The CEO’s AI Dashboard, Part 4: Infrastructure Readiness

Welcome to a new edition of The Board: Distillation Aftershots ^(*).

This online version of ouot newsletter, written by Constellation's Chief Distiller and Board Advisor Esteban Kolsky, shares curious and interesting insights and data points distilled from enterprise technology to identify what’s notable. To receive the newsletter with updates to The Board issues, please click here.

In this issue, I want to move from data readiness to the infrastructure that will determine whether AI can operate at enterprise scale with enough speed, control, and resilience to matter. Infrastructure readiness is often treated as a technology upgrade. It is closer to a long-term operating choice. The enterprise has to leverage what it already built, support the business models it is trying to create, and give users enough flexibility to move faster without losing governance, security, or control.

First, my take.

Two misconceptions keep showing up in this discussion. The first is that the digital transformation work of the last decade is enough to carry the enterprise into the AI era. It helped and built a foundation, but it was not enough in the long run. The second is that AI readiness depends on perfectly clean data. Better data improves outcomes over time, but infrastructure readiness turns more on whether the right data can be accessed, governed, moved, and used in context at the right moment.

AI is changing what infrastructure has to do. The earlier generation of cloud and data modernization was built to stabilize applications, reliably store data, and distribute access. The next generation must support near-real-time execution, continuous inference, and highly personalized interactions across workflows. Deloitte’s 2026 infrastructure survey found that 97% of leaders are confident they can scale AI workloads in the next three years. It also shows that 64% have started limited or at-scale deployments of AI factories, 88% expect the same by 2028, and 72% expect AI at the edge to scale by then as well. The demand curve is moving fast.

The enterprise needs a mix of brains and brawn. Brains means CPUs, logic, orchestration, workflow control, and the systems that decide how work should move. Brawn means GPUs, dense compute, memory, networking, storage, and the capacity to process large volumes of data and inference requests at speed. Deloitte’s compute-strategy work says wafer costs are expected to rise by 20%, pushing both GPU and CPU prices higher, while recurring inference workloads are already forcing enterprises to rethink the economics of API-heavy architectures. The infrastructure bill is about running the business.

That is why infrastructure readiness should be focused less on raw hardware and more on plumbing. The issue is whether the enterprise can deliver the right content in context, with sufficient speed, control, and meaning. In an assessment of 216 cloud estates, Accenture found that 59% of core workloads remain on-premises or in aging environments, only 8% are dedicated to experimenting with advanced technologies, and over 60% of cloud strategies are still not aligned with long-term business goals. It also found that only 2% of companies have fully integrated data and AI for real-time insights.

The next business flow depends on the context layer: .memory, semantic structure, metadata, access rights, confidentiality, observability, and the ability to retrieve the right knowledge or data at the right time are becoming mission-critical. Enterprises will overpay for compute if they use more data and more processing to compensate for weak semantics, weak classification, or poor workflow design. They will also overbuild if they assume more storage or more model access can fix weak infrastructure logic. The better path is narrower and more disciplined: determine what data the chosen models and workflows actually need, secure it, classify it, and make it available with the right controls. Observability belongs in that same layer. Traditional infrastructure monitoring is not enough when the operating risk now includes compliance drift, inference cost, retrieval quality, latency variance, and model-driven failure paths.

All of this can be reduced to four tenets for infrastructure readiness: context, governance, latency, and economics. Context determines whether the right knowledge, data, and meaning are available when needed. Governance determines whether access, confidentiality, security, and control remain in place as the infrastructure scales. Latency determines whether the system can support near real-time execution, personalization, and distributed decision-making. Economics determines whether the enterprise can afford to operate the model, the workflow, and the surrounding infrastructure over time. If one of those four breaks, the infrastructure is not ready, no matter how modern it looks.

The cloud principles that got enterprises here still matter: use cloud architectures, retain central control, and mpower business operations. Those principles now need updating for an environment in which workload placement is driven simultaneously by latency, economics, regulation, risk, and confidentiality. Accenture makes the point clearly: cloud now spans public, private, hybrid, multi-cloud, sovereign cloud, and edge. IBM describes the answer as hybrid by design. That is closer to the right operating model than the usual cloud-versus-on-premises argument; the infrastructure has to work across both.

There is one more practical issue that deserves more attention: dependency chains. Infrastructure readiness now depends on external constraints the enterprise cannot control well, including power, chips, specialized networking, and capacity concentrated among a small number of providers. Those constraints are now part of strategy. If the enterprise sequences its investments poorly, buying compute before deciding which workflows, semantics, and control layers actually matter, it can lock itself into an expensive architecture before it knows what the business really needs.

Infrastructure readiness is a business initiative with heavy plumbing amidst a paradigm shift. AI is the first in a line of frontier technologies that will place new demands on the enterprise over the next two to three decades. The infrastructure built now will not only support AI. It will support whatever comes next. Smart CEOs will oversee this directly, use executives and board members with both technology and business judgment, and resist advisers who reduce the issue to cleaner data or another round of technology-first modernization.

Recommended CEO actions:

1. Make infrastructure readiness a business-led operating priority, with the CIO and CFO jointly accountable for timing, investment sequencing, and long-term architecture choices.
2. Require infrastructure decisions to be tested against four conditions: context, governance, latency, and economics. If a platform cannot satisfy those together, it will become a bottleneck in the future.
3. Invest in a private platform to address hybrid private-public cloud infrastructure that preserves central control over sensitive data, semantic assets, and workflow logic while still giving the business elastic access to compute and services.
4. Treat “clean data” and “technology-first modernization” as secondary screens. The primary screen is whether the infrastructure can support the business model the enterprise is trying to build for the next 20-30 years.

Here are some reading resources: