Most organizations are still in AI's early stages, and the gap between adoption and impact is widening

Two major research bodies published findings this summer that, read together, define the actual scale of the AI challenge facing enterprise operators: the physical infrastructure required to run AI is outpacing grid capacity, and the organizational capability to extract value from it is lagging just as badly.

BloombergNEF, in analysis published July 29, 2026, projects that global data center electricity demand will reach 194 gigawatts by 2035. That figure is not a warning about a distant future. It is already reshaping procurement decisions today. According to BloombergNEF, grid connection wait times have grown so long that many data center operators are bypassing traditional interconnection queues entirely and building on-site natural gas generation to power facilities independent of the grid.

Meanwhile, McKinsey's People and Organizational Performance practice published a global survey in July 2026 finding that most organizations are still early in their AI journeys, with adoption concentrated at the level of individual tools rather than scaled into enterprise-wide performance gains. The dual picture is clarifying: enterprises are racing to deploy AI while simultaneously under-investing in the two things that make it work, power and organizational readiness.

The infrastructure ceiling is real and arriving fast

The 194-gigawatt demand projection from BloombergNEF represents the combined draw of AI compute, storage, and cooling at scale across the United States and globally. The firm's US Data Center Capacity Outlook, released July 21, 2026, adds regional texture: demand is not evenly distributed, and certain markets face acute constraints in both grid capacity and developable land. According to BloombergNEF, the biggest challenge facing data center growth right now is not computing power, it is electricity.

The practical consequence for operators is that natural gas has become the near-term bridge fuel of choice. BloombergNEF's July 2026 podcast analysis documents a pattern where data center developers, unable to secure utility-scale grid connections on a viable timeline, are turning to on-site gas turbines and generators to bring capacity online faster. This is a procurement and site-selection decision with multi-year operational implications, not a temporary workaround.

For enterprise CIOs and real estate or facilities teams evaluating new data center leases or co-location agreements, this shifts the evaluation criteria. Power availability, on-site generation capacity, and grid interconnection status now belong in the vendor due-diligence checklist alongside latency, compliance certifications, and SLA terms. Bank of America, in a Q&A published by BloombergNEF on August 6, 2026, described its AI infrastructure financing pipeline as robust, signaling that capital continues to flow into the sector despite the power constraints.

Most enterprises haven't crossed from adoption to impact

On the organizational side, McKinsey's global survey data published in the McKinsey Quarterly in July 2026 frames the problem with a three-horizon model. The first horizon is individual adoption, employees using AI tools to increase their own productivity. The second is workflow integration, where AI is embedded into team processes. The third is enterprise transformation, where AI changes the fundamental structure of how value is created. According to McKinsey, most organizations have made progress on horizon one but remain stuck before horizon two.

The diagnosis matters because the investment patterns don't match the ambition. Companies are licensing AI platforms, running pilots, and measuring adoption rates. But according to McKinsey's August 2026 research on closing the agentic adoption gap, AI transformations at their core require a reinvention of how work gets done, and that demands change leadership, not just change management. The distinction is meaningful: change management deploys a tool; change leadership redesigns the organization around it.

A separate McKinsey Quarterly article from July 14, 2026, surfaces a downstream risk that operations and HR leaders should take seriously. As AI absorbs entry-level tasks, the traditional pipeline for building organizational expertise narrows. Junior roles that once trained the next generation of analysts, engineers, and operators are being automated before organizations have built replacement development pathways. McKinsey argues that knowledge management, mentorship structures, and deliberate expertise development must be built into AI transformation plans from the start, not retrofitted later.

Agentic AI is raising the stakes on team structure

McKinsey's 2026 research on agentic AI, systems that can plan and execute multi-step tasks without continuous human direction, identifies a new management role emerging inside enterprises: the agent manager. According to McKinsey's July 27, 2026 analysis, companies are reassessing team structures and what it means to be a manager in an environment where AI agents handle a growing share of execution work. This is not a future scenario. It is happening now inside companies that have moved past horizon one.

For VP-level operators, the practical implication is that org design, workforce planning, and AI infrastructure investment are no longer separate workstreams. A company that secures power capacity for its AI compute but fails to redesign the workflows and management layers around it will not realize the returns it is financing. Conversely, a company that trains employees on AI tools without securing the compute and power infrastructure to run production-grade agentic systems at scale will hit a ceiling quickly.

What this means for your team

• Audit your data center and co-location contracts for on-site power generation capacity and grid interconnection status, BloombergNEF's 194 GW demand projection signals that power availability is a first-order selection criterion, not a secondary spec.
• Evaluate where your organization sits across McKinsey's three horizons: individual adoption, workflow integration, or enterprise transformation, and identify the specific bottlenecks blocking the move from horizon one to horizon two.
• Build explicit expertise development pathways into your AI transformation roadmap before entry-level task automation removes the traditional on-ramps for growing the next generation of domain experts.
• Assign clear accountability for agent manager roles and agentic workflow oversight now, before agentic systems are deployed at scale, so governance structures exist before the need is acute.