Digital Transformation • 1 day ago • Shruti Das
For much of the digital transformation era, cloud migration was treated as a relatively straightforward strategic direction. Enterprises moved applications out of their data centers, adopted SaaS platforms and shifted infrastructure toward public cloud providers in pursuit of greater flexibility, scalability and speed.
The underlying assumption was often simple: moving more technology into the cloud represented progress.
That assumption is becoming more complicated in the AI era. As enterprises move beyond experimentation and begin deploying AI across business processes, the question is no longer simply whether a workload belongs in the cloud. It is increasingly about which workload should run where, and why.
AI is forcing organizations to reconsider infrastructure decisions through a wider set of variables, including data location, latency, cost, security, sovereignty and the technical requirements of different workloads. The result is not necessarily a retreat from the cloud. Instead, digital transformation is moving away from a broad cloud-first philosophy toward a more selective workload-first approach.
Recent industry developments reinforce that shift. Gartner has argued that cloud architects need to rebalance hybrid workload placement policies for an AI-enabled future, while ISG has reported that enterprises are increasingly recasting hybrid cloud strategies around AI infrastructure requirements, sovereignty and governance. At the same time, surging demand for AI servers and networking is demonstrating that enterprise AI is creating significant infrastructure needs beyond traditional cloud software consumption.
The cloud is not becoming less important. In many ways, it is becoming more strategic. But AI is making enterprises much more deliberate about where their technology actually runs.
Cloud-first is giving way to workload-first
Cloud-first strategies emerged at a time when many enterprises were struggling with aging infrastructure and slow technology delivery. Public cloud platforms offered the ability to provision computing resources quickly, scale applications without buying hardware in advance and access new technology services without building everything internally.
For many workloads, that model remains highly effective. SaaS applications, customer-facing digital services, development environments and variable workloads can all benefit from the flexibility of public cloud infrastructure.
AI, however, does not behave like every other enterprise workload. Training, inference, retrieval and data processing can each have different requirements, while AI systems may depend heavily on proximity to large volumes of enterprise data. Moving information repeatedly between environments can introduce additional costs, latency and security considerations.
This is pushing infrastructure decisions toward a more granular model. Instead of beginning with the question, “Should we move this to the cloud?”, enterprises increasingly need to ask what the workload requires in terms of performance, data access, economics and governance.
That shift is reflected in Gartner’s 2026 guidance, which calls on cloud architects to rethink cloud strategies for an AI-enabled future and rebalance hybrid workload placement policies around business outcomes and sovereign constraints.
The destination matters less than the fit. A workload running in a public cloud is not automatically more modern than one running in a private environment, and an on-premises workload is not automatically a legacy system. The increasingly important question is whether the chosen infrastructure supports the workload effectively.
AI is bringing data location back into the conversation
One of the biggest reasons AI is changing infrastructure strategy is data.
Traditional cloud transformation often focused heavily on moving applications. AI introduces a stronger need to consider where the information required by those applications already exists. Enterprise data can be distributed across operational databases, private infrastructure, cloud platforms, SaaS applications and edge environments.
For an AI system, moving all of that information into a single location may not always be practical or desirable.
Some data may be too sensitive to move freely. Other information may be subject to sovereignty or regulatory requirements. Large datasets can also be expensive and time-consuming to transfer, while certain AI applications may require low-latency access to information or computing resources located close to where the data is generated.
This is reviving the importance of data gravity in enterprise infrastructure decisions. Rather than assuming that data should always move toward computing resources, organizations are increasingly considering situations where computing and AI capabilities should move closer to the data.
The same principle is becoming increasingly important for AI inference. Recent distributed AI infrastructure initiatives, including Equinix’s Inference Exchange, reflect the growing interest in bringing AI computation closer to enterprise data and applications rather than treating AI processing as something that must always happen in a centralized environment.
This does not mean every enterprise needs to build private AI infrastructure. It does mean that data location is once again becoming a strategic architecture decision.
The economics of AI are changing cloud decisions
Cost is another reason enterprises are reassessing infrastructure choices.
The cloud’s traditional economic advantage was closely tied to flexibility. Organizations could avoid large upfront investments and pay for computing resources as they used them. That model remains valuable, particularly when demand is unpredictable.
AI changes the equation because some workloads can become intensive, continuous and expensive. A large-scale AI application may require sustained access to GPUs, high-performance networking and specialized infrastructure. As usage grows, enterprises need to consider whether the economics of consuming those resources through external cloud services remain appropriate for every workload.
This does not create a universal argument for owning infrastructure. Buying and operating AI hardware brings its own costs, including capital expenditure, power, cooling, skills and capacity planning. The economics will vary significantly depending on workload utilization and scale.
What is changing is the nature of the conversation. Infrastructure decisions can no longer be based only on whether cloud or on-premises technology is cheaper in isolation. Organizations increasingly need to understand the total cost of supporting a workload over time.
HPE’s latest results provide a useful signal of how quickly this infrastructure demand is growing. The company raised its forecasts after reporting strong growth driven by demand for AI-related servers and networking equipment, with its executives highlighting continued enterprise adoption of AI as an important driver of future growth.
The broader implication for digital transformation leaders is that AI infrastructure is becoming a business planning issue, not simply a technical procurement decision.
Hybrid is becoming more deliberate
Hybrid cloud has been an enterprise reality for years, often because organizations had no practical alternative. Legacy applications remained on-premises while newer workloads moved to public clouds, creating a hybrid environment through necessity rather than deliberate architecture.
AI could change the meaning of hybrid.
The emerging model is less about accepting infrastructure fragmentation and more about intentionally placing workloads in different environments according to their requirements. A company may use public cloud infrastructure for highly elastic workloads, private infrastructure for sensitive data and AI applications, edge computing for low-latency services and SaaS platforms for standardized business functions.
That creates a more distributed technology environment, but not necessarily a more chaotic one. The objective is to make workload placement a strategic decision rather than the accidental result of years of technology purchases.
ISG’s recent research reflects this direction, reporting that enterprises are investing in hybrid operating models that prepare environments for AI while addressing requirements around observability, governance, FinOps and data sovereignty.
This represents an important evolution in digital transformation thinking. The future infrastructure strategy may not be about choosing between public and private environments. It may be about developing the ability to operate effectively across both.
Sovereignty is making architecture more strategic
AI is also making questions around data sovereignty and control more prominent.
Enterprises operating across multiple countries may need to consider where sensitive information is stored and processed. Governments and regulated industries may impose additional requirements around data handling, while organizations themselves may want greater control over critical systems and intellectual property.
These concerns existed before generative AI, but AI expands their importance because enterprise information is increasingly being used to train, ground or inform intelligent systems.
The question is therefore not simply where an organization’s data is stored. It can also involve where AI processing takes place, which infrastructure providers are involved and how information moves between systems.
Gartner’s 2026 cloud strategy guidance specifically identifies sovereign constraints as a growing factor in workload placement and calls for cloud strategies that balance AI ambitions with broader geopolitical and regulatory realities.
For digital transformation leaders, this means architecture decisions are increasingly connected to business risk and compliance strategy. Infrastructure can no longer be viewed purely as an IT efficiency issue.
The cloud operating model also has to change
The infrastructure question is not limited to where workloads run. Enterprises also need to reconsider how they manage increasingly distributed environments.
A traditional cloud operating model may have been designed primarily around human users and application teams provisioning resources. AI introduces new consumers of infrastructure, including automated systems and agents that may require access to data, computing resources and business applications.
This creates additional governance challenges.
Organizations will need stronger visibility into how infrastructure resources are being consumed, where AI workloads are running and how costs are accumulating. FinOps practices will become increasingly relevant as enterprises attempt to manage potentially expensive AI consumption across multiple environments.
At the same time, security and identity controls will need to operate consistently across distributed infrastructure. An enterprise cannot treat public cloud, private infrastructure and AI platforms as completely separate security domains if data and workflows increasingly move between them.
The hybrid environment therefore requires a more mature operating model. The goal is not simply to connect multiple infrastructure environments but to manage them as parts of a coordinated technology strategy.
AI could expose the weaknesses of earlier cloud transformation
Many enterprises have spent years modernizing applications without necessarily modernizing the architecture around them.
They may have moved workloads into public cloud environments but retained fragmented data, complex integrations and poorly understood dependencies. AI can make those weaknesses more visible because intelligent systems often need to work across the boundaries created by those disconnected environments.
An organization may have dozens of cloud applications, but an AI agent cannot easily create an end-to-end workflow if every system has different interfaces, permissions and data definitions. Similarly, a company may have successfully migrated infrastructure but still struggle to deploy AI if critical information remains inaccessible or poorly governed.
This is why the AI era may force enterprises to revisit earlier transformation decisions.
The question is not whether cloud migration was a mistake. In many cases, cloud platforms provided essential capabilities and created the digital foundations enterprises now depend on. The challenge is that the requirements have changed.
Digital transformation focused heavily on moving technology into more flexible environments. AI transformation requires organizations to think about how those environments work together.
The new cloud strategy is about placement, not destination
The most important shift may be conceptual.
For years, cloud transformation was often discussed as a journey toward a destination. The enterprise would move from traditional infrastructure to the cloud, gradually increasing the percentage of applications and workloads running there.
AI makes that model less useful.
The modern enterprise may have workloads permanently distributed across public clouds, private infrastructure, edge environments and specialized AI platforms. The objective is not necessarily to move everything to one place. It is to determine the best place for each workload.
This makes workload placement one of the more important strategic capabilities of the AI era.
CIO and infrastructure leaders will increasingly need frameworks for evaluating workloads based on factors such as data sensitivity, latency, utilization, regulatory requirements, cost and integration needs. Those decisions will need to evolve as AI applications move from experimentation into production and their infrastructure requirements become clearer.
The result is a more complex environment, but potentially a more mature one. Instead of applying a single infrastructure philosophy across the enterprise, organizations can build architecture around business and technical requirements.
Digital transformation is becoming architecture transformation
The cloud remains one of the most important technologies of the digital transformation era. AI is not replacing it.
What AI is doing is forcing enterprises to become more sophisticated about how they use it.
The next phase of transformation will not be defined by how much infrastructure an organization has moved into the cloud. It will increasingly be defined by how effectively the enterprise can place workloads across multiple environments while maintaining performance, security, governance and economic discipline.
That is a significant shift from the cloud-first thinking that dominated much of the last decade.
AI workloads are making infrastructure decisions more consequential because they sit at the intersection of data, computing, business processes and increasingly autonomous systems. As a result, the enterprise architecture of the future is likely to be more distributed, more specialized and more intentional.
The cloud was once presented as the destination of digital transformation. In the AI era, it is becoming one part of a much larger infrastructure strategy.
The future of digital transformation may not be cloud-first. It may be workload-first.
Key Takeaways
- Cloud-first is evolving into workload-first. AI is forcing enterprises to decide where workloads should run based on performance, data, cost, governance and business requirements.
- Data location is becoming strategically important again. AI workloads often depend on large volumes of sensitive enterprise information, making data gravity and proximity increasingly relevant.
- AI is changing infrastructure economics. Continuous, compute-intensive workloads require enterprises to reassess the long-term cost of different deployment models.
- Hybrid infrastructure is becoming more intentional. The goal is increasingly to place workloads deliberately across public cloud, private infrastructure and edge environments.
- Sovereignty and governance influence architecture. Where AI systems process enterprise information is becoming a strategic risk and compliance consideration.
- Earlier cloud transformation decisions may need to be revisited. AI is exposing fragmented data, disconnected applications and complex integrations that can limit intelligent automation.
- The next phase of digital transformation is increasingly about architecture. Success will depend less on choosing a single destination and more on making better workload placement decisions.