The Hidden Line Item in Your 2026 AI Bill
The six-question Power Pass-Through Audit to send your AI vendor before you renew.
So here's the line nobody put in your AI budget: electricity. Gartner just said data center power demand jumps 26% this year, and they were blunt about who pays — those costs get passed down to the companies selling you AI, and then to you. Your token price looks fixed. It isn't. It's riding on a power bill that's climbing fast, and your vendor gets to decide how much of that lands on your invoice. The move below turns you from someone who reacts to a price hike into someone who priced the risk before signing.
Paste this into an email to your AI or cloud vendor before your next renewal:
Subject: Power and pricing terms — [CONTRACT NAME] renewal
Before we renew, I need clarity on how energy costs affect our pricing.
Please answer:
1. What portion of our unit price ([$/1M tokens] or [$/GPU-hour]) is
exposed to your energy costs, and how has that shifted in the last
12 months?
2. Do our rates include any energy or power surcharge — now, or
contractually allowed later? If so, what caps apply?
3. Where does the compute for our workload physically run, and what is
that region's grid reliability and power-cost trend?
4. If you hit a power shortage, what is our priority tier for capacity
versus [ENTERPRISE TIER / other customers]?
5. Can we lock current pricing with a [24–36 month] commit, and what's
the discount for signing before [RENEWAL DATE]?
6. What's your contracted power plan (nuclear PPA, on-site generation),
and when does it come online?
Sample output (a straight answer tells you exactly where you stand):
"~40% of your rate tracks our energy and hosting costs; those rose ~18%
last year. No surcharge today, but the MSA permits a pass-through above
a 20% power-cost increase. Your workload runs in [REGION], grid
constrained. A 30-month commit locks current rates at 12% off if signed
before Sept 30."
What changes: You stop treating the AI bill as a fixed SaaS line and start treating it as an energy-exposed contract — one you can cap, lock, or move before the increase hits.
Where else this works: GPU cloud renewals, hyperscaler committed-use agreements, colocation and data center leases, and any managed-inference vendor repricing on 2026 power.
On June 10, Gartner put data center electricity demand up 26% for 2026 — to 565 TWh — and said operators will use their leverage to secure power and pass the rising cost down the chain to AI providers, and to you. The vendors already know their power exposure. This is the week to make them tell you yours.
Your AI Sherpa,
Mark R. Hinkle
Founding Publisher, The AIE Network
Follow me on LinkedIn
