Brian Gruttadauria, CTO Hybrid Cloud, HPE.
getty
For years, the dominant enterprise architecture model was relatively straightforward: Move data to a central location, run applications there and scale infrastructure as demand grows. AI is making that model increasingly difficult to sustain.
AI, real-time analytics and data-intensive applications are changing both where data is created and where it can be processed efficiently. Data now accumulates across data centers, public clouds, factories, hospitals, retailers and other edge environments. At the same time, access to GPUs, power, cooling and specialized infrastructure is increasingly constraining where demanding workloads can run.
These factors create a new architectural tension. Data gravity pulls applications and services toward the location of the data. Compute gravity pulls workloads toward environments with the capacity to process them. Increasingly, those locations are not the same, leaving enterprises to operate between the two.
The result is a shift from static hybrid cloud environments to dynamic, distributed architectures capable of continuously determining where data should remain, where workloads should run and how the two should connect.
The enterprise is distributed by design.
Distributed IT is not new. What has changed is the speed, scale and performance of the workloads operating across it.
Traditional applications could often tolerate manual provisioning, fixed network capacity and operational handoffs between infrastructure teams. AI pipelines and real-time analytics expose the limitations of that approach. Latency, bandwidth constraints, duplicated data and fragmented visibility can quickly become barriers to performance and cost control.
This means a hybrid cloud approach can no longer be treated as a collection of separate environments. It must operate as a coordinated system spanning compute, storage, networking, data and applications.
That coordination matters because there is rarely one correct location for every workload, and that location may change over time. Some data needs to remain on premises because of governance, sovereignty, security, latency or cost. Other workloads may need access to specialized cloud services or accelerated compute. Still others must run near the edge to support real-time decisions.
The challenge is no longer simply deciding where a workload belongs. It’s building an architecture capable of continually reassessing that decision as organizations’ needs change.
The network is becoming part of the control plane.
Many organizations still manage the network mainly as a way to move data from one place to another. Capacity is provisioned in advance, changes require manual processes and connectivity is managed separately from applications and infrastructure.
That model does not fit a world in which workload requirements can change by the hour.
As enterprises become more distributed, the network must become programmable, observable and responsive to application needs. Bandwidth, latency, routing and connectivity policies should be managed as part of the workload lifecycle rather than through a separate operational process.
This is an important industry shift. The network is evolving from a static path into an active part of the hybrid cloud control plane. Telemetry can help teams understand how applications are behaving across environments, while automation can adjust connectivity and capacity as requirements change.
This does more than improve data transfer; it can enable new operating models. When a remote operation can transmit large datasets immediately rather than storing them locally or moving them through physical handoffs, analysis can begin earlier, duplication can decline and operational decisions can happen faster.
AI operations must work across infrastructure silos.
The next step is coordinated intelligence across the full technology stack.
Most enterprises already have automation within individual domains. Compute platforms can optimize resources. Storage systems can manage capacity. Network tools can identify performance issues. The problem is that these systems often operate independently.
AI-native operations require a broader view. An application slowdown may originate in compute, storage, connectivity, data placement or a combination of factors. Optimizing one layer without understanding the others can simply move the bottleneck.
The industry is moving toward operational models that can correlate telemetry across domains, interpret changing conditions and recommend or take action based on business intent and policy. The goal is not autonomy for its own sake. It is faster, more consistent decision-making across environments that are too complex for teams to manage one component at a time.
Governance becomes more important as infrastructure becomes more autonomous.
Greater autonomy also creates greater responsibility.
When systems can dynamically move workloads, adjust capacity, modify connectivity or remediate problems, organizations need clear boundaries around what those systems are allowed to do. Policies for security, compliance, data residency, cost and operational risk must be embedded into the architecture.
Actions should also remain observable, explainable and auditable. Technology leaders should be able to understand what changed, why it changed, what information informed the decision and which policy authorized the decision.
This is particularly important in regulated industries, but it is quickly becoming a general requirement. Ultimately, trust, not technical capability alone, will determine how much operational autonomy enterprises are willing to adopt.
Here are four priorities for technology leaders.
As this shift accelerates, leaders should focus on four areas.
First, map where critical data is created, stored, processed and consumed. Data strategy must account for gravity, not only governance.
Second, develop workload placement policies that consider latency, compute availability, security, sovereignty, resilience and cost together.
Third, treat connectivity as a programmable resource rather than a fixed constraint. Network capacity and policy should be able to adapt alongside applications and infrastructure.
Finally, build observability and governance across domains before introducing greater autonomy. AI cannot coordinate what the enterprise cannot see, and it should not act beyond clearly defined trust boundaries.
The next generation of hybrid cloud will not be defined by how many environments an organization uses. It will be defined by how effectively those environments function as one system.
Enterprises that can continuously align data, compute and connectivity will be better positioned to scale AI, respond to changing business needs and turn distributed infrastructure from a source of complexity into a strategic advantage.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
