Capital is flooding into AI data center development at a pace that has few precedents in digital infrastructure. Hyperscalers, developers, and financial sponsors are committing tens of billions of dollars to campuses that, in many cases, did not exist as a line item in anyone’s capacity plan three years ago. Much of the investor scrutiny on these deals has understandably focused on demand durability, tenant credit quality, and the risk of stranded compute as chip architectures evolve. Power has moved from an afterthought to the central constraint in data center investment decisions, and it is often the single largest determinant of whether a project reaches energization on schedule, and at what cost.
That gap is what prompted me to formalize a framework I have been using informally across several site evaluations. This framework scores grid and energy risk for large baseload data center projects and applies consistently across candidate sites and portfolios. What follows is a summary of that framework, along with the reasoning behind it.
Why Power Risk Is Different for AI Loads
A conventional large industrial customer, such as a smelter, a refinery, or a mine, draws power against a demand curve that flexes with production schedules, shift patterns, and market conditions. Grid operators have decades of experience integrating loads like these, and the load itself offers some give.
A hyperscale AI campus does not behave this way. It runs at or near full load continuously, 24 hours a day, 365 days a year, with essentially zero tolerance for curtailment once training or inference workloads are running. That is a fundamentally different integration problem for a grid operator, and it means the standard due diligence questions that apply to conventional industrial siting are necessary but not sufficient. A site can look attractive on paper, with proximity to fiber routes, favorable land costs, and a receptive local jurisdiction, and still carry a materially higher power risk profile than a superficial reading would suggest.
Ten Categories of Grid and Energy Risk
The framework scores each site across ten categories, weighted according to how much each one tends to drive actual delivery risk in practice. None of these are exotic. What matters is asking them consistently and specifically enough that the answers are comparable across sites.
1. Load to grid ratio
What percentage of the local or regional peak demand, not the broader national or continent wide system, does the new load represent? A 300 MW facility is immaterial to a grid with tens of gigawatts of capacity, but it can be a defining event for a constrained regional pocket that imports a meaningful share of its own power. The relevant denominator is almost always smaller and more local than the headline system capacity figure a developer will cite.
2. Baseload dispatchability
Not all committed capacity is equal. Firm, dispatchable resources such as hydro, gas, and nuclear behave very differently from intermittent wind and solar when a facility needs guaranteed output every hour of every day. A supply plan that leans on capacity additions without a firm commissioning date, or on legacy generation that has been kept in reserve status under regulatory or environmental review, carries meaningfully more risk than one built on contracted, operating capacity.
3. Timing alignment
This is the category investors most often underweight. It is not enough that the generation and transmission upgrades a project depends on will eventually be built. The real question is whether they will be built before the facility energizes. Projects where new supply, new transmission, and a large new load all land in the same window are absorbing simultaneous risk rather than sequential risk, and that distinction matters enormously when something slips.
4. Transmission and interconnection adequacy
A dedicated substation at the fence line solves the last mile, not the whole problem. The more relevant questions sit upstream. Is there a known constraint feeding that substation? Is a reinforcement project actually funded and under construction? Is there redundancy in the path, or does a single line failure take the facility offline?
5. Regulatory and political exposure
Was environmental and regulatory review completed in full, or was a step bypassed, deferred, or rejected on a technicality? Is there an active legal or policy dispute affecting any generation source the project depends on? Political risk to a power arrangement rarely shows up in the term sheet, but it shows up in delivery timelines and in the durability of rate agreements.
6. Price and rate structure volatility
A regulated, utility fixed rate is a structurally different risk than merchant market exposure, and this is one of the clearer differentiators between jurisdictions. Within regulated markets, it is worth distinguishing predictable, capital driven rate increases from commodity indexed pass throughs, since the latter reintroduces volatility that a fixed rate structure was supposed to eliminate. It is worth being precise about what a regulated rate actually protects against. It shields a developer from commodity price swings, but it does not automatically shield a developer from a utility restructuring its own tariff to shift buildout risk onto large new customers. A regulated utility with no generation of its own, facing a wave of speculative interconnection requests, has every incentive to use the one lever it does control, the terms of service, to manage that risk. The result can look a great deal like merchant market exposure even inside a fully regulated jurisdiction, just relabeled as a service tariff rather than a spot price.
7. Contractual commitment structure
This is a distinct risk from rate volatility, and it deserves its own line of diligence. The relevant questions are about the terms a utility or grid operator requires before it will guarantee service, not about the price of the electricity itself. Does the agreement require a minimum monthly bill based on contracted or historical peak demand, regardless of actual usage? Is there a multi year take or pay commitment, and how long is it? What exit fee applies if the project is delayed, resized, or cancelled after the utility has begun construction? Is collateral required from the developer or its financial sponsor? These terms can turn a nominally low risk, regulated jurisdiction into a high commitment one in practice, and they are increasingly common as utilities across North America respond to a surge of data center interconnection requests that may not all materialize. An investor should treat these contract terms with the same scrutiny given to a lease term or a debt covenant, since they function the same way, as a binding financial obligation that survives even if the underlying project economics change.
8. Backup power and carbon or emissions liability
On site backup generation is usually the single most important mitigant against grid and regulatory risk, since it can render a facility self-sufficient in an outage. But backup capacity is not a free option. Chronic use of fossil fueled backup can trigger material carbon costs and undercut a sponsor’s sustainability commitments. Backup should be modeled as insurance with a cost attached if it is ever drawn on heavily, not simply as a mitigant that neutralizes risk elsewhere in the plan.
9. Equipment and supply chain lead time
This is the category most site evaluations still miss, and it now sits upstream of almost everything else on this list. Gas turbine wait times have stretched from a historical 1 – 3 years to as long as 7, with prices up sharply for units delivering at the end of the decade. Large power transformers, once available in under 2 years, now carry lead times of 4 – 5 years, constrained by a small number of global manufacturers and a raw materials supply chain under pressure from every other sector electrifying at the same time. A committed generation plan or a well-designed backup strategy means little if the physical hardware has not already been ordered and slotted into a manufacturer’s production queue. This risk applies as much to a facility’s own on-site backup turbines as it does to utility-scale generation, which means it can quietly undermine the backup power mitigant described above. The question to ask is not whether a generation or transmission plan looks sound on paper, but whether the equ`ipment behind it has an actual delivery date from a manufacturer.
10. Scale and precedent risk
Has the grid operator successfully integrated a load of this size before, or does this project exceed anything in its operating history? A load that is an order of magnitude larger than any prior industrial customer on that system carries risk that has no track record to lean on, regardless of how sound the underlying engineering looks.
From Qualitative to Comparable
Each category is scored on a simple 1 to 5 scale, with a written rubric behind each score to keep the exercise from becoming a subjective gut check. Weighting and summing the scores produces a single comparable risk rating, low, low medium, medium, medium high, or high, for each site. The rating itself matters less than the discipline behind it. It forces every site evaluation to answer the same ten questions, with the same specificity, which makes comparing a portfolio of candidate sites, or competing offers from different developers, an apples to apples exercise rather than a narrative one.
Concluding Thoughts
A handful of conclusions fall out of building this framework that are worth stating plainly.
The most underpriced risk in data center investing right now may not be the utility relationship at all. It is the assumption that hardware ordered on a normal timeline will arrive on a normal timeline. A site with a difficult regulatory story but equipment already in a factory queue may be a better bet than a site with a clean regulatory story and no turbines or transformers on order.
Regulated markets are not automatically the safe choice, and deregulated markets are not automatically the risky one. What actually matters is which party ends up holding the risk that demand does not materialize as forecast, and that allocation is being renegotiated in real time through tariffs, take or pay contracts, and exit fees, regardless of which market structure a jurisdiction started with.
Most financial models for data center deals are more rigorous about lease durability and tenant credit than they are about whether the power will actually show up on schedule. That is backwards. A lease is a promise between two parties who can renegotiate. A turbine backlog is a physical constraint that no amount of negotiation shortens.
The industry is behaving as if power risk is a site selection problem, solved once at the diligence stage. It is increasingly a portfolio management problem, since the same equipment scarcity, the same regulatory repricing, and the same grid constraints are showing up across supposedly unrelated sites at the same time. A framework built to score one site in isolation should also be read across a portfolio, to see whether the same risk is quietly concentrated in more places than it first appears.
If you are working through pre-investment technical and commercial due diligence on a data center opportunity, whether power, site selection, contractual structure, or the broader investment case, feel free to reach out. I am happy to compare notes.