The market reality: power, not square feet
If you are sourcing AI-ready capacity in 2026, your primary constraint is no longer real estate, construction labor, or even capex.
It is power that can be delivered on time, at a predictable cost, under a rate structure that does not blow up your P&L.
Two signals from the past week show how fast this has become the central issue.
First, US data center construction is showing signs of friction even as demand accelerates. CBRE reported that US capacity under construction fell to 5.99 GW at year-end 2025, down from 6.35 GW at year-end 2024, the first decline since 2020, driven by delays in permitting, zoning, and power procurement (Los Angeles Times).
Second, the “who pays” question is moving from utility commission dockets into national politics. On March 4, Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI signed a White House-brokered Ratepayer Protection Pledge committing to build/procure/fund new generation and pay for grid upgrades needed to serve their data centers, with the stated goal of not shifting these costs onto other ratepayers (POWER Magazine).
Even if you are not a hyperscaler, these developments matter.
- Constraints are real: permitting and power procurement are now gating timelines in major markets.
- Cost allocation is changing: tariffs, special rate classes, and new contracting norms increasingly aim to assign incremental grid costs to large loads.
- Procurement must adapt: the classic “pick a market, then pick a site” playbook underperforms when interconnection and rate design are the binding constraints.
This article is an enterprise procurement playbook for buying AI capacity under these conditions.
Construction is slowing in legacy hubs
In the CBRE data cited by the Los Angeles Times, construction dropped materially in traditional hubs—down 29% in Northern Virginia, 15% in Hillsboro, and 14% in Silicon Valley—while growth moved to markets like Chicago (+169%) and Dallas-Fort Worth (+15%) (Los Angeles Times).
The implication for enterprise buyers is straightforward: if your internal roadmap assumes you can always “just add MW” in the same two or three legacy markets, your plan is likely to fail.
Your portfolio will become multi-market by necessity.
Policy pressure is shifting who pays for grid upgrades
The Ratepayer Protection Pledge includes a critical commercial concept: signatories will negotiate separate rate structures and commit to pay for the power and related infrastructure brought online to serve their data centers, whether they use the electricity or not (POWER Magazine).
This is not a niche clause. It is the future direction of large-load contracting.
For enterprise tenants, it foreshadows a world where:
- utilities and regulators push for minimum demand charges, collateral, and multi-year commitments, and
- landlords and developers increasingly ask tenants to share interconnection and delivery upgrade costs.
In parallel, there is real skepticism about what a federal pledge can enforce, since most implementation power sits with state utility regulators (WIRED).
That uncertainty creates a new category of risk: policy-driven volatility in rate design and cost recovery.
A procurement “north star” for 2026: secure power first, then negotiate the building
In a constrained market, your procurement sequence matters.
If you start with space and location preferences, you will waste cycles on options that cannot secure power inside your decision window.
Instead, treat power as the scarce asset and build the procurement around it.
The three questions every CIO/CFO should ask before issuing an RFP
What is the earliest date we need firm MW delivered, and what is the “drop-dead” date? – This determines whether you should prioritize existing powered shells, near-term colocation, or build-to-suit.
What is our tolerance for rate variability and pass-through costs? – In many markets, capacity, transmission, and interconnection costs can move materially.
What operational flexibility do we need (and what can we trade away)? – Examples: curtailment participation, backup generation, or staged ramp schedules.
When these three are clear, you can run a power-first shortlist process that produces bankable options.
The 8-line-item risk checklist for AI capacity contracts
Below is a risk checklist we use when reviewing colocation, wholesale, and build-to-suit terms for AI capacity.
1) Power delivery and curtailment rights
As grids tighten, curtailment programs and load management are becoming structural.
Contract for: – explicit curtailment triggers, – compensation mechanics, – notice periods, and – whether your operations can use on-site backup during curtailment.
The White House pledge itself references coordination with grid operators to make backup generation available during system emergencies (POWER Magazine). Enterprises should assume curtailment language will increasingly appear in tariffs and leases.
2) Cost pass-throughs (capacity, transmission, and interconnection)
Many enterprise buyers underwrite deals using a “blended $/kW all-in” assumption.
That is not sufficient in 2026.
Break out: – capacity cost exposure, – transmission upgrade allocation, – distribution upgrades, – interconnection study and construction charges, – and any “true-up” mechanisms.
If your landlord cannot explain the rate class and cost recovery pathway, treat it as a risk flag.
3) Schedule realism
The Los Angeles Times describes power procurement and permitting as the gating items behind the construction decline (Los Angeles Times).
Translate that into contracting discipline:
- require schedule milestones tied to utility deliverables,
- include liquidated damages aligned to business impact,
- and insist on a transparent critical path.
4) Cooling and water constraints
AI density discussions often fixate on cooling technology.
The bigger enterprise risk is water availability, permitting, and community acceptance.
Require: – water source clarity, – cooling mode specs (air, liquid, hybrid), – and constraints under extreme temperature days.
5) Upgrade scope (substation, feeders, transformers)
Most deal slippage hides inside “utility upgrades.”
Define: – what is in-scope vs out-of-scope, – who owns procurement risk for long-lead electrical gear, – and the handoff between utility and EPC.
6) Optionality: expansion rights and step-in rights
If you are securing a scarce MW tranche, you want options.
Negotiate: – rights of first offer on adjacent capacity, – step-in rights if the developer fails, – and conversion rights between product types (colo → wholesale → BTS).
7) Credits, security, and “pay whether you use it” terms
The pledge’s “pay whether you use it” framing is effectively a take-or-pay concept (POWER Magazine).
Enterprises should expect more: – minimum demand structures, – collateral requirements, – and credit enhancement requests.
Model it explicitly.
8) Auditability and proof of compliance
Because state regulators and politics are now involved, you should assume counterparties may make claims about cost isolation or compliance that are hard to verify.
Require: – reporting rights, – audit rights, – and copies of relevant tariff orders or utility agreements.
How enterprises should re-structure the RFP process
Replace “site-first” with a “power-first shortlist”
A practical approach:
- Build a shortlist of markets based on utility posture and feasible interconnection timelines.
- Pre-qualify options by confirming: available MW, delivery window, upgrade scope, and rate class.
- Only then evaluate: latency, tax, labor, and physical security considerations.
This sequence reduces the amount of time your team spends evaluating non-viable sites.
Pre-negotiate utility posture and rate class alignment
Your landlord can sell you a building.
They cannot always sell you a workable power contract.
Given the increasing focus on separate rate structures for large loads (POWER Magazine), enterprises should:
- request documentation on the applicable tariff / rate class,
- understand whether a special contract is required,
- and ask what assumptions the utility is making about your ramp.
Use a two-track strategy: near-term colocation + medium-term build-to-suit
In a constrained environment, many teams need capacity in two horizons:
- 0–18 months: “good enough” capacity to keep product and model roadmaps on track.
- 18–36 months: bespoke capacity with better economics.
A two-track strategy avoids betting the business on an uncertain utility delivery date.
What this means for board-level planning (and how Stacked AI helps)
Boards increasingly ask two questions:
- “Are we secure on compute?”
- “Are we exposed to rate shocks or community backlash?”
Answering both requires the same capability: an independent view of where power is feasible, how it will be priced, and how risk is allocated in contracts.
Stacked AI supports enterprise buyers and investors with:
- power-first market shortlists,
- contract risk reviews (tariffs, pass-throughs, curtailment),
- and supplier orchestration across colo, wholesale, and build-to-suit pathways.
If you are planning AI capacity for 2026–2028, the most valuable output is not a site tour schedule.
It is a bankable plan for MW delivery.