Inference has emerged as the leading force in enterprise AI, bringing with it a significant challenge: the transport of data to compute environments. Each inference call necessitates the movement of sensitive enterprise information from its native systems to external platforms that prioritize GPU throughput over data governance. This process introduces friction that escalates with scale, leading to increased costs, heightened security risks, and a proliferation of data copies that can become misaligned with operational realities.
What enterprises truly desire is to maintain the integrity of their data and intellectual property within their databases, avoiding the creation of multiple copies and the complexities that arise from managing inconsistencies.
According to research involving over 2,050 senior executives from major global enterprises, 95% of organizations plan to develop their own AI and data platforms within the next 780 working days. However, only 13% have successfully achieved this goal. Those that have succeeded are realizing nearly five times the return on investment compared to their counterparts still grappling with AI operationalization.
The distinction between leaders and followers lies not in the quality of their models, but in their infrastructure strategy.
Leading organizations have embraced a sovereign-by-design approach, with more than 75% operating across multiple clouds and on-premises environments rather than depending on a single hyperscale provider. They are tailoring their AI frameworks to align with their unique business, regulatory, and operational needs, rather than conforming to the architectures dictated by cloud vendors.
The shift from training to inference
Training represents a singular event, while inference constitutes an ongoing business process. A model may undergo training once, yet it can be invoked millions of times daily. Every instance of fraud detection, insurance claim assessment, customer service interaction, medical recommendation, sanctions check, or predictive maintenance relies on inference executed against live operational data.
“What separates the leaders from the followers is not model quality. It is infrastructure strategy.”
This fundamental distinction reshapes the requirements for enterprise infrastructure.
While training workloads emphasize compute density and GPU availability, inference workloads prioritize latency, governance, reliability, and cost management. These workloads must function where business data resides and where compliance requirements can be enforced.
In heavily regulated sectors such as financial services, healthcare, telecommunications, energy, and the public sector, inference cannot simply be executed in the region with the lowest compute costs. Data sovereignty requirements, audit obligations, and security mandates often dictate the specific locations where workloads must operate.
Consequently, the challenge extends beyond AI itself. Organizations require an operational model that integrates compute, data, and governance without sacrificing flexibility.
Why neoclouds are emerging as a critical layer to cross the chasm to production
In this context, neoclouds have gained prominence.
Unlike traditional hyperscalers, neoclouds are specifically designed for AI infrastructure. Their emphasis is not on delivering a multitude of generic cloud services but on optimizing GPU access, enhancing AI performance, and offering flexible consumption models.
For many enterprises, neoclouds present a compelling solution to the increasing demand for specialized AI compute. They provide access to the latest accelerator technologies while allowing organizations to scale their workloads without the complexities often associated with large cloud environments.
“The future of AI architecture therefore depends on bringing models closer to data rather than moving data closer to models.”
However, neoclouds address only one aspect of the enterprise AI equation.
AI does not generate value in isolation; models require context. They need access to customer records, transaction histories, operational workflows, policy documents, supply chain information, and enterprise knowledge. Transferring these assets into separate AI environments leads to duplication, latency, and governance challenges.
Thus, the future of AI architecture hinges on the principle of bringing models closer to data rather than relocating data closer to models.
Why Postgres has become the enterprise AI foundation
As organizations seek a unified platform that accommodates both operational and AI workloads, Postgres has emerged as a natural foundation.
Postgres already serves as the operational backbone for many critical applications worldwide. It combines transactional reliability, extensibility, and scalability with the openness that enterprises increasingly demand. Over 70% of AI-related application development is occurring on Postgres.
What makes Postgres particularly relevant in the AI landscape is its capacity to function beyond a mere database. It can act as a governed memory layer for AI systems, integrating operational data, application context, permissions, observability, and retrieval capabilities into a cohesive architecture.
This integration significantly reduces complexity.
Instead of managing separate infrastructures for transactional systems, vector stores, AI memory layers, and governance frameworks, organizations can consolidate around a trusted operational platform that already supports their mission-critical workloads.
For CIOs aiming to balance innovation with control, this architectural simplification presents a considerable strategic advantage.
Why sovereignty matters more than ever
Sovereignty has emerged as a defining theme in enterprise technology.
For financial institutions, sovereignty entails maintaining control over financial data and regulatory obligations. For healthcare organizations, it means safeguarding patient information while fostering innovation. For governments, it involves ensuring that national and citizen data remains under appropriate jurisdictional control.
The rise of AI has intensified these concerns.
Organizations increasingly require assurance that models, data, policies, and operational controls can remain within designated environments while still leveraging the advancements of AI technology.
This necessity is driving the demand for sovereign AI architectures capable of functioning across clouds, private infrastructures, and on-premises environments.
The challenge lies in achieving consistency across these environments without introducing operational complexity.
EDB Postgres AI: connecting sovereign data and sovereign AI
EDB Postgres AI addresses this challenge by integrating operational Postgres, AI capabilities, and hybrid infrastructure management into a unified platform.
Rather than compelling enterprises to choose between innovation and control, EDB Postgres AI empowers organizations to deploy AI where their data already resides. With capabilities spanning operational databases, analytics, agentic AI workloads, and hybrid management, organizations can establish a consistent operating model across sovereign environments.
This approach is especially pertinent for regulated industries, where transferring sensitive information to external AI services may raise compliance, security, or governance concerns.
By facilitating inference close to operational data, organizations can minimize data movement, enhance performance, and fortify their compliance posture. Simultaneously, they retain the flexibility necessary to leverage emerging AI technologies and modern infrastructure models.
“By enabling inference close to operational data, organizations reduce data movement, improve performance, and strengthen their compliance posture.”
The outcome is a platform that aligns with the realities of enterprise AI rather than the assumptions of consumer AI.
“The reality is that the new AI at scale world needs a new infrastructure. That isn’t just the compute; it’s the governance, heuristic data access, and level of observational and orchestration control that are absolute, governed, agile, and work for humans and agents.” — Nancy Hensley, CPO, EDB
The new enterprise AI stack
The evolving enterprise AI architecture is increasingly constructed around complementary rather than competing technologies.
Together, these layers form an architecture capable of supporting the entire AI lifecycle—from experimentation and model training through production inference and continuous optimization.
The CIO imperative
Organizations that are extracting the most value from AI are no longer focused solely on how to train better models. They are concentrating on how to operationalize AI across the enterprise while maintaining control over costs, governance, and risk.
Their strategies are becoming increasingly uniform.
They are adopting multi-cloud and hybrid approaches instead of relying on a single cloud provider. They prioritize sovereign architectures over centralized data movement. They build on open operational foundations rather than succumbing to proprietary lock-in. Most critically, they recognize that AI success hinges on bringing intelligence to data, not the other way around.
Neoclouds supply the compute layer essential for modern AI, while Postgres offers the operational foundation necessary for trusted enterprise systems. EDB Postgres AI bridges these realms through a sovereign architecture tailored to the realities of regulated industries.
As AI transitions from a phase of experimentation to one of operational necessity, the enterprises that thrive will be those that can ensure inference is secure, governed, low-latency, and economically sustainable at scale.
In the forthcoming era of enterprise AI, the most significant business value will not stem from model selection or raw GPU access. It will arise from an infrastructure strategy centered around data—keeping intelligence close to where data resides, governed, trusted, and ready to act.
