By Paul Hirst, Richard Palmer and Sanjeev Bahl
Data centres in the European Union consumed an unprecedented volume of electricity in 2024, usage that is expected to rise more than 60% by 2030[1], with the increase equating to the annual electricity consumption of several major European cities combined. AI is disproportionately driving this growth, with AI-related data centre power demand projected to grow 47.9% annually between 2023 and 2027.[2]
This surge in energy demand, and AI’s role in driving it, is straining the systems that data centres depend on, and crucially, the deals that govern them. Operators are relying more heavily on AI-specific infrastructure that must scale fast and run near-continuously, just as power supply, hardware availability and construction capacity are coming under pressure. Together, these variables are complicating the processes surrounding site selection, contract negotiations and managing delivery risk.
Limits of the old risk model
With conventional data centre contracts, each risk is typically allocated to whichever party is best placed to manage that risk. The developer handles construction and delivery. The utility deals with grid connection. The equipment suppliers bear responsibility for their own timelines. Insurance and liability caps are then added to absorb the rest. Increasingly, though, this model is being tested. Pressures like site readiness, grid connection delays, commissioning constraints and supply chain timelines affect multiple parties simultaneously. Risk allocation remains important, but contracts need mechanisms that manage the interdependencies between parties from the outset.
Traditionally, the operator commits to performance targets, such as uptime, network connectivity or power availability, under service level agreements. Those targets are usually described in measurable terms, if uptime falls below an agreed threshold (for example, 99.9% in a given month), the customer receives a credit calculated against that month’s service or rack fees. Service level agreement credits are formulaic and often defined in the contract itself, so both sides understand roughly what a lapse in service will cost. The credits are designed to compensate for small, contained service interruptions, not lost revenue. For example, an outage claim is a fairly contained event and resolved through credits against future bills rather than large cash damages. Historically, this has been sufficient.
But fast forward to the data centres of tomorrow and the picture looks different. Claims against operators could stem from a variety of sources, such as a power outage or a delay in commissioning. As the value of AI compute itself – namely tokens, the basic unit of AI processing – become tradable, the result is a potential loss of revenue throughout the value chain, from energy supplier through to data centre, developer, tenant and customer.
As the value of claims rises, the service level agreement credit model may provide an insufficient remedy. Any damages calculation would need to account for the pricing schedule for tokens, the technology used, utilisation of facilities at the time of an outage and the ability to mitigate a loss by sending processing tasks elsewhere. Even a seemingly straightforward claim would be a complex exercise.
Measuring losses based on lost compute revenue could significantly increase claim sizes. This risk isn’t the only thing that has changed, the way in which these facilities are built and delivered has changed too.
Construction and delivery
AI data centres are far more complex than the data centres of 10 to 15 years ago, which were fairly small and stable in design and requirements. Developers or data centre operators were able to appoint a single company to deliver the project. Now, the scale and money involved mean getting a data centre online is far from straightforward. For grid connection alone, operators often need to consider behind-the-meter arrangements, such as gas turbines, battery storage and nuclear power, complicating the contracting approach.
Data centres can take years to build; meanwhile, the technology required continues to evolve. In turn, the design, the racks and the cooling all need continual updates. Likewise, a single, fixed-price contract for a build will be slow, costly to vary and difficult to update to meet changing design requirements.
Now, operators procure individual parts of a data centre build, which gives them the flexibility and speed to adapt to ongoing changes to the requirements. But that flexibility comes at a price, as the operator must integrate those components to create the whole facility and owns increased delivery risk.
In short, this exposes limitations in the conventional single point, single contract risk transfer model. Increasingly, whether a data centre can actually deliver depends on factors outside any one operator’s control — energy availability, hardware supply, supply chain resilience, grid capacity — not just how well construction is managed. Compliance and performance increasingly depend on the wider delivery system, rather than the performance of individual assets in isolation.
A new governance model
Tightly drafted contracts remain critical, but operators also need more internal delivery capability to manage this increased risk directly in this new world. A coordinated risk management system supports this: an approach built around coordinating interdependencies across construction, energy, hardware and delivery. This coordination allows the project to adapt as conditions change, rather than relying on any single fixed contract to absorb the risk.
A key mechanism for this is stage-gate governance, where work is divided into phases with a gate, or formal review point, between each. These gates force different workstreams to align before the project moves forward, driving better integration and a more robustly developed project. In turn, there are fewer integration issues once construction is under way. In the early phase, requirements gathering, finance raising and project costing must be developed together, with each informing the others, before the project can pass through the first gate. At each gate, operators should ask three questions: Are critical dependencies still achievable? Does risk remain with the party best able to manage it? And have changes elsewhere in the programme created new exposures?
In addition to stage-gate governance, operators need robust supply chain management arrangements and programme controls to act as early-warning systems to monitor costs, risks and schedules in real time.
Towards a more fragmented market
It is likely there will be a much more fragmented contracting approach rather than a single design and build or turnkey solution. As outages start to translate into meaningful revenue loss rather than the contained disruptions traditional service level agreements were designed for, operators face a new problem: That loss can exceed what any individual vendor can absorb or is proportionate to that vendor’s share of the project’s value. Fragmentation also adds its own burden. Splitting a project across more contracts and stage gates means constant coordination, resetting schedules and renegotiating terms whenever one party falls behind. It also places greater emphasis on interface management between contracts, as disputes may increasingly arise where obligations fall between multiple contractual arrangements rather than from the failure of a single contractor.
AI data centre pipelines now exceed the scale of portfolios being delivered in other capital-intensive sectors — hundreds of billions of dollars a year for each of the major players, from hyperscalers through to specialist operators.[3] At that scale, a strong client-side delivery capability will be crucial to winning in this market and minimising the downside risks inherent in building projects of this complexity.
The contracts being signed today will govern disputes that play out under different conditions. Understanding how loss is defined, how risk is allocated and where liability sits across a fragmented contractor base has become a critical and an underused discipline. For law firms advising operators, developers or tenants, that understanding is becoming as important as the contract terms themselves.
Authors
Paul Hirst is a chartered civil engineer with 25 years of industry and consulting experience supporting clients to deliver some of the world’s largest construction projects. As a Senior Managing Director in FTI Consulting’s Construction, Projects and Assets practice, Paul enables clients to improve the set up and delivery of their complex capital projects. His expertise includes delivery model selection, transforming program controls, delivery team set up, project turnaround and optimization and helping clients to harness technology to improve delivery. The team is currently supporting a number of data centre developers to scale up their delivery set up to keep pace with their expanding capex portfolios.
Richard Palmer brings more than 20 years of experience to FTI Consulting’s Technology segment. Focusing on modernisation and the use of advanced technologies to transform legal operations, Richard works with global corporate organisations to deliver against business case objectives for reducing cost and risk. Richard has expertise supporting widespread organisational objectives in digital transformation, leveraging industry-leading legal, risk and compliance-focused software-as-a-service platforms. He serves as a trusted partner to his clients, often supporting multiyear managed services engagements that deliver ongoing value, with a focus on business transformation and reducing data-related risks.
Sanjeev Bahl is an expert witness specialising in the quantification of losses and asset valuation across the global energy sector. With more than 20 years of experience, Sanjeev has been appointed and prepared expert reports in high-value arbitrations before ICC, LCIA, UNCITRAL, PCA and SIAC tribunals. His expertise spans the oil and gas value chain, as well as emerging sectors such as offshore wind, hydrogen and carbon capture, utilisation and storage (“CCUS”).
*This article includes contributions from Chase D’Agostino, Managing Director, FTI Technology.
References
[1] “In focus: Data centres – an energy-hungry challenge,” European Commission (November 17, 2025) https://energy.ec.europa.eu/news/focus-data-centres-energy-hungry-challenge-2025-11-17_en.
[2] “Current Power Trends and Implications for the Data Center Industry,” FTI Consulting (June 14, 2024) https://www.fticonsulting.com/insights/articles/current-power-trends-implications-data-center-industry.
[3] “Sector Review: U.S. Tech Earnings: Hyperscalers Again Are Hyperspending,” S&P Global (February 12, 2026) https://www.spglobal.com/ratings/en/regulatory/article/sector-review-us-tech-earnings-hyperscalers-a-s101669934.