Artificial Intelligence is leading data center providers to offer an increasingly diverse portfolio of digital infrastructure facilities catering to the varying needs of different clients. This “different strokes for different folks” approach has led to the emergence of blended infrastructure. In this piece, we examine the myriad facets of this exciting trend.
What is “blended infrastructure”?
Blended infrastructure refers to a bouquet of different types of data centers that are designed, developed and deployed based on demand and user requirements.

“By blended infrastructure, we mean an environment where private infrastructure, colocation, public cloud, edge deployments and increasingly AI-specific platforms work together as one architecture,” explains Simon Lockington, Senior Director of Solutions Engineering, APAC, Equinix. “This extends beyond technology preference to a new business reality where enterprises need to balance performance, control, resilience and regulatory requirements across markets that are developing at different speeds.”
Lockington has observed a marked uptake in blended infrastructure adoption as it transitions into becoming what he feels is the practical operating model for many organizations across Asia-Pacific. He explains that most enterprises cannot replace their existing IT estate overnight. “Instead, modernization is done in stages where core systems, sensitive data and predictable workloads may remain in private environments, while other operations are conducted on cloud platforms to tap into its agility and access to digital services,” he says. “AI adds another layer, since training, inference, data pipelines and enterprise applications each have different requirements. The result places greater emphasis on a more intentional placement strategy based on where data is created, where it needs to be processed and where users and partners sit. This is particularly relevant in Asia-Pacific, where enterprises often operate across multiple jurisdictions and must accommodate different data residency, security and connectivity requirements.”
He says that a blended approach allows them to retain local control where necessary while still accessing global clouds, network providers and digital ecosystems. The added advantage is the flexibility to adapt as AI workloads mature and demand patterns change.

Ben Tulloch, Executive Managing Director, Advisory Services Asia-Pacific, NTT Data has also observed a similar trend. “Enterprises are taking a more deliberate approach to blended infrastructure as they modernize for AI across the region. The key shift observed is away from treating cloud, on-premises, and edge as separate destinations, and towards placing each workload where it fits best,” he says. “Organizations understand the need for greater control, but many are still working out how to adapt to their existing environments. In practice, the right approach is to match each workload to a mix of cloud, on-premises, and edge that makes the most sense.”
Out with the old?
Contrary to popular belief, traditional digital infrastructure is still valuable, and continues to serve a wide variety of clients spread across industries like banking, healthcare, retail, hospitality, and many more.

“We do not see blended infrastructure as a transitional phase,” says Sridhar Pinnapureddy, Founder & CEO, CtrlS Datacenters. “AI factories will become an important part of the digital infrastructure landscape, but they will complement, not replace, traditional enterprise IT, cloud and colocation infrastructure.”
He explains that different workloads have fundamentally different compute, power and latency requirements. “AI model training requires massive GPU clusters and extremely high-density infrastructure, while AI inference, enterprise applications, cloud services and business-critical systems often require very different operating environments,” says Pinnapureddy. “The future is therefore not about one type of datacenter replacing another; it is about building infrastructure optimized for specific workloads. We expect traditional datacenters, hyperscale cloud environments, AI factories and edge infrastructure to coexist and work together through seamless interconnection.”
Ankit Saraiya, Director & CEO, Techno Digital agrees. “Not every workload needs a GPU, just as not every rack needs 100 kW-plus of power. An enterprise application operating at conventional rack densities and a large AI training cluster are fundamentally different infrastructure problems. Designing both as if they were the same would add cost and complexity without necessarily adding value,” he says.
NTT Data’s Ben Tulloch concurs. “We expect blended infrastructure to remain the long-term operating model as enterprises become more AI-native across use cases,” he says, elaborating, “For example, public cloud remains at the forefront for frontier model training, rapid entry into broad geographic markets, and direct access to frontier AI APIs. Edge provides sub-10ms latency operations to eliminate bandwidth costs and function offline which is essential for scaled manufacturing. On-premises and sovereign environments provide low-latency integration with core legacy transaction systems, proprietary trade secret retention, and strict compliance with local data residency laws.”
Meanwhile according to Lockington, AI itself reinforces the need for a blended model. “Large-scale model training may happen in highly specialized, centralized environments with concentrated accelerated compute. But once AI moves into production, inference becomes much more distributed. It needs to sit closer to users, data sources, devices and business processes. That could mean a financial institution running fraud detection close to transactional systems, a manufacturer processing data from industrial sites, or a retailer using real-time intelligence to improve customer experiences. These use cases cannot always rely on a distant centralized environment without introducing latency, data movement costs or governance challenges,” he says.
Managing complexity
Planning and operating a blended infrastructure facility comes with a multitude of challenges, stemming from layers upon layers of interconnected complexities.

“What AI is changing is the range of workloads a modern data center may need to accommodate. That creates a fundamental engineering challenge: compute evolves much faster than physical infrastructure,” says Saraiya. “A data center is expected to operate for decades, while chip architectures, rack densities and thermal requirements can change within a few years. Moving from a facility designed around 8-10 kW per rack to supporting high-density AI deployments at 100 kW and beyond is not simply a server upgrade. It can affect the power train, electrical distribution, cooling architecture, floor loading and ultimately the economics of the facility.”
Tulloch adds, “Just as AI has transformed the software development lifecycle, it is starting to transform data and data pipeline management, thus unlocking data faster in legacy systems and enabling quicker use in AI-native applications. Snowflake, Google Cloud, Azure and Databricks’ agentic data platforms are leading the way here.” He says, “Shifting mindsets and cultivating skills to use these new capabilities, that rely less on traditional data engineering and increasingly on mastery of agentic tools, may emerge as internal hurdle. But, in the long run, it will be important for organizations to clean up legacy applications and infrastructure as these continue to be a drag on costs and agility.”
Meanwhile, Pinnapureddy points out that the challenge is not simply upgrading existing infrastructure; it is determining where retrofitting is practical and where purpose-built AI campuses are the better long-term solution. “Existing datacenters can be enhanced with high-density zones and upgraded cooling capabilities where feasible, while dedicated AI-ready campuses can be designed from the ground up to support next-generation GPU clusters. The objective is to deliver infrastructure that combines scalability, efficiency and reliability while remaining adaptable to rapidly evolving AI technologies,” he says.
Powering blended infrastructure
In a world ravaged by climate change, the need to find sustainable, energy efficient renewable power for data centers cannot be overstated. These days, many data center operators are increasingly investing in developing captive renewable power generation units such as solar and wind power farms. While some such plants directly supply power to the data centers, others engage in grid offsets.
But it is also important to make intelligent power storage choices.

“AI workloads create large and rapid swings in electricity demand. Storage placed close to the load can act as a shock absorber, charging when demand falls and discharging when it rises,” says Bill Mitchell, Senior Business Unit Director of Renewables & Energy Infrastructure, Jabil. “Depending on the load profile and system design, this could reduce peak incoming power requirements by 20 to 40 percent, helping operators use existing infrastructure more effectively and address power availability constraints.”
So, the bigger question is: how can operators manage peak demand, reduce grid constraints, and support the coexistence of AI and traditional data center workloads, all the while striving for sustainability?
Mitchell points out that reliability depends on storage, renewables, inverters or solid-state transformers, and controls operating as one coordinated system. “The energy management system senses changes in demand and directs the battery when to charge or discharge, while the inverter or solid-state transformer delivers usable power, and the battery management system protects the cells. Together, these components can respond within milliseconds, stabilize power, and support critical loads during grid disruptions,” he says.
Staying connected
Connectivity is an integral part of digital infrastructure, and it too must evolve to meet the growing and diverse needs of blended infrastructure.
“As enterprise workloads, cloud platforms and AI environments become increasingly distributed, performance depends not only on compute capacity but also on how efficiently data moves between these environments,” says Pinnapureddy. “This is particularly important for AI, where training and inference workloads rely on moving massive volumes of data with extremely low latency. High-performance interconnection enables seamless communication between enterprise systems, cloud platforms, AI clusters, networks and edge locations, ensuring that applications perform efficiently regardless of where the workloads reside.
Meanwhile, Lockington warns, “Hybrid AI environments depend on data, models, applications and users that are spread across different locations. A model may run in one environment, draw data from another, and feed results into an enterprise application or partner network. If those connections are slow, unpredictable, or difficult to manage, the performance of the overall system suffers regardless of how powerful the underlying compute may be.”
Conclusion
Therefore, traditional and AI data centers will continue to coexist. The challenge lies in managing their diverse operational needs and complexities.