On September 30, 2026, a subsidiary of a small Nasdaq-listed bitcoin mining company signed a power contract in Mississippi that may prove to be one of the best-timed contracts of the year. PowerCompute’s subsidiary, US Digital Mining Mississippi LLC, signed a new contract with Columbus Light and Water raising contracted demand at its Columbus, Mississippi site from 8.5 megawatts to 11 megawatts. The next day, October 1, the Tennessee Valley Authority’s new data center tariff took effect — including a capacity commitment charge of approximately $1.5 million per megawatt on new and expanded data center load above the first five megawatts.
The company says it was not assessed that charge on any of its 11 megawatts, because the contracted load predated the October 1 cutoff. The stakes are clear from the arithmetic: a brand-new 11-megawatt load would face the charge on the six megawatts above the threshold, roughly $9 million. PowerCompute stated no total dollar figure; the $9 million is straightforward multiplication from the company’s own description of the rate. A 2.5-megawatt increase — modest by hyperscale standards — arrived just ahead of a tariff designed to make exactly that kind of incremental reservation expensive.
This is a story about much more than one 11-megawatt site. TVA’s tariff is a major attempt by a large American power supplier to solve the problem at the center of this desk’s two foundational explainers: how do you plan a grid, and price access to it, when customers can reserve enormous amounts of capacity they may never use? The answer TVA has chosen — charge upfront for the reservation itself — is likely to echo well beyond the Tennessee Valley.
What TVA adopted
TVA’s Board of Directors approved the new data center rate at its quarterly meeting in Memphis on August 20, 2026. The package moved data center load out of the general industrial manufacturing rate class and into its own dedicated rate structure, with four rate classes depending on contract demand. TVA said average billing for data center customers would rise approximately 10 percent, phased in over three fiscal years. The new rate took effect October 1, 2026, the first day of TVA’s fiscal 2027.
The centerpiece is the Capacity Commitment Rider. Beginning in fiscal 2027, new or expanded data center load above the first five megawatts of contracted demand faces a charge of approximately $1.5 million per megawatt, described by TVA as an upfront commitment paid over three to five years to fund the grid upgrades the new load requires. TVA CFO Tom Rice said the provision covers the incremental capacity cost the data center rate alone does not cover.
The rate applies broadly. Service under the new schedules is available only where the major use of electricity at the delivery point falls under 2022 NAICS subsectors 518 or 519, NAICS codes 522320 or 541214, or is used, in TVA’s sole judgment, for computational equipment. Data center load under the new rate includes bitcoin mining — which is why PowerCompute’s Columbus mining site moved onto the new tariff rather than remaining on the manufacturing rate. TVA is classifying an entire category of compute load, from crypto mines to colocation halls, as a distinct kind of customer with distinct cost responsibilities.
The problem it is trying to solve
To understand why TVA would impose a charge this large, return to the forecasting and interconnection problems this desk has covered. In the old world, load growth was slow and predictable, and a utility could extrapolate from history. Now, a handful of corporate siting decisions, each worth hundreds of megawatts, determine a territory’s growth — and the interconnection queue is full of requests that will never be built. Developers routinely file multiple requests for alternative sites, and some are purely speculative, filed to hold a place in line.
That speculation is not harmless. When a utility plans around a queue full of phantom load, two bad things happen. First, the forecast gets inflated, and the utility may build generation and transmission for load that never materializes — leaving existing customers to pay for stranded capacity. Second, the queue itself clogs: every request must be studied against the other queued loads ahead of it, and when a speculative project drops out, the studies behind it may need to be redone.
The cost-shift mechanics are the heart of it. Serving a new large load requires real infrastructure — generation capacity, substation capacity, sometimes transmission lines that take years to build. If the customer that triggered those investments walks away, the cost does not vanish. It lands on the customers who remain: households and existing businesses that never asked for the expansion. TVA’s framing is explicit: the rate is designed so data center customers pay for the system builds required to serve their loads, rather than leaving residential customers carrying the cost.
TVA has paired that logic with a promise of restraint: interim president and CEO Mike Skaggs said TVA would build capacity and transmission to sit just below demand, not above it. The message is symmetrical: TVA will not speculate on future growth with other people’s money, and developers should not expect it to. The Sierra Club criticized the package, arguing it tilted toward serving large industrial loads — but the direction of the policy is not in dispute. TVA intends to make the cost causer pay, and to make the commitment real before the concrete is poured.
How the capacity commitment charge works
The charge is a price on the reservation itself, layered on top of the ordinary rate for power consumed. It is the tariff version of the staged-commitment mechanisms described in our planning explainer — minimum bills, take-or-pay provisions, deposits — which align the customer’s commitment with the utility’s investment. The charge does the same work upfront, in a single visible number, at the moment the load enters the system.
First, the five-megawatt threshold. Load at or below five megawatts escapes the charge entirely, which keeps small facilities and modest expansions outside the regime. The charge bites only on the megawatts above the threshold — so for an 11-megawatt load, six megawatts are chargeable; for a 100-megawatt campus, 95 megawatts would be. Second, the payment timeline: TVA has described the charge as paid over three to five years rather than in a lump sum, which spreads the burden but does not reduce it. Third, the cohort limit: for fiscal 2027, TVA set a 2,000-megawatt cohort threshold, available first-come, first-served — a rationing device that rewards early movers.
What the charge does not do is equally important. It does not replace the underlying rate: customers still pay for power used under the Large Data Service schedules, including the phased-in increase of roughly 10 percent. It does not guarantee faster interconnection; it prices the reservation, but engineering studies, upgrade timelines, and the grid’s physical constraints remain. A developer who must pay roughly $1.5 million per megawatt to reserve capacity has a strong reason to request only the capacity it genuinely intends to use. That honesty is the point.
The Columbus case study, line by line
PowerCompute’s Columbus contract illustrates how the new regime works, precisely because the site is small enough to see whole. The subsidiary signed with Columbus Light and Water — the local power company distributing TVA power in Columbus — on September 30, 2026. The contract moved the site from TVA’s general industrial Manufacturing Service Rate, Schedule MSB, to Columbus Power’s Large Data Service Rate, Schedule DCB, which implements TVA’s new data center rate. Service continued without interruption, and the contract runs through September 16, 2030, with a five-year renewal if both parties agree.
The financial terms are as revealing as the timing. Columbus Light and Water required no capital contribution for the additional 2.5 megawatts; the company’s estimated share of the utility’s capital costs was zero. The utility is also returning the company’s $300,000 power deposit, which PowerCompute is replacing with a utility bond, and management said it intends to apply the returned cash toward buildout at Columbus. The site’s operating practice did not change: the company curtails during on-peak hours and draws contracted demand during the remaining hours, lowering the demand-charge component of its power cost, with on-peak power still available at the tariff’s on-peak and excess demand rates.
Read that package together and it tells a coherent story about what the old regime looked like. Under the manufacturing rate, a compute load could hold 8.5 megawatts, expand to 11, post a modest deposit, and contribute nothing to the utility’s capital costs — while the utility carried the obligation to serve the full reservation. That is the world TVA’s new tariff is designed to end. The company said it moved to the data center rate on its own, before October 1, ahead of the changes taking effect. Columbus now carries the tariff classification the company says any high-performance-computing facility on TVA power must have — obtained, in effect, at the old price.
There is a broader lesson in the grandfathering. Any cutoff creates a cliff, and cliffs reward whoever gets across first. Load under contract effective before October 1 escaped the charge; load contracted after it pays. That is a defensible line — contracts signed under the old regime cannot fairly be repriced retroactively — but it also means the first weeks of the new tariff will be full of stories like this one, as developers rush to lock in reservations under expiring terms. The Columbus contract is unlikely to be the last example of pre-deadline positioning, but it is an early one disclosed in an SEC filing.
What it means for siting and contracting
TVA’s tariff changes the math for developers choosing where to build. First, it makes the TVA region’s true cost of entry visible upfront. A developer comparing sites across regions can no longer treat TVA capacity as cheap to reserve and cheap to abandon; the reservation itself now has a price tag of roughly $1.5 million per megawatt above five megawatts. That will push some developers toward smaller initial reservations, staged expansions tied to real construction milestones, and more careful queue discipline — exactly the behaviors the tariff is designed to encourage.
Second, it puts TVA in a national context where every region is tightening the same screws in its own way. Texas, facing an interconnection queue of roughly 474 gigawatts — about 90 percent of it attributed to data centers — enacted Senate Bill 6 in 2025, requiring large-load customers to demonstrate site control, self-fund upgrades, disclose duplicate requests, and accept curtailment during grid stress. TVA’s approach differs in mechanism — a tariff charge rather than a statute — but matches it in logic: make the reservation costly enough that only serious projects file for it.
The era in which a data center could sign a loose reservation, hold it cheaply, and decide later whether to build is closing. Expect to see more of what TVA’s tariff incentivizes: earlier, firmer commitments; initial tranches sized to what the developer can actually use; expansions contracted in stages as construction milestones are met; and more sophisticated on-peak behavior, like the curtailment PowerCompute practices at Columbus, as customers manage demand charges under their new schedules. The 5-megawatt threshold will also shape project design at the margin: small expansions just above the threshold will face disproportionate scrutiny, while genuinely small facilities stay outside the regime entirely.
There is also a competitive question. TVA’s charge is large enough to influence siting decisions, and interstate competition for data center investment is real. If neighboring regions offer cheaper reservation terms, some development will migrate — TVA will have traded growth for cost protection. That is a legitimate policy choice, and TVA’s leadership has signaled comfort with it. The first test will be whether the 2,000-megawatt fiscal 2027 cohort fills quickly, slowly, or not at all.
What to watch
Three things will determine whether TVA’s experiment becomes a template. The first is the cohort itself: how quickly the 2,000 megawatts of fiscal 2027 capacity is claimed, and by whom — hyperscale AI campuses, crypto miners converting to high-performance computing, colocation providers, or some mix. The second is whether other utilities follow. TVA is the nation’s largest public power supplier, serving parts of seven states, and its move carries weight beyond its territory. If the charge survives its first year without driving away the load it seeks to serve, expect other utilities to study it closely — and developers to start pricing reservation costs into siting models as a standard line item.
The third is the distributional question that never goes away. Whether it protects households depends not just on its text but on its administration: how rigorously TVA enforces eligibility, how honestly developers report intended loads, and whether the capacity funded by the charge arrives on the timeline the reservations imply. The tariff is a tool, not a verdict. Its success will be measured in the gap between the load developers reserve and the load they actually build — and in whether customers who never asked for the data center boom end up paying for it anyway.