New Demand, Familiar Fundamentals
Since the first data center was built in the 1940s, the data market has changed significantly.
Across the globe, artificial intelligence (AI) has shifted to a daily operating necessity for many companies, and the data center infrastructure required to support that shift is expanding at an extraordinary speed. McKinsey estimates that global data center investment could reach $6.7 trillion by 2030, with roughly 70% of demand tied to AI demands.[1]
That scale has encouraged the market to describe the development of AI-era data centers as something fundamentally new: a type of project unlike the industrial or infrastructure projects that came before it. However, while the type of demand is new with unprecedented technical requirements, the projects themselves are governed by many of the same delivery fundamentals that determine success or failure. Project outcomes are still determined by funding availability, scope definition, schedule realism, procurement sequencing, labor availability, project governance, and risk allocation.
While AI‑era data centers are still capital projects at their core, these projects are differentiated by their distinct challenges particularly around power, scale, speed, integration, and uncertainty in a rapidly evolving digital market.
Familiar Fundamentals Still Apply
Core project delivery principles remain unchanged. Data centers are not a new type of project; they are complex capital projects marketed under a new narrative. The same forces that shape outcomes in other mission-critical programs across the power, advanced manufacturing, oil and gas, and petrochemical industries also shape outcomes in data center projects. Success or failure still depends on identifying risks early, clearly assigning responsibilities, and managing execution with discipline from planning and development through turnover.
While headlines may identify AI-era data center delivery risks as unique, in practice, many of the most consequential failures are familiar: unrealistic assumptions during development, incomplete planning and scheduling, resource constraints, poorly defined contracting structures, and operational readiness issues that are discovered too late. What has changed in AI-era data center projects is the speed at which those risks compound and the limited opportunity to recover once performance begins to slip.
More than ever, the design of data centers is influenced by the requirements of AI demands and workloads that are rapidly changing. In the current market, once a data center breaks ground, the window of relevance for its originally specified technology begins to shrink. The risk of obsolescence is more significant than traditional concerns like cost overruns or delayed interim milestones. Delay to a data center project schedule can mean that, by the time the facility is operational, the computing infrastructure may no longer match the needs of current AI workloads. Impacts to schedule not only threaten profitability, but they also risk rendering the facility obsolete before commercial operation.
The impact of delays goes above and beyond profit losses in a market where technology advances rapidly and the ability to process AI workloads at speed and scale is crucial. However, those risks show up in familiar ways across every stage of the project lifecycle.
1. Planning and Pre-Development
Initial feasibility assessments can have a similar focus to other development projects. Teams may overestimate power availability, rely on incomplete utility assumptions, understate site-readiness challenges, and build schedules around energization dates that have not been validated against interconnection realities, procurement timelines, or permitting requirements.
These risks are not unique to data centers. Major capital projects fail when early assumptions later prove to be incorrect. For example, in the AI-era data center market, where power demands are substantial, insufficient diligence on power availability, environmental conditions, equipment requirements, and local permitting constraints can significantly undermine project feasibility before construction begins.
2. Project Execution and Delivery
Risks during execution are also similar to other complex projects. Realistic schedules still depend on accurate projections for the procurement of long lead time items like transformers, generators, switchgear, chillers, and controls packages. Project delay and increased costs are resulting from labor availability — especially shortages in the skilled trades — incomplete design coordination, weak project controls, and poor planning across trade disciplines.
However, AI-era data centers can be highly compressed technical capital projects. Small delays can quickly lead to significant schedule delays and overall cost increases given critical infrastructure systems, complex mechanical, electrical, and plumbing (MEP) design, fixed interim milestones for phased turnover, and compressed project delivery timelines.
3. Facility Operations
A data center does not generate revenue when construction is substantially complete. Revenue generation of a data center comes with how quickly the first token is achieved or how quickly the facility can process AI workloads.
Data centers require a disciplined transition from construction completion to operational readiness. When that transition is weak, the result is familiar: underprepared operations teams, unstable systems, incomplete or missing documentation, unplanned and deferred maintenance, and unresolved interface issues between construction and facility operations. A 2025 report published by the Uptime Institute showed that power-related issues remain the leading cause of significant data center outages, accounting for nearly half of serious incidents. The outages are most often associated with the facility’s uninterruptable power supply (UPS) system.[2]
The 2025 data from the Uptime Institute report suggests that most outages were preventable. Nearly 87% of respondents who experienced an impactful outage in the last three years stated it could have been avoided if better management processes were in place.[2]
Where AI-Era Data Center Delivery Differs
Data center construction is not new, but some of its governing principles have changed in ways that materially affect strategy, delivery, and overall risk. Power availability now drives site selection and schedule credibility, AI workloads are changing technical design assumptions, speed to first token is becoming the driving milestone, and partnership models are evolving to assemble the capital, power, land, and execution capability required by these fundamental shifts.
4. Power Is the Controlling Constraint
For AI-era data center projects, the primary constraint is often no longer access to capital or customer demand, it is access to a large amount of power. Analysis from the Lawrence Berkeley National Laboratory shows that median time from interconnection request to commercial operation for new power generation projects is now exceeding five years, meaning that access to power can control the critical path to first token.[3] Today’s hyperscale AI-driven data centers require vastly higher power capacity than other project types, often by a factor of 10X or more, rivaling heavy industrial plants. For context, a large data center in 2010 had a capacity on the order of 30 MW and a major hospital campus has a capacity on the order of 15 MW.
Figure 1 – Design Power by Facility Type
This amount of power requires grid planning, interconnection studies, and possibly new on-site generation or dedicated renewable energy projects to meet demand without compromising reliability or climate goals. Key power agreements, and the stakeholder disputes that arise from them, are typically governed by interconnection agreements, service level agreements (SLAs) between operators and tenants, power purchase agreements (PPAs) addressing energy supply, long-term capacity commitment agreements, and long-term service agreements (LTSAs).[4]
Even when a data center project secures an interconnection agreement, the hardest work often begins after entering the queue for receiving and being able to utilize the power. For example, in Florida, Senate Bill 484 — signed May 7, 2026, and effective July 1, 2026 — creates the regulatory framework for “large load customers,” defined to include any data center with anticipated peak loads of 50 MW or more. The law preserves local control over land-use decisions, mandates impact studies, tightens water-use rules, and explicitly prevents utilities from shifting capital cost of data center infrastructure onto residential ratepayers.[5]
That shift fundamentally changes the logic of development: A site is not truly viable because land is available and fiber is nearby; it is only viable if power can be secured and delivered while providing minimal impact to the ratepayers.
5. AI Workloads Changes Design Assumptions
AI workloads have shifted traditional data center designs toward a hyperscale infrastructure model, where improvements in efficiency are now being leveraged to meet the intensive requirements set upon AI-era data centers. The Uptime Institute has tracked power usage effectiveness (PUE) since 2007.[i] As seen in Figure 2 of the 2025 annual report, improved PUE has contributed to better energy efficiency in data centers.
Figure 2 – Weighted Average Annual PUE (2007-2025)
The report indicates that AI workloads require higher rack densities, more intensive cooling requirements, and greater electrical complexity. As a result, many of the assumptions and specifications from previously constructed data centers are outdated. Designs standards that were acceptable for lower-density computing environments of typical enterprise and colocation data centers no longer meet the increased requirements for hyperscale facilities being built to support AI workloads.
The scale of hyperscale facilities justifies significant investment in efficient power and cooling infrastructure that provide cost savings at scale. This increased complexity in the AI-era means design decisions carry greater consequences than in traditional data center projects. As a result, owners and investors must treat early design decisions as strategic planning decisions with long-term project implications, not just value engineering.
6. Speed to First Token as an Industry Metric
Where the critical path on complex projects is through known or typical milestones (e.g. First Fire and Commercial Operation at power plants are typically critical), the critical path on data centers is measured to the ability to achieve first token. Speed to first token, the point at which a facility can begin processing AI workloads has become an industry-wide metric to measure the value of AI-era data center projects. Schedule impacts directly influence how fast the facility can begin to process AI workloads and ultimately start generating revenue.
The financial stakes behind the achievement of first token are substantial. Industry analysis from CMIC Global indicates that commissioning delays on a typical 60 MW facility can translate to roughly $14 million per month in lost revenue and related impacts.[6] This reality has changed how owners evaluate project decisions where procurement, phasing, modular construction methods, commissioning protocols, and partner selection are now assessed not only by whether they reduce construction risk, but by whether they accelerate time to first token.
7. Partnership Models Are Changing
The final major shift is structural rather than technical. Ankura’s Joint Ventures & Partnerships practice has found data center development increasingly depends on partnership models that bring together power access, capital, land, operating capability, and execution expertise in combinations that a single party often cannot assemble alone.[7] Ankura’s analysis of more than 100 AI-infrastructure transactions from 2021 to 2025 shows that joint ventures (JVs) now account for over half of all strategic partnerships in the sector, following a 429% surge in 2024.
This is a meaningful change in market organization. Even so, it does not eliminate familiar risks. Governance deadlock, timeline optimism, misaligned incentives, and unclear accountability remain just as significant in a JV as they are in any other capital-intensive partnership. Successful data center delivery depends on structuring those relationships correctly from the beginning.
8. External Risks Require Agility
Owners, lenders, and developers must consider external risks that evolve rapidly in today’s market. Regulatory and legal challenges surrounding data center siting and construction face mounting pressure from conflicting federal, state, and municipal requirements. There is a growing movement to prohibit data centers within certain municipalities or states, creating the potential for significant expenditures when late-stage regulatory obstacles emerge for previously approved sites.
The impact of this regulatory uncertainty is measurable. According to a CBRE Group Inc report in February 2026, U.S. data center construction capacity declined from 6.35 GW at the end of 2024 to 5.99 GW by the end of 2025, marking the first downturn since the surge of AI demand in 2020.[8]
Sources:
- McKinsey & Company, “The Cost of Compute: A $7 Trillion Race to Scale Data Centers,” April 28, 2025, https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-cost-of-compute
- Donnellan, D., Lawrence, A., Bizo, D., Rogers, O., Judge, P., Davis, J., Smolaks, M., & Weinschenk, R. (2025). Uptime Institute global data center survey results 2025. Uptime Institute Intelligence. https://uptimeinstitute.com/resources/research-and-reports/uptime-institute-global-data-center-survey-results-2025
- Lawrence Berkeley National Laboratory. (2025, December). Queued up: 2025 edition — Characteristics of power plants seeking transmission interconnection. https://emp.lbl.gov/publications/queued-2025-edition-characteristics
- Barratt, James, (2026). The Coming Wave of Disputes in Data Centre and AI Infrastructure, JD Supra. https://www.jdsupra.com/legalnews/the-coming-wave-of-disputes-in-data-7224642
- Florida Legislature. (2026). CS/CS/SB 484: Data centers (Chapter No. 2026-65). https://www.flhouse.gov/Sections/Bills/billsdetail.aspx?BillId=84225
- CMIC Global. (2026). Key data center construction trends in 2026. https://cmicglobal.com/resources/article/data-center-construction-trends
- Ankura Consulting Group. (2026). Joint ventures as the operating system of AI infrastructure: Partnering for speed and scale in a constrained world.
- CBRE Group Inc. (2026, February 24). North America data center trends H2 2025. https://www.cbre.com/insights/books/north-america-data-center-trends-h2-2025
[i] PUE was introduced in 2007. It is a metric that is used to track a facility’s energy efficiency overtime. PUE is calculated by dividing the total power of a facility by the power consumed by the facility’s IT infrastructure.
© Copyright 2026. The views expressed herein are those of the author(s) and not necessarily the views of Ankura Consulting Group, LLC, its management, its subsidiaries, its affiliates, or its other professionals. Ankura is not a law firm and cannot provide legal advice.
