Last week, Amazon posted one of its cloud division’s best-ever quarters. The tech giant told investors that AWS’s operating margin jumped 6.5 percentage points from a year earlier.
The straightforward read is that cloud demand is outrunning cost. About 1.3 of those 6.5 points, though, came from a single accounting line: a gain on energy contracts for electricity Amazon may never actually use. Amazon’s own guidance for next quarter assumes that number goes back to zero.
The $551-million gain didn’t come from Amazon buying and selling power. It’s what shows up on the books when a company holds something like an option on roughly 270 million megawatt-hours of electricity, weighted toward delivery nine or more years out. This quarter, the expected future price of that power moved enough in Amazon’s favor to be really worth something on paper.
It works like this: Amazon signs long-term power contracts, some running 20 years, to secure electricity for data centers it hasn’t finished building. Normally, buying power a company intends to use doesn’t normally require restating its value every quarter. But Amazon’s 10-Q says the company “may make or receive net cash payments, rather than take delivery of electricity, when our consumption is less than committed quantities due to operational variability.” Because that possibility exists, the contracts get treated more like a running bet on future power prices than an ordinary purchase. Their value gets recalculated every quarter, whether that bet is currently paying off or not.
None of it is money changing hands, so there’s no real transaction to check the $551-million estimate against.
Every large power buyer in the country is now facing the same underlying questions or whether the load will show up the way the contract assumed it would. Amazon’s derivative accounting is one company’s way of living with that uncertainty on its books. But the utilities serving these loads are grappling with the same question.
The utility hedge
Dominion Energy’s new data center rate class, approved in November, will take effect for new contracts starting January 2027, and requires large customers to post $1.5 million in collateral for every megawatt of contracted capacity.
Dominion doesn’t dress up why the rule exists. The contract term and collateral requirement, the company wrote, ensure that if a data center shuts down completely, the minimum revenue threshold is still met. Just like Amazon’s energy contracts, these fees are a hedge.
The Virginia proceeding that produced the $1.5 million figure was a nine-day evidentiary hearing in September 2025 that drew filings from Google, Microsoft, the U.S. Department of Defense — and Amazon. What they argued about wasn’t whether the risk was real, it was who should get paid for carrying it. Dominion requested a higher return on equity, meaning the profit rate regulators let it earn on shareholders’ money, while the industry trade group the Data Center Coalition argued for lower. The commission ultimately split the difference, landing at 9.8%.
Both sides agreed that the hedge moves risk off Dominion’s books and onto its customers. They disagree only about what that shift is worth. Dominion requested a higher return on equity of at least 10.40%, while the Data Center Coalition, the industry’s own trade group, argued for 8.90%.
The commission split the difference and set 9.8%. That’s the return Dominion earns for that hedge, and by extension, what everyone paying Dominion’s rates ends up covering.
Latitude Intelligence tracks collateral terms across large-load tariffs nationally, and almost every one of them measures collateral the same way, as some multiple of the customer’s own monthly bill. Dominion, however, is the exception. It states a flat price per megawatt instead, which makes it the only publicly legible number in the country for what hyperscaler demand risk is actually worth. It’s also the only one you can multiply by a company’s stated capacity and get an answer.
The risk being hedged isn’t that a well-capitalized company defaults on its bills; it’s that a well-capitalized company changes its mind about how much power it needs. That’s the same operational variable Amazon’s own filing names as the reason its contracts are structured as derivatives.
But a stronger balance sheet doesn’t make that kind of change any less likely. If anything, the companies most able to walk away from a reservation without financial consequence are the best-capitalized ones, exactly who the waivers are most generous to.
Amazon is one of the highest-rated purchasers of electricity anywhere — and the company’s own 10-Q is the best evidence in the industry that a top-tier credit rating and demand certainty aren’t the same thing.
That said, the $1.5-million fee is a sticker price, not a cost. What a data center developer actually pays to hold that collateral depends on two things: how much it would otherwise earn on that money if it weren’t tied up, and how large a discount it gets for having strong credit.
For a top-rated customer, the real cost over fourteen years lands around $169,000 per megawatt, nearly a ninth of the sticker price. That number only holds if the contract runs its full term, though. A data center that defaults early, before most of its collateral has been released, faces a cost much closer to the full $1.5 million.
Even the discounted number carries a gap. Dominion’s own order says the collateral accrues interest, but never states a specific rate. So the real cost of the largest, most explicit collateral requirement in the country still isn’t fully knowable from the public record.
The county that hosts the risk
There’s a third party in this story who never filed a brief and isn’t collecting collateral, and its position may be the least hedged of anyone’s.
On July 22, Virginia’s Loudoun County’s Board of Supervisors voted to prepare a moratorium on new data center applications, site plans, and substation permits, to be taken up at the board’s September 15 meeting. Loudoun hosts the largest concentration of data centers in the world, and the same meeting rejected a 780-megawatt campus outright. The board carved out an exception for 24 applications already filed before February 2025, on the condition they proceed without substantial changes and sit at least 500 feet from residential property, and even that narrower exception only passed 5 to 4.
The county attorney told the board a blanket moratorium likely lacks legal standing under Virginia law, which requires each application to be judged on its individual merits rather than paused wholesale. The supervisor who cast the lone dissenting vote didn’t think the county would get away with it. If you pass this in September, he said, somebody will sue.
Loudoun isn’t the only Virginia locality reaching for this lever anymore. Other localities are pausing too. Chesapeake has already approved an eight-month pause on new applications, and Suffolk is rewriting its own rules under a similar pause. Front Royal is going further, drafting language that would ban data centers from its zoning code entirely.
None of these existed a year ago. Moratoria are happening now because the localities hosting this build-out are, in their own way, trying to buy the same kind of protection Dominion charges hyperscalers for and Amazon books on its own balance sheet.
A locality’s version of that insurance might not survive a legal challenge, though. The county’s own attorney told the board as much.
Every hedge in this story protects the party holding it and pushes the risk somewhere else. Amazon’s derivative accounting protects Amazon from any real cash consequence, whether this quarter’s number is a gain or a loss. Dominion’s collateral is held, interest-bearing in theory, and returned over 14 years, protecting the utility and, in the commission’s framing, its ratepayers. Loudoun’s moratorium, if it survives its own legal review, would protect the county’s grid and water supply.
These tariffs use credit rating to decide who gets a break on collateral, on the assumption that a stronger balance sheet means less risk. Amazon has about as good a rating as exists — but its own filings admit that rating doesn’t guarantee it will use all the energy it has committed to buy.
Meanwhile, Virginia priced that exact risk twice in the same case, once at $1.5 million per MW it had secured in collateral, once in the gap between the return Dominion argued it deserved and the lower one that regulators actually approved.
Neither price reaches the people who never had a seat at that table. A residential ratepayer is protected only as well as a formula that measures credit instead of demand happens to work. A county weighing a moratorium is relying on a legal theory its own attorney won’t vouch for. Every party to that case spent nine days arguing about exactly this risk, yet nobody in the room was representing whoever ends up holding what’s left over.
