AI’s Power Bill: Making Data Centers Pay for Capacity They May Not Use

An 80% minimum billing demand changes who pays when a data center needs less power than expected. It does not mean the customer has promised to purchase 80% of forecast electricity consumption.

That distinction anchors Michigan’s October 1, 2026 conditional approval of DTE Electric’s Google service contracts. The Michigan Public Service Commission described a 20-year contract and termination protection ensuring at least 15 years of minimum monthly charges in total, including amounts already paid—not another 15 years of payments upon departure. The terms are described in the commission’s announcement and its data-center service guidance.

The economic question is larger than electricity sales: can stronger payment commitments protect existing customers while making speculative capacity reservations less attractive?

The mechanism is plausible. The outcome is not yet established by these contracts.

Capacity is a commitment, not consumption

Electricity consumption measures energy used over time. Billing demand concerns the capacity basis on which demand charges are calculated. A minimum billing demand puts a floor under that calculation even when actual use falls short.

That matters because infrastructure spending can precede computing demand. A utility may need to expand substations, connections or other grid assets before servers begin operating. Once built, those assets carry costs even if the customer’s expansion slows.

Without adequate customer commitments, the resulting exposure must land somewhere: on the utility’s balance sheet, on other customers through approved rates, or on another party under a separate agreement. The allocation depends on contracts and regulatory decisions; households do not automatically inherit every shortfall.

Nor does every large new customer create a burden. FactCheck.org’s October 29, 2026 analysis distinguishes constrained grids, where additional demand can increase prices, from systems with spare capacity, where new customers can spread fixed costs.

The relevant test is therefore incremental revenue against incremental cost—not the size of the data center’s electricity bill alone.

Payment floors need financial backing

Minimum charges, collateral and exit provisions address different risks.

A minimum charge supports revenue while a customer remains connected but uses less capacity than anticipated. An exit provision addresses early departure. Collateral or other financial security addresses collectability: whether a payment obligation can become cash when needed.

These protections complement one another. A substantial termination obligation is less useful if the obligated entity cannot pay. Conversely, collateral may cover only part of the potential exposure. Its amount, duration and draw conditions matter.

This is a framework for evaluating protection, not a claim that the supplied DTE–Google summaries establish every financial backstop. They do not provide enough detail to endorse the arrangements as eliminating all household exposure.

The same caution applies to asset life. A contractual payment floor need not match the full period over which infrastructure remains useful or recoverable. What matters at termination is the remaining cost, the payments available and whether the capacity can serve another customer. Reusable infrastructure presents a different risk from a connection with little alternative demand.

Charging for the option to expand

A capacity reservation gives a developer flexibility: secure access now, decide how much computing capacity to deploy later. If abandoning that reservation is inexpensive, some cost of keeping the option open can fall outside the developer’s own accounts.

Stronger commitments could change that calculation.

A developer expecting to pay for unused capacity has more reason to align its request with a credible deployment schedule. It might reserve less, phase its expansion or choose a location requiring less new infrastructure. The potential gain is better capital allocation before construction, not merely compensation after a forecast fails.

This remains a hypothesis. The cited evidence does not demonstrate that Google has reduced reservations, that speculative applications have declined, or that Michigan’s terms have improved forecasting.

There is also a trade-off. Tougher commitments can disadvantage smaller operators with less balance-sheet capacity, even when their projects are viable. Financial discipline may reduce speculative demand while favouring established companies. Regulators should distinguish protection against unpaid costs from requirements that unnecessarily restrict entry.

Better computing productivity adds another uncertainty. Less electricity per task could reduce a project’s capacity needs, but cheaper computing could also stimulate more use. A durable contract allocates forecasting risk rather than assuming either effect will dominate.

Lifetime benefits are not a billing schedule

In its October 1, 2026 announcement, the commission reported DTE’s projected $1.7 billion benefit over the contract’s 20-year term. That is a company projection, not measured household savings.

A lifetime total cannot, by itself, establish when customers benefit. Infrastructure expenditure may occur early; contributions toward shared costs may accumulate later. Financing costs and the timing of regulatory pass-through affect the result.

Households need an annual reconciliation: what costs are attributable to the project, what revenue offsets them, and how the net amount enters rates. A projected benefit relative to a no-project scenario also differs from an absolute bill reduction. Other system costs can rise at the same time.

Projected savings do not establish realized bill reductions.

The test is recovery, not reassurance

Michigan Attorney General Dana Nessel’s office announced an October 30, 2026 rehearing petition, seeking stronger cost recovery, tracking and financial backstops. Those are the office’s positions, not findings that customers have suffered losses. The petition alone does not settle the contracts’ adequacy.

The economic standard is clearer than the dispute: commitments should cover attributable exposure, remain collectible under stress and produce transparent treatment of costs and revenues.

Minimum charges can make a demand forecast financially consequential. They may also discourage developers from reserving capacity they are unlikely to need. But contractual protection and household savings are separate propositions. The first requires enforceable, adequately backed obligations. The second requires demonstrated cost recovery and regulatory pass-through.

Sources and References

• Michigan Public Service Commission: October 1, 2026 conditional approval announcement.
• Michigan Public Service Commission: Data centers and electric service.
• Michigan Department of Attorney General: October 30, 2026 rehearing request announcement.
• FactCheck.org: Data centers and electricity prices, October 29, 2026.