimental projects common in more mature markets. Such an approach reflects a broader regional trend where foundational investments in data infrastructure and platform modernization take precedence over speculative AI ventures.

Despite projected APAC tech spending growth of 9.3% in 2026, much of this increase masks diminished purchasing power as AI-embedded software renewals surge nearly five times faster than the consumer price index in Australia and hardware costs spike by 10–20%. This inflationary squeeze compels CIOs to mature their FinOps practices by embedding cost signals into engineering workflows, optimizing AI workloads through right-sizing and pipeline reuse, and strategically selecting models to balance cost with performance. Concurrently, volatile supply chains and geopolitical tensions have shifted planning from price sensitivity to capacity assurance, with CIOs locking in critical infrastructure through longer-term agreements and deferring non-essential refreshes to manage escalating costs.

Amid mounting economic pressures and a volatile procurement landscape, organizations are urged to anchor AI investments in measurable business outcomes and agile procurement processes to avoid costly missteps. Gartner’s 2024 survey revealing that 79% of technology buyers regretted their purchases underscores the risks of overinvesting in AI infrastructure without clear ROI. To build resilient AI capabilities, CIOs are advised to establish platform-centric models, form AI centers of excellence, and deploy fully automated delivery pipelines with stringent cost controls—practices achievable within 90 days using predominantly open source tools. This disciplined approach is critical given Gartner’s prediction that by 2027, half of AI projects designed for service desks will be abandoned due to unforeseen costs and risks.

Layered atop these economic challenges are evolving AI regulations around transparency, ethics, and data protection, which compel CIOs to embed responsible AI governance from the outset. This regulatory complexity adds another dimension to budget resilience, as organizations must mitigate privacy risks while scaling AI competitively. Integrating governance early not only safeguards compliance but also strengthens stakeholder confidence, enabling CIOs to sustain AI initiatives amid cost inflation and geopolitical uncertainty.

By mid-2026, a clear majority of enterprises—56% globally—have shifted or plan to shift their AI inference workloads from public to private clouds, driven by a blend of cost, security, and sovereignty concerns. Broadcom’s report highlights a notable 15-point drop in public cloud usage for AI inference, from 56% to 41%, underscoring a decisive market reset as organizations seek greater control and predictability in their AI infrastructure.

Cost inefficiencies in public cloud environments have emerged as a primary catalyst for this repatriation trend, with 31% of IT leaders citing cost as their top concern and a staggering 97% acknowledging wasted public cloud spend—over half estimating waste exceeds 25% of their budgets. Paul Turner of Broadcom encapsulates this dynamic, noting AI acts as a ‘cost multiplier’ that inflates infrastructure expenses, compelling enterprises to pursue more cost-effective private cloud solutions optimized through virtualization and GPU utilization.

Security, data sovereignty, and operational control remain critical forces steering AI workloads back on-premise, especially outside the U.S. where regulatory compliance is stringent. With 54% of IT leaders emphasizing data residency and sovereignty, and 37% prioritizing data protection, enterprises are demanding private cloud architectures that not only safeguard the data plane but also the control plane—enabling offline operation and tighter governance, as Chris Wolf articulates. This is particularly pronounced in the Asia Pacific and Japan region, where 82% have considered repatriation and 54% have already moved workloads to private clouds.

The evolving AI lifecycle is reshaping cloud strategies: while public clouds remain favored for training and experimentation, day-to-day AI inference is increasingly anchored in private clouds to address token costs, data gravity, and compliance concerns that have ascended to boardroom priorities. Prashanth Shenoy of Broadcom observes that enterprises now prefer keeping AI models closer to data sources within on-premise private clouds, signaling a break from the long-held assumption that all workloads would eventually migrate to public clouds.

By mid-2026, enterprises operationalizing sovereign AI are prioritizing governance models that secure both the data plane and control plane, enabling disconnected operation and uninterrupted control over AI workloads within private cloud environments. As Chris Wolf from Broadcom explains, this dual sovereignty ensures organizations can 'disconnect from the internet and continue to run,' a critical evolution driven by the need for resilience and autonomy in AI deployments. Complementing this, Craig McLellan of ThinkOn highlights that AI sovereignty extends beyond connectivity to encompass stringent data classification, access controls, and model governance, all within trusted private clouds to effectively manage data gravity and compliance risks.

The rising tide of AI-driven cyberattacks—reported by Paul Turner as impacting 73% of enterprises across industries—has accelerated the shift toward private cloud infrastructures fortified with enhanced governance and operational controls. This security imperative compels organizations to implement microsegmentation and strict access controls as foundational defenses, particularly in regulated sectors where AI sovereignty is non-negotiable. These measures not only mitigate compliance and risk challenges but also underpin the broader strategic move from AI pilots to scaled production, as noted by Prashanth Shenoy, who observes that robust governance frameworks are now essential to balance cost, security, and compliance in hybrid and private cloud deployments.

By mid-2026, enterprise AI infrastructure strategies are decisively moving away from a cloud-first default toward a compositional approach that prioritizes workload-specific, cost-aware, and governed placements. This shift acknowledges that AI workloads—characterized by sustained high-intensity compute demands, massive data volumes, and latency sensitivity—disrupt the traditional economic and architectural assumptions underpinning public cloud attractiveness. As a result, organizations must cultivate strong operational discipline, ensuring clear ownership, cost transparency, and alignment with business goals to prevent fragmented architectures and spiraling expenses, effectively treating cloud as a governed business capability rather than a mere utility.

The emerging cloud paradigm emphasizes intentional workload placement where efficiency and compliance are maximized, moving beyond traditional multi-cloud redundancy models. Public clouds remain optimal for elastic, globally distributed workloads, but private, on-premises, and sovereign or regional environments are gaining prominence for sustained, high-intensity AI compute needs. This nuanced landscape fosters the rise of specialized providers and regional clouds that address regulatory and latency constraints, signaling a future where compositional cloud strategies are tailored to the unique demands of AI workloads and sovereignty requirements.

To navigate the complexity and escalating costs of AI integration, enterprises must transition to platform-centric operating models and establish dedicated AI centers of excellence. Gartner’s guidance highlights that within 90 days, organizations can leverage mostly free and open-source tools to build fully automated delivery pipelines with strict cost controls. Complementing this, experts like Autumn Stanish advocate for creating platform teams led by product owners who shift focus from traditional uptime metrics to business outcomes such as revenue growth and customer satisfaction, effectively absorbing legacy server and storage roles into a more agile, outcome-driven structure.

Economic discipline in technology procurement emerges as a critical success factor amid rising cloud costs and AI complexity. Luke Ellery references a 2024 survey revealing that 79% of technology buyers regretted their purchases, underscoring the need for senior-level engagement, agile procurement processes, and risk-informed decision-making. This approach ensures cloud and AI investments are tightly coupled to measurable business value, preventing costly misalignments and enabling enterprises to modernize infrastructure with both financial prudence and strategic clarity.