Summary

  • The new Columbus contract does not make 11 MW equally firm in every tariff period: it specifies 300 kW on-peak and 11,000 kW off-peak, with an 11 MW maximum contract demand.
  • The 300 kW figure is not an absolute cap on peak consumption. PowerCompute says additional on-peak use remains available at excess-demand rates, while its current practice is to curtail mining during those hours.
  • Schedule DCB also creates minimum billing exposure. On an 11 MW off-peak contract, and assuming no higher demand in the preceding year, its formula implies a 3.9 MW off-peak billing floor and at least 429 MWh of off-peak energy per month.

An 11-megawatt power agreement sounds like a simple block of electricity. The filed agreement for PowerCompute’s Columbus, Mississippi, site divides that block by clock. It promises 300 kilowatts of firm power during the utility’s on-peak hours and 11,000 kilowatts during off-peak hours. The second number is 36.7 times the first. The company’s expansion is real as a higher off-peak contract quantity; it is not a statement that the same 11 MW is firm through every weekday peak window.

That distinction matters because the facility is not a generic data centre with one stable load. PowerCompute uses the site primarily for Bitcoin mining, a business that can turn machines down when power is costly or when selling energy is more attractive. The company says it does so during on-peak periods. It is also developing an enterprise high-performance computing (HPC) offer. A customer paying for compute usually values a machine that stays available, which creates a different relationship with a tariff that separates the day into two firm-power blocks.

The contract does not say the site is barred from drawing more than 300 kW at peak. The 300 kW line is the on-peak contract demand. PowerCompute says more on-peak electricity remains available under the schedule’s ordinary and excess-demand rates, and the contract sets a maximum demand of 11 MW. The commercial question is therefore not simply “Can the site run?” It is what it would cost to run above the firm on-peak allowance, how often the operator would choose to do so, and whether a compute customer would accept that operating pattern. (PowerCompute’s announcement; filed contract, Exhibit 10.1)

The 11 MW quantity belongs to particular hours

PowerCompute said on October 6 that its subsidiary US Digital Mining Mississippi LLC had signed a replacement power contract with Columbus Light and Water on September 30. Contracted demand at the site rose from 8.5 MW to 11 MW. The company says the service moved from the utility’s manufacturing schedule to its Large Data Service Schedule DCB, took effect for service on October 1 and continued without interruption. The SEC filing identifies the contract, the parties and an initial term reported to end September 16, 2030. The complete contract is attached as Exhibit 10.1. (Form 8-K)

The schedule defines peak by the calendar, not by a live wholesale-price signal. On weekdays from April through October, on-peak hours run from 1 p.m. to 7 p.m. In January through March and November through December, they run from 4 a.m. to 10 a.m. Saturdays, Sundays, November 1 and the named federal holidays are off-peak all day. All times are Central. The Tennessee Valley Authority (TVA) can change the windows with at least a year’s notice. A “peak” hour in this contract is thus a tariff category; the document does not prove that each such hour has the highest system cost, or that an off-peak hour is always cheap in a real-time market.

The company says its on-peak contract demand did not change when total contracted demand increased. That is a useful limit on what the 2.5 MW increase represents. The announcement does not publish the old hourly schedule, so it would be too strong to say exactly how the old 8.5 MW was divided between time periods. What the new exhibit does establish is that the current 11 MW firm quantity is explicitly attached to off-peak hours, while only 300 kW is firm on-peak.

Mining can follow the schedule; HPC has to price its exceptions

Mining machines can be curtailed as a group. When a miner’s expected Bitcoin revenue falls below the cost of using electricity—or when another power sale is more valuable—switching machines off can preserve cash rather than waste output. The timing rules make that flexibility commercially legible: PowerCompute can reduce its load in the specified windows, then use the larger off-peak allotment for the hours in which it chooses to mine. The contract does not show the site’s hourly load, how often machines are actually curtailed, or the revenue from each curtailment event.

HPC changes the customer promise. A GPU server can be interrupted or its workload moved, but the buyer may pay for completion time, availability, a reserved instance or a service-level commitment. A 300 kW firm peak allocation can support some continuous load, and excess demand may allow more use at a price. It does not establish that a customer can count on 11 MW of firm power around the clock at the same economics as the off-peak block. An operator considering a full conversion would need to price peak draw, workloads that can move, backup power or storage, and the lost mining or power-sale opportunity.

None of those costs can be calculated from the contract alone.

The company’s own disclosures describe a transition in progress rather than an HPC campus already using all 26 MW. Its September 30 release says about 22.5 MW across the Oklahoma and Mississippi portfolio primarily supports mining, with a portion allocated to an enterprise HPC pilot. In August it reported that one GPU in Oklahoma had passed Vast.ai’s machine-level “Verified” status and that a second unit had been ordered. The facility-level certification it described required ISO/IEC 27001 and five GPU servers; the company said it was working toward the requirements and gave no assurance it would qualify. These are signs of commercial experimentation, not evidence of sustained HPC demand at Columbus. (September 30 operating update; Vast.ai GPU update)

Curtailing load does not remove the minimum bill

A flexible operator still has a utility bill in the hours when it consumes less. Under Schedule DCB, the minimum off-peak billing demand is built from a percentage of contracted demand: 30% of the first 5,000 kW and 40% of the next 20,000 kW, using the greater of current contract demand or the highest relevant billing demand from the preceding 12 months. At 11,000 kW, if that prior-year high is no greater than the current contract amount, the arithmetic is 1,500 kW plus 2,400 kW, or 3,900 kW.

The schedule also says that off-peak energy for the month cannot fall below off-peak billing demand multiplied by 110 hours. With the same 3,900 kW assumption, that is 429,000 kWh, or 429 MWh. It is not a claim that PowerCompute actually used that amount in October; it is a tariff floor before the full demand, energy and adjustment calculation. If the prior-year demand was higher, the floor could differ. The tariff also measures monthly demand over 30-minute intervals, so short peaks matter to a charge even if the average over a day looks low.

Listed energy rates are only one part of the bill. For FY2027 the schedule lists DCB summer on-peak energy at 6.084 cents per kWh and winter on-peak energy at 4.873 cents. It separately lists customer and administrative charges, on-peak and maximum demand charges, excess demand, energy blocks and adjustments tied to TVA. These figures should not be compressed into an “all-in” site price without interval data and an invoice. PowerCompute has separately described a 3.3-cent-per-kWh blended net power cost across its 26 MW portfolio after energy sales; that company-wide net figure is not the same measure as a gross DCB line item for one Mississippi site. (Schedule DCB, Exhibit A; company portfolio-cost statement)

The contract also sets a commitment over time. It describes an initial five-year term from the In-Service Date and can charge remaining fixed and demand-related costs, plus direct unrecoverable utility costs, following early termination. The utility, not just the miner, needs a predictable load to plan around facilities and wholesale purchases. That obligation does not disappear because the operator curtails for six hours on a weekday.

But the public documents use several date formulations: the contract cover is dated September 16; both signatures are dated September 30; section 2.1 makes service effective October 1 subject to facilities being complete and in service; and the 8-K exhibit description says the contract is effective as of September 16. The 8-K summarizes the initial term as ending September 16, 2030. The record does not reconcile those references to the old and new service periods, so the exact date relationship should not be inferred from the headline alone.

The capacity charge is a company-reported outcome

The new data-centre rate introduced a separate capacity commitment charge for qualifying new or expanded loads. PowerCompute’s release describes the FY2027 amount as about $1.5 million per megawatt above the first 5 MW, with a 2,000 MW first cohort. The attached schedule prints $1.521 million per MW. The company says it has not been assessed the charge on any of its 11 MW and attributes this to load already under contract before October 1.

That outcome is important to the economics, but the public material reviewed here does not include a TVA or Columbus Light and Water eligibility letter or a bill showing the assessment. The date descriptions also need to be read carefully: the signed instrument and cover, the delivery-service effective date and the 8-K exhibit description do not use one unambiguous formulation. This is not evidence that the exemption is invalid; it means an outside reader cannot verify the utility’s application from the announcement alone. A written eligibility determination and the first itemized invoice would settle the question.

The company’s claim that the capacity charge was not assessed should remain attributed to the company, not upgraded into an independently verified finding.

The same distinction applies to the reported $300,000 deposit return. The company says it is replacing the cash deposit with a utility bond and intends to use the returned cash for site buildout. The 8-K says the company’s estimated share of Columbus Light and Water’s facility capital costs is zero. Those terms reduce cash tied up in the deposit and avoid a direct contribution to the stated utility facilities; they do not reveal the bond’s cost, the utility’s overall rate recovery or the site’s total electricity bill.

An 11 MW connection and an 11 MW compute service are different assets

PowerCompute’s contract may be attractive precisely because mining can shape its load around the tariff. The customer is getting a large, firm off-peak quantity, a much smaller firm quantity in defined peak windows, and the option to take more at excess rates subject to the contract. That can raise the value of electrical infrastructure whose current use can be switched quickly. It also makes the energy price and the power schedule part of the same operating decision.

A buyer of HPC capacity would ask a different set of questions: how much power remains firm in the hours when its workload cannot move, what redundancy sits behind the meter, what is the expected availability of the actual servers, how long does the customer contract run, and who pays when demand crosses the 300 kW firm peak level? The 11 MW contract provides evidence of a utility relationship and a tariff classification. It does not disclose the customer-specific combination of transmission, cooling, backup, network links, GPUs, staffing and uptime needed to turn that relationship into a dependable compute product.

The cost of mining also sets a real alternative. Q2 2026 results show $2.1 million in quarterly revenue, a $4.6 million net loss and a 29% mining margin, down from 41% a year earlier; the company reported $145,000 in curtailment and energy sales that quarter. Those figures are historical and consolidated, not the Columbus contract’s first bill. A September 30 debt repayment sharply reduced secured borrowings, but the company also said a $1.25 million secured note remained due December 31. The combination means energy flexibility is a potential operating advantage, not proof that the site has already converted into a stable compute business. (Q2 results; Form 10-Q; debt update)

For investors and operators, the better unit is not contracted megawatts in isolation. It is megawatts by hour, the rate for exceeding each firm block, the monthly billing floor, the actual revenue earned by a curtailment decision, and the cost of leaving the contract. PowerCompute has published enough to show that its 11 MW figure is a shaped power right. The remaining test is whether the company can produce the load data, tariff treatment and customer uptime evidence that would make that right valuable to a larger class of compute users.

Evidence register

  • PowerCompute’s October 6 announcement supports the announced 8.5-to-11 MW change, the company's account of peak curtailment, the charge assessment, deposit and equipment count.
  • The SEC Form 8-K identifies the parties, filed contract, reported term and utility-facility capital estimate. The full Exhibit 10.1 contract copy supports the separate firm quantities, Schedule DCB periods, rate lines, billing floor, minimum energy and termination terms.
  • PowerCompute’s September 30 debt and operating release, August 26 GPU update, and Q2 results provide the company’s stated load mix, early HPC milestone, mining performance and financial context.
  • The SEC Form 10-Q is the primary filing for Q2 financial statements and risks. None of these sources contains the site’s utility invoice, interval usage, a written rider determination or a customer contract for Columbus HPC service.