RFP Template: Who Pays for Power? Data Center Contracts in the Age of AI

Who Pays for Power? How to Write RFPs and Contracts that Allocate AI-era Energy Costs

Hook: If your operations team is buying colocation or cloud capacity for AI inference or training, you’re facing a new cost reality: power and demand charges that can explode your TCO. In 2026, utilities and regulators have tightened the rules — and some regions now make data centers bear the incremental cost of new grid capacity. The result: unclear allocation of who pays for new feeders, interconnection upgrades, batteries, and the steep demand charges that come with GPU farms. This article gives operations leaders a ready-to-use RFP and contract playbook to allocate energy responsibility without sacrificing performance.

Executive summary — What you need to build into your RFPs right now

In one paragraph: demand charges and grid upgrade costs are increasingly being shifted to data center customers and owners. To protect your budget, include explicit clauses in RFPs and contracts that define:

• Metering & telemetry — who measures the demand and with what granularity.
• Demand charge allocation — a transparent, auditable formula for dividing demand charges and coincident peaks.
• Grid upgrade & interconnection cost sharing — cap, amortization, and triggers for customer contribution.
• SLA for committed kW/kVA with burst allowances for AI loads and remedies for shortfalls.
• Change-in-law/change-in-regulation and price reopener mechanics (essential in 2026).
• Energy management obligations — workload scheduling, load curtailment coordination, and DR participation.

Why this matters in 2026: macro trends reshaping power responsibility

Late 2025 and early 2026 accelerated two trends that directly affect RFPs: first, regulators in key markets (PJM and several state utilities) signaled that incremental grid investments tied to data center build-outs should be funded by the beneficiaries — i.e., data center operators and large customers. Second, utilities increasingly rely on demand charges and time-of-use (TOU) pricing instead of pure volumetric rates to recover costs. For AI-heavy loads, this is a double hit: sustained high draw during training spikes both increases kWh and creates large coincident peaks that drive demand charges.

Practical effect for buyers: an initially attractive $/kWh quote can hide multi-thousand-dollar monthly demand charge line items and even capital allocations for substation or feeder upgrades. If your RFP doesn’t make vendors commit to measurement, allocation and caps, you could be billed for building a new substation.

Power cost components to address in contracts

Make sure every RFP itemizes these components so vendors respond with clarity:

• Energy (kWh) — the volume charge, often time-varying.
• Demand charges (kW) — billed on peak interval (15-min or 5-min) and often the dominant cost for AI workloads.
• Capacity & ancillary services — charges passed through by ISOs for capacity obligations.
• Interconnection and grid upgrade costs — capital contributions, one-time fees, or amortized customer charges.
• Resilience costs — on-site generation, batteries, and their O&M or capital recovery.
• Environmental attributes — RECs and carbon accounting (important for ESG clauses).

Practical RFP & contract clauses — language to include today

Below are sample clause headings and concise, negotiable language you can paste into your RFP or contract draft. Treat these as templates: have counsel review and adapt to local law.

1. Metering, telemetry, and measurement standards

Clause objective: Provide a single source of truth for demand and energy measurement so charges are auditable.

Sample clause: Provider shall install and maintain ANSI/IEC-compliant interval metering capable of recording energy (kWh) and demand (kW) in five-minute intervals. Meter readings, timestamped and remote-accessible, shall be delivered daily via secure API. Provider and Customer shall accept this meter as the sole basis for energy and demand billing unless proven inaccurate by >0.5% in a jointly agreed meter test.

2. Demand charge allocation & coincident peak definition

Clause objective: Define how monthly demand charges and ISO coincident peaks are allocated when multiple tenants or customers share infrastructure.

Sample clause: Monthly demand charges assessed by the utility shall be allocated to Customer based on the Customer’s contribution to the facility’s coincident peak, calculated using five-minute interval data for the utility-defined billing window. Where the Provider aggregates multiple customers, Provider shall provide a transparent allocation worksheet and allow the Customer to audit allocation up to twice per year.

3. Grid upgrade, interconnection and capital cost sharing

Clause objective: Prevent surprise capex bills for new substations or feeders.

Sample clause: Provider shall notify Customer in writing within 10 business days of any utility requirement for interconnection or grid upgrades triggered by Customer’s requested capacity. Provider may not invoice Customer for capital upgrades unless: (a) an independent engineering study demonstrates the upgrade is solely attributable to Customer’s incremental load above a mutually agreed baseline; and (b) Provider documents an amortization schedule and caps Customer’s share to [X]% of total upgrade cost or $[Y] per kW, whichever is lower.

4. Demand charge cap and true-up mechanics

Clause objective: Limit Customer exposure while allowing Providers to recover legitimate incremental costs.

Sample clause: Provider may pass through demand charges attributable to Customer up to a monthly cap equal to [Z] times Customer’s committed kW. Amounts above the cap will be borne by Provider for the first [N] months. Parties shall true-up charges quarterly and reconcile any disputed allocation within 30 days.

5. SLA: Committed power, burst allowances, and remedies

Clause objective: Guarantee availability of contracted power and specify financial remedies for shortfalls.

Sample clause: Provider warrants continuous delivery of up to [A] kW (the "Committed Power") per Customer deployment with an additional burst allowance of [B]% for [C] minutes per 24-hour period. Failure to deliver Committed Power due to Provider-managed faults shall accrue service credits equal to [D]% of monthly service fees for each hour of outage beyond the first [E] minutes, up to [F]% of monthly fees. Provider’s total liability for power availability shall not be limited for damages resulting from Provider’s gross negligence or willful misconduct.

6. Change in law and regulatory pass-through

Clause objective: Protect the customer from new regulations that shift novel costs unless clearly allocated.

Sample clause: If, after Agreement Effective Date, a change in law requires Provider to pay additional charges (including mandatory grid costs assigned to data centers), Provider shall notify Customer within 10 business days. Parties shall meet to allocate new costs; absent agreement, Provider may not unilaterally pass through such costs until an arbitration process is concluded. Any agreed pass-through shall contain an amortization schedule and be subject to Customer audit.

7. Energy management & AI workload coordination

Clause objective: Require cooperation on load shaping, demand response, and scheduling to reduce shared demand peaks.

Sample clause: Provider and Customer shall jointly adopt an energy management plan that includes workload scheduling, demand response participation, and notification windows for planned ramp-ups >[X] kW. Provider shall provide 48-hour forecast windows for potential coincident peaks; Customer agrees to reasonable requests to shift non-time-sensitive AI workloads to off-peak windows, subject to agreed service credits where shifting impairs performance.

Technical specifics every operations team must confirm

Insist on these measurable technical items in your RFP and score vendor responses:

• Interval length for demand measurement — 5-minute preferred (many utilities use 15-min but AI peaks can be shorter).
• Meter accuracy and certification — ANSI/IEC compliance and third-party calibration.
• Definition of peak — rolling maximum vs. utility-determined coinci­dent peak.
• Normalization rules — weather adjustments and baseline normalization when new infrastructure changes underlying load profiles.
• Load diversity and ratchet rules — how historical peaks affect future bills.

Negotiation strategies and levers

Use these levers to trade off cost, risk and performance:

• Longer term commitments can lower capital allocation for grid upgrades; offer multi-year contracts in exchange for lower pass-through caps.
• Co-investment in on-site storage or transformers reduces per-customer capital and gives you control of operational policies.
• Reserve capacity or rights to burst during scheduled windows to control demand profiles and avoid unplanned peaks.
• Shared savings — structure incentives so Provider shares cost savings from energy efficiency projects or peak shaving with Customer.
• Audit and transparency — insist on meter-level data access and monthly allocation worksheets as non-negotiable.

AI workload-specific considerations

AI training runs and inference clusters create a unique mix of sustained high kW and short intense spikes. Your RFP should clarify:

• Committed and burst power per rack or cluster, in kW and kVA.
• How cooling PUE is measured and reported; set maximum PUE targets for billing tiers.
• Coordination for planned training campaigns — notification windows and pre-booked burst windows to avoid coincident peaks.
• Allowances for GPU cold-start or warm-up cycles affecting power draw.

Sample evaluation matrix (RFP scoring)

Score vendor responses on these weighted criteria (example):

1. Metering & data transparency — 20%
2. Demand charge allocation & caps — 20%
3. Grid upgrade policy & caps — 15%
4. SLA for power and remedies — 15%
5. Energy management and AI workload support — 10%
6. ESG/REC and carbon accounting — 10%
7. Price competitiveness (kWh + demand) — 10%

Real-world example (anonymized)

Company X, a mid‑sized SaaS firm running BERT-scale model training, issued an RFP in PJM in late 2025. The initial colocation quotes showed competitive $0.06–$0.08/kWh but also included vague language about utility charges. Using the clauses above, Company X required:

• 5‑minute interval metering with direct API access;
• an explicit cap of $300/kW on any grid upgrade cost allocable to them;
• a burst window of 120 minutes/day at 20% above committed kW;
• Provider-shared savings for any demand charge reductions below a defined baseline.

Negotiation outcome: Company X paid a slightly higher base kW rate but avoided a $1.2M capital allocation for a new substation. They added batteries co-funded 60/40 with the Provider and cut peak exposure by 35% within six months.

Operational playbook: immediate steps for RFP teams

Follow these steps as you prepare procurement documents:

1. Map expected AI workload profiles (peak kW, average kW, burst frequency).
2. Request 5-minute interval metering and API access in the RFP requirements.
3. Insert demand charge allocation, cap and true-up clauses into your standard contract template.
4. Score vendor proposals using the matrix above — don’t over-index on $/kWh alone.
5. Build optional co-investment options (storage, transformers) into your commercial term sheet for negotiation leverage.
6. Have legal draft a clear change-in-law clause that forces vendor engagement rather than unilateral pass-throughs.

Risk checklist — what to watch for in vendor responses

• Vague metering or ‘Provider meters’ with no API access.
• Open-ended language: “Customer may be required to contribute to utility upgrade costs.”
• No true-up or audit rights for allocation of coincident peaks.
• Ratchet clauses that lock you into historical peaks without escape or normalization.
• Providers refusing to support demand response or workload shifting coordination.

Future-proof clauses and 2026-proofing your contracts

Given the policy momentum in 2026, include clauses that survive rapid regulatory change:

• Sunset review: A scheduled contract review every 24 months to reallocate energy costs based on evolving utility tariffs and grid rules.
• Arbitration path: Fast-track arbitration for disputes over allocation of grid upgrade costs.
• Technology reopener: Allow for renegotiation if new feasible technologies (e.g., more efficient GPUs, cheaper batteries) materially change peak profiles.

Final considerations: legal, accounting and ESG

Work with finance and legal early. Capitalizing grid upgrade payments, accounting for shared storage, and tracking RECs require cross-functional alignment. Also, if your company has public ESG goals, negotiate for RECs/attribute transfer or offsets so that energy responsibility doesn’t undo sustainability claims.

Quick take: In 2026, power is no longer a footnote in data center contracts — it’s a line item that can make or break project economics. RFPs that leave measurement, allocation or caps vague will expose you to major financial and operational risk.

Actionable takeaways

• Include five-minute interval metering and API access as mandatory RFP requirements.
• Require explicit demand charge allocation formulas and caps tied to committed kW.
• Push back on open-ended grid upgrade pass-throughs; require documented causality and caps.
• Negotiate SLA language that defines committed power and burst allowances for AI workloads.
• Use co-investment and shared-savings models to align incentives and lower costs.

Next steps — where to get help

If you’re preparing an RFP this quarter, start with the clauses above and run them by your procurement, legal and cloud-ops teams. For vendor selection, use a scoring matrix that weights power transparency and allocation as heavily as $/kWh. If you want a ready-to-run RFP template pre-populated with the clauses above and a vendor shortlist vetted for energy transparency, our marketplace at outsourceit.cloud curates providers with proven experience managing AI workloads and power contracts.

Call to action: Download the full RFP clause pack and evaluation spreadsheet from outsourceit.cloud, or request a vendor shortlist tailored to your region and AI power profile. Don’t let power costs surprise your budget — build the rules into your RFP now.

Related Reading

• Can Big-Name Festival Promoters Turn Dhaka Into a Regional Music Hub?
• Protect Yourself from Deal Scams: How to Verify Deep Discounts on Tech and Collectibles
• Pancake Pop-Ups: How to Launch a Weekend Brunch Stall Using Affordable Tech and Cozy Packaging
• Micro‑Event Idea: Sound + Light Pairings for a Multi‑Sensory Ice‑Cream Tasting
• How Fragrance and Flavor Companies Define 'Fresh' — And Why That Matters for Relaxation Scents